<?xml version="1.0" encoding="UTF-8" standalone="no"?><rss xmlns:atom="http://www.w3.org/2005/Atom" xmlns:blogger="http://schemas.google.com/blogger/2008" xmlns:gd="http://schemas.google.com/g/2005" xmlns:georss="http://www.georss.org/georss" xmlns:openSearch="http://a9.com/-/spec/opensearchrss/1.0/" xmlns:thr="http://purl.org/syndication/thread/1.0" version="2.0"><channel><atom:id>tag:blogger.com,1999:blog-7258374652086543351</atom:id><lastBuildDate>Sun, 01 Nov 2026 09:13:09 +0000</lastBuildDate><category>LLM</category><category>Keras</category><category>Text Generation</category><category>LSTM</category><category>Text Classification</category><category>Multi-Class Text Classification</category><category>TF Data Pipeline</category><category>Classification</category><category>Murat Karakaya Akademi</category><category>Text Preprocessing</category><category>Yapay Zeka</category><category>Seq2Seq</category><category>Transformers</category><category>Classification Metrics</category><category>Index</category><category>Keras TextVectorization</category><category>Language Model</category><category>Ollama</category><category>Encoder  _Decoder</category><category>Metrics</category><category>Multi-Topic Text Classification</category><category>Agents</category><category>Açık Kaynak</category><category>Image</category><category>Open Weights</category><category>RAG</category><category>Türkçe</category><category>Agentic AI</category><category>Artificial Intelligence</category><category>Büyük Dil Modelleri</category><category>Embedding Layer</category><category>Exploratory Data Analysis</category><category>Loss Functions</category><category>VS Code</category><category>AI</category><category>AI Agents</category><category>Activation Functions</category><category>Araçlar</category><category>Controllable Text Generation</category><category>Fine-Tuning</category><category>Free API</category><category>GPT</category><category>Gemini</category><category>Open Source AI</category><category>Prompt Engineering</category><category>Python</category><category>Resources</category><category>Yerel Yapay Zeka</category><category>AGI</category><category>AI Governance</category><category>AI Safety</category><category>AI Tools</category><category>AUC</category><category>Accuracy</category><category>Ar-Ge</category><category>Benchmark</category><category>Binary Classification</category><category>Binary Text Classification</category><category>Claude Fable 5</category><category>Derin Öğrenme</category><category>EDA</category><category>End-to-End Model</category><category>F1</category><category>Frontier AI</category><category>GGUF</category><category>GPT-5.6</category><category>GPU</category><category>Gemma</category><category>Generative AI</category><category>Inkling</category><category>Keras Model Load</category><category>Keras Model Save</category><category>Kimi K3</category><category>Llama</category><category>Local AI</category><category>MCP</category><category>Multi-Class Classsification</category><category>Multi-Label Text Classification</category><category>Precision</category><category>QLoRA</category><category>Qwen3.8</category><category>Recall</category><category>Recurrent Neural Network</category><category>Sampling</category><category>Teknoloji</category><category>Unsloth</category><category>Yapay Zekâ</category><category>free</category><category>ücretsiz</category><category>ABD</category><category>ABD Çin Rekabeti</category><category>AI Engineering</category><category>AI Regulation</category><category>AI Sovereignty</category><category>Abonelik Modelleri</category><category>Acik Modeller</category><category>Agent Skill</category><category>Agentic</category><category>Attention</category><category>Açik Agirlik</category><category>Açık Ağırlık</category><category>Açık Ağırlık Modeller</category><category>BERT</category><category>Bahdanau</category><category>Baise</category><category>CPU ile Yapay Zeka</category><category>CV</category><category>Career Strategy</category><category>China</category><category>CoPilot</category><category>Conv1D</category><category>Cost</category><category>Cyber Security</category><category>DICOM Veri İşleme</category><category>Deep Learning</category><category>Early Specialization</category><category>Education</category><category>Eleştirel Düşünme</category><category>Embedding</category><category>Engineering</category><category>Erken Uzmanlaşma</category><category>Evren</category><category>Future of AI</category><category>Gemma 3 Vision</category><category>GitHub</category><category>Google AI Studio</category><category>Gradium Phonon</category><category>Görüntü İşleme</category><category>Imbalanced Dataset</category><category>Jev</category><category>Kaggle</category><category>Kariyer Stratejisi</category><category>Keras Custom Layer</category><category>Kurumsal</category><category>LLAMA 3</category><category>Label Encoding</category><category>Large Language Models</category><category>Llama.cpp</category><category>Luong</category><category>M31 Flash</category><category>MLOps</category><category>MLP</category><category>Masking</category><category>Meta Learning</category><category>Meta Öğrenme</category><category>Metinden Sese</category><category>MiniMax</category><category>MiniMax M3.1 Flash</category><category>Model Eğitimi</category><category>Multimodal AI</category><category>Murat Karakaya</category><category>Murat Karakaya Academy</category><category>NVIDIA Nemotron</category><category>Nemotron 3.5 Super VL</category><category>Ngrok</category><category>Nutuk</category><category>OCR</category><category>Ollama Cloud</category><category>Open Source</category><category>Open WEBUI</category><category>Open Weight</category><category>OpenClaw</category><category>Otomatik Fonksiyon Çağırma</category><category>PDF</category><category>Padding</category><category>Pay As You Go</category><category>Proje Yönetimi</category><category>Project Management</category><category>Q&amp;A</category><category>QWEN</category><category>R&amp;D</category><category>ROC</category><category>Reasoning Models</category><category>Research</category><category>Research and Development</category><category>Savunma Sanayi</category><category>Sağlıkta Yapay Zeka</category><category>Sesli Erişilebilirlik</category><category>Siber Güvenlik</category><category>Software</category><category>SoruCevap</category><category>Sosyal Medya</category><category>Systems Engineering</category><category>Teacher Forcing</category><category>Tech Policy</category><category>Technology</category><category>Teknofest</category><category>Teknoloji Politikası</category><category>TensorBoard</category><category>TogetherAI</category><category>USA</category><category>Vibe Codding</category><category>Voice AI</category><category>Worthless Test</category><category>Yapay Zeka Ajanları</category><category>Yapay Zeka Balonu</category><category>Yapay Zeka Güvenliği</category><category>Yapay Zeka Yarışması</category><category>Yapay Zekâ Ajanları</category><category>Yazılım</category><category>Yazılım Mühendisliği</category><category>Yeni Mezun</category><category>Yerel LLM</category><category>Yerel LLM Kurulumu</category><category>YouTube</category><category>YouTube AI</category><category>apay Zeka</category><category>dataset</category><category>open</category><category>price</category><category>Çanakkale Zaferi</category><category>Çin</category><category>Üretken Yapay Zeka</category><category>Üretken Yapay Zekâ</category><category>İnce Ayar</category><title>Deep Learning Tutorials by Murat Karakaya Akademi</title><description>Deep Learning Solutions implemented by Python &amp; Keras</description><link>https://www.muratkarakaya.net/</link><managingEditor>noreply@blogger.com (Murat Karakaya)</managingEditor><generator>Blogger</generator><openSearch:totalResults>115</openSearch:totalResults><openSearch:startIndex>1</openSearch:startIndex><openSearch:itemsPerPage>25</openSearch:itemsPerPage><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-1687070476432364967</guid><pubDate>Sun, 01 Nov 2026 09:12:48 +0000</pubDate><atom:updated>2026-11-01T12:12:48.172+03:00</atom:updated><title>YouTube AI Weekly Trend Report (October 25 – November 1, 2026)</title><description>&lt;!doctype html&gt;&lt;html lang="en"&gt;&lt;head&gt;&lt;meta charset="utf-8"&gt;
&lt;meta name="viewport" content="width=device-width,initial-scale=1"&gt;
&lt;title&gt;YouTube AI Weekly Trend Report&lt;/title&gt;&lt;style&gt;
:root{--bg:transparent;--sf:#fff;--sf2:#eef1f4;--ln:#d7dce2;--ink:#111;--mu:#555;
--ac:#00796b;--am:#b45309;--mono:ui-monospace,Consolas,monospace}
*{box-sizing:border-box;margin:0;padding:0}
body{background:var(--bg);color:var(--ink);font:15px/1.5 -apple-system,'Segoe UI',Roboto,sans-serif;padding:28px 24px 60px}
.wrap{max-width:1100px}
h1{font-size:32px;letter-spacing:.3px;text-align:center;margin-top:8px} .datebar{text-align:center;margin:12px 0 6px}
.dtc{display:inline-block;font:600 15px/1 var(--mono);color:#0b0e12;background:#35d0ba;padding:7px 16px;border-radius:999px}
.meta{color:var(--mu);font-family:var(--mono);font-size:13.5px;margin-top:2px;text-align:center}
.dseg{cursor:pointer;transition:stroke-width .12s} .dseg:hover{stroke-width:36}
#tip{position:fixed;left:0;top:0;z-index:9;pointer-events:none;background:#f6f7f8;border:1px solid #c9d0d8;color:#111;font:12.5px/1.45 var(--mono);padding:7px 11px;border-radius:8px;opacity:0;transition:opacity .12s;max-width:340px;box-shadow:0 6px 20px rgba(0,0,0,.15)}
.meta b{color:var(--ac);font-weight:600}
nav{margin:14px 0 26px;display:flex;flex-wrap:wrap;gap:8px}
nav a{color:var(--mu);text-decoration:none;font-size:12px;padding:4px 10px;border:1px solid var(--ln);border-radius:999px;line-height:1.35} nav a b{color:var(--ac);font-family:var(--mono);font-weight:600;margin-left:4px}
nav a:hover{color:var(--ink);border-color:var(--ac)}
.card{background:var(--sf);border:1px solid var(--ln);border-radius:12px;padding:16px 18px}
.card h3,.stats h3{font-size:12px;text-transform:uppercase;letter-spacing:1.2px;color:var(--mu);margin-bottom:10px;font-weight:600}
.top{display:grid;grid-template-columns:2fr 1fr;gap:14px;margin-bottom:14px}
.champ{background:linear-gradient(135deg,#f7f9fa,#eef2f5);border:1px solid #d7dce2;border-radius:12px;padding:20px;text-decoration:none;color:var(--ink);display:block}
.champ:hover{border-color:var(--am)}
.ck{font-size:11px;letter-spacing:1.5px;text-transform:uppercase;color:var(--am)}
.cv{font:600 44px/1.1 var(--mono);color:var(--am);margin:6px 0}
.ct{font-size:17px;font-weight:600} .cs{color:var(--mu);font-size:13px;margin-top:4px}
.cd{color:#444;font-size:12.5px;line-height:1.55;margin-top:10px;border-top:1px solid var(--ln);padding-top:10px}
.stats{background:var(--sf);border:1px solid var(--ln);border-radius:12px;padding:14px 18px;display:flex;flex-direction:column;justify-content:space-between}
.st{display:flex;justify-content:space-between;align-items:baseline;padding:5px 0;border-bottom:1px solid var(--ln)}
.st:last-child{border:0} .st b{font:600 19px var(--mono)} .st span{color:var(--mu);font-size:12.5px}
.credit{text-align:center;margin-top:14px;font:600 14px var(--sans);color:#333;letter-spacing:.2px}
.credit b{color:var(--am);font-weight:800}
.rowg{display:grid;gap:14px;margin-bottom:14px}
.g23{grid-template-columns:2fr 1fr} .g11{grid-template-columns:1.2fr 1fr}
@media(max-width:820px){.top,.g23,.g11{grid-template-columns:1fr}}
svg{width:100%;height:auto;display:block}
.ax{font:11px var(--mono);fill:#555} .bn2{font:600 11px var(--mono);fill:#333}
.dn{font:700 30px var(--mono);fill:#111}
.donut{max-width:220px;margin:0 auto}
.lgs{display:flex;flex-wrap:wrap;gap:6px 14px;margin-top:12px}
.lg{font-size:11.5px;color:var(--mu)} .lg i{display:inline-block;width:9px;height:9px;border-radius:2px;margin-right:5px} .lg b{color:#111;font-family:var(--mono)}
.cbar{display:grid;grid-template-columns:130px 1fr 90px;gap:10px;align-items:center;padding:7px 0;border-bottom:1px solid var(--ln)}
.cbar:last-child{border:0} .cn{font-size:12.5px;color:#333;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
.ct2{background:var(--sf2);height:10px;border-radius:4px;overflow:hidden} .cf{display:block;height:100%;background:var(--ac);opacity:.85;border-radius:4px}
.cv2{font:600 12.5px var(--mono);text-align:right} .cv2 i{font-style:normal;color:var(--mu);font-size:10.5px;margin-left:6px}
h2.sec{font-size:13px;text-transform:uppercase;letter-spacing:1.4px;color:var(--mu);margin:26px 0 10px}
ol.heads{list-style:none;display:flex;flex-direction:column;gap:8px}
ol.heads li{display:flex;gap:14px;background:var(--sf);border:1px solid var(--ln);border-radius:10px;padding:13px 16px}
ol.heads .hr{font:700 20px var(--mono);color:var(--ac);min-width:26px}
ol.heads li.h2 .hr{color:#0369a1} ol.heads li.h3 .hr{color:#475569}
ol.heads li.h1{border-color:#bfe3dd}
ol.heads p{font-size:14.5px} ol.heads li.h1 p{font-size:16px;font-weight:600} ol.heads li.h3 p{color:var(--mu)}
.bars{display:flex;flex-direction:column;gap:4px}
.bar{display:grid;grid-template-columns:190px 1fr 30px;gap:12px;align-items:center;text-decoration:none;color:var(--ink);padding:4px 0}
.bar:hover .bn{color:var(--ac)} .bn{font-size:12.5px;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
.bt{background:var(--sf2);border-radius:4px;height:10px;overflow:hidden} .bf{display:block;height:100%;opacity:.9;border-radius:4px}
.bv{font:600 12.5px var(--mono);text-align:right;color:var(--mu)}
@media(max-width:640px){.bar{grid-template-columns:130px 1fr 28px}}
.theme{margin-top:30px} .theme h2{font-size:17px;display:flex;align-items:baseline;gap:10px}
.tn{font:700 15px var(--mono)} .dot{width:9px;height:9px;border-radius:3px;display:inline-block}
.cnt{font:400 12.5px var(--mono);color:var(--mu);margin-left:auto}
.insight{color:#444;font-size:13.5px;margin:6px 0 10px;padding-left:2px;max-width:860px}
.rows{border:1px solid var(--ln);border-radius:10px;overflow:hidden}
.row{display:grid;grid-template-columns:64px 56px 150px 1fr;gap:12px;align-items:baseline;padding:9px 14px;border-bottom:1px solid var(--ln);text-decoration:none;color:var(--ink);background:var(--sf)}
.row:last-child{border:0} .row:hover{background:var(--sf2)}
.vw{font:600 12.5px var(--mono);text-align:right}
.v1{color:var(--am)} .v2{color:var(--ac)} .v3{color:var(--mu)}
.dt{font:12px var(--mono);color:var(--mu)} .ch{font-size:12px;color:#2c5f8a;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
.tt{font-size:13.5px} .other .tt{color:var(--mu)}
footer{margin-top:34px;color:#555;font:12px/1.7 var(--mono);border-top:1px solid var(--ln);padding-top:12px}
&lt;/style&gt;&lt;/head&gt;&lt;body&gt;&lt;div class="wrap"&gt;

&lt;div class="datebar"&gt;&lt;span class="dtc"&gt;2026-10-25 &amp;rarr; 2026-11-01&lt;/span&gt;&lt;/div&gt;
&lt;div class="credit"&gt;This data was compiled by &lt;b&gt;Murat Karakaya Academy&lt;/b&gt;. &lt;p&gt;&lt;em&gt;Turkish edition: &lt;a href="https://www.muratkarakaya.net/2026/11/youtube-ai-haftalk-trend-raporu-25-ekim.html"&gt;YouTube AI Haftalık Trend Raporu&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;&lt;/div&gt;

&lt;div class="meta"&gt;&lt;b&gt;27&lt;/b&gt; active channels / 43 &amp;nbsp;|&amp;nbsp; &lt;b&gt;135&lt;/b&gt; videos &amp;nbsp;|&amp;nbsp; &lt;b&gt;3.1M&lt;/b&gt; total views&lt;/div&gt;
&lt;h2 class="sec"&gt;Headlines of the Week&lt;/h2&gt;&lt;ol class="heads"&gt;&lt;li class="h1"&gt;&lt;span class="hr"&gt;1&lt;/span&gt;&lt;p&gt;Google shipped its second frontier model in the span of two weeks: Gemini 4 Carbon took the top spot on the intelligence index, and the launch video reached 421,000 views.&lt;/p&gt;&lt;/li&gt;&lt;li class="h2"&gt;&lt;span class="hr"&gt;2&lt;/span&gt;&lt;p&gt;DeepSeek is back on the marquee: &amp;#x27;DeepSeek crushed everyone&amp;#x27; and &amp;#x27;the new model is an absolute monster&amp;#x27; drove a wave past 300,000 views among the week&amp;#x27;s most-watched videos.&lt;/p&gt;&lt;/li&gt;&lt;li class="h3"&gt;&lt;span class="hr"&gt;3&lt;/span&gt;&lt;p&gt;The local AI camp got loud: allegations that NVIDIA GPUs quietly throttle local AI, plus a 3,000 tokens/sec local run, set the hardware agenda for the week.&lt;/p&gt;&lt;/li&gt;&lt;/ol&gt;
&lt;h2 class="sec"&gt;Themes of the Week&lt;/h2&gt;&lt;nav&gt;&lt;a href="#t1" style="font-size:12.8px"&gt;The Gemini 4 Carbon era&lt;b&gt;9&lt;/b&gt;&lt;/a&gt; &lt;a href="#t2" style="font-size:11.5px"&gt;Genie 4 and world models&lt;b&gt;4&lt;/b&gt;&lt;/a&gt; &lt;a href="#t3" style="font-size:11.8px"&gt;The DeepSeek storm&lt;b&gt;5&lt;/b&gt;&lt;/a&gt; &lt;a href="#t4" style="font-size:11.5px"&gt;GPT-6.1 and OpenAI&amp;#x27;s impossible claim&lt;b&gt;4&lt;/b&gt;&lt;/a&gt; &lt;a href="#t5" style="font-size:11.2px"&gt;Kimi and the raw benchmarks&lt;b&gt;3&lt;/b&gt;&lt;/a&gt; &lt;a href="#t6" style="font-size:11.0px"&gt;Mistral Large 4 and cyber sovereignty&lt;b&gt;2&lt;/b&gt;&lt;/a&gt; &lt;a href="#t7" style="font-size:13.2px"&gt;OCR and document intelligence&lt;b&gt;11&lt;/b&gt;&lt;/a&gt; &lt;a href="#t8" style="font-size:13.0px"&gt;Voice AI and speech&lt;b&gt;10&lt;/b&gt;&lt;/a&gt; &lt;a href="#t9" style="font-size:15.2px"&gt;Small models and efficiency&lt;b&gt;19&lt;/b&gt;&lt;/a&gt; &lt;a href="#t10" style="font-size:11.2px"&gt;Local hardware and compute&lt;b&gt;3&lt;/b&gt;&lt;/a&gt; &lt;a href="#t11" style="font-size:13.5px"&gt;Coding agents and developer tools&lt;b&gt;12&lt;/b&gt;&lt;/a&gt; &lt;a href="#t12" style="font-size:17.0px"&gt;Agent orchestration and the LangChain ecosystem&lt;b&gt;26&lt;/b&gt;&lt;/a&gt; &lt;a href="#t13" style="font-size:11.8px"&gt;Image and video generation and multimodal&lt;b&gt;5&lt;/b&gt;&lt;/a&gt; &lt;a href="#t14" style="font-size:11.5px"&gt;OpenAI&amp;#x27;s product and enterprise front&lt;b&gt;4&lt;/b&gt;&lt;/a&gt; &lt;a href="#t15" style="font-size:11.0px"&gt;Anthropic and the balance of power&lt;b&gt;2&lt;/b&gt;&lt;/a&gt; &lt;a href="#t16" style="font-size:13.0px"&gt;Practical AI on the job&lt;b&gt;10&lt;/b&gt;&lt;/a&gt; &lt;a href="#t17" style="font-size:12.0px"&gt;Learn and level up&lt;b&gt;6&lt;/b&gt;&lt;/a&gt;&lt;/nav&gt;
&lt;div class="top"&gt;&lt;a class="champ" href="https://www.youtube.com/watch?v=kMpTBy3g1N0" target="_blank" rel="noopener"&gt;&lt;div class="ck"&gt;Record of the week&lt;/div&gt;&lt;div class="cv"&gt;421K&lt;/div&gt;&lt;div class="ct"&gt;Gemini 4 Carbon: The New Frontier&lt;/div&gt;&lt;div class="cs"&gt;Prompt Engineering &amp;middot; 28 Oct&lt;/div&gt;&lt;div class="cd"&gt;Google&amp;#x27;s Gemini 4 Carbon, its second frontier model introduced within two weeks, reached number one on the Artificial Analysis intelligence index with 59 points; the launch video argues the model stands out across the entire cyber-defence workflow, from finding a vulnerability to proving it is exploitable to writing the patch.&lt;/div&gt;&lt;/a&gt;
&lt;div class="stats"&gt;&lt;h3&gt;Weekly Figures&lt;/h3&gt;&lt;div class="st"&gt;&lt;span&gt;Active channels&lt;/span&gt;&lt;b&gt;27/43&lt;/b&gt;&lt;/div&gt;
&lt;div class="st"&gt;&lt;span&gt;Total videos&lt;/span&gt;&lt;b&gt;135&lt;/b&gt;&lt;/div&gt;
&lt;div class="st"&gt;&lt;span&gt;Total views&lt;/span&gt;&lt;b&gt;3.1M&lt;/b&gt;&lt;/div&gt;
&lt;div class="st"&gt;&lt;span&gt;Theme count&lt;/span&gt;&lt;b&gt;17&lt;/b&gt;&lt;/div&gt;&lt;/div&gt;&lt;/div&gt;
&lt;div class="rowg g23"&gt;
&lt;div class="card"&gt;&lt;h3&gt;Weekly rhythm — daily video count (bars · left axis) + views (line · right axis)&lt;/h3&gt;&lt;svg viewBox="0 0 960 250" role="img"&gt;&lt;line x1="44" y1="26" x2="904" y2="26" stroke="#d7dce2"/&gt;&lt;text x="36" y="30" text-anchor="end" class="ax"&gt;33&lt;/text&gt;&lt;line x1="44" y1="74" x2="904" y2="74" stroke="#d7dce2"/&gt;&lt;text x="36" y="78" text-anchor="end" class="ax"&gt;25&lt;/text&gt;&lt;line x1="44" y1="121" x2="904" y2="121" stroke="#d7dce2"/&gt;&lt;text x="36" y="125" text-anchor="end" class="ax"&gt;16&lt;/text&gt;&lt;line x1="44" y1="168" x2="904" y2="168" stroke="#d7dce2"/&gt;&lt;text x="36" y="172" text-anchor="end" class="ax"&gt;8&lt;/text&gt;&lt;line x1="44" y1="216" x2="904" y2="216" stroke="#d7dce2"/&gt;&lt;text x="36" y="220" text-anchor="end" class="ax"&gt;0&lt;/text&gt;&lt;rect x="63.3" y="193.0" width="68.8" height="23.0" rx="3" fill="#2b5f58"&gt;&lt;title&gt;4 videos&lt;/title&gt;&lt;/rect&gt;&lt;text x="97.8" y="188.0" text-anchor="middle" class="bn2"&gt;4&lt;/text&gt;&lt;text x="97.8" y="238" text-anchor="middle" class="ax"&gt;25 Oct&lt;/text&gt;&lt;rect x="170.8" y="106.6" width="68.8" height="109.4" rx="3" fill="#2b5f58"&gt;&lt;title&gt;19 videos&lt;/title&gt;&lt;/rect&gt;&lt;text x="205.2" y="101.6" text-anchor="middle" class="bn2"&gt;19&lt;/text&gt;&lt;text x="205.2" y="238" text-anchor="middle" class="ax"&gt;26 Oct&lt;/text&gt;&lt;rect x="278.4" y="89.3" width="68.8" height="126.7" rx="3" fill="#2b5f58"&gt;&lt;title&gt;22 videos&lt;/title&gt;&lt;/rect&gt;&lt;text x="312.8" y="84.3" text-anchor="middle" class="bn2"&gt;22&lt;/text&gt;&lt;text x="312.8" y="238" text-anchor="middle" class="ax"&gt;27 Oct&lt;/text&gt;&lt;rect x="385.9" y="26.0" width="68.8" height="190.0" rx="3" fill="#2b5f58"&gt;&lt;title&gt;33 videos&lt;/title&gt;&lt;/rect&gt;&lt;text x="420.2" y="21.0" text-anchor="middle" class="bn2"&gt;33&lt;/text&gt;&lt;text x="420.2" y="238" text-anchor="middle" class="ax"&gt;28 Oct&lt;/text&gt;&lt;rect x="493.4" y="77.8" width="68.8" height="138.2" rx="3" fill="#2b5f58"&gt;&lt;title&gt;24 videos&lt;/title&gt;&lt;/rect&gt;&lt;text x="527.8" y="72.8" text-anchor="middle" class="bn2"&gt;24&lt;/text&gt;&lt;text x="527.8" y="238" text-anchor="middle" class="ax"&gt;29 Oct&lt;/text&gt;&lt;rect x="600.9" y="112.4" width="68.8" height="103.6" rx="3" fill="#2b5f58"&gt;&lt;title&gt;18 videos&lt;/title&gt;&lt;/rect&gt;&lt;text x="635.2" y="107.4" text-anchor="middle" class="bn2"&gt;18&lt;/text&gt;&lt;text x="635.2" y="238" text-anchor="middle" class="ax"&gt;30 Oct&lt;/text&gt;&lt;rect x="708.4" y="141.2" width="68.8" height="74.8" rx="3" fill="#2b5f58"&gt;&lt;title&gt;13 videos&lt;/title&gt;&lt;/rect&gt;&lt;text x="742.8" y="136.2" text-anchor="middle" class="bn2"&gt;13&lt;/text&gt;&lt;text x="742.8" y="238" text-anchor="middle" class="ax"&gt;31 Oct&lt;/text&gt;&lt;rect x="815.9" y="204.5" width="68.8" height="11.5" rx="3" fill="#2b5f58"&gt;&lt;title&gt;2 videos&lt;/title&gt;&lt;/rect&gt;&lt;text x="850.2" y="199.5" text-anchor="middle" class="bn2"&gt;2&lt;/text&gt;&lt;text x="850.2" y="238" text-anchor="middle" class="ax"&gt;01 Nov&lt;/text&gt;&lt;polyline fill="none" stroke="#d97706" stroke-width="2" points="97.8,183.5 205.2,158.0 312.8,130.3 420.2,26.0 527.8,162.1 635.2,99.4 742.8,207.8 850.2,216.0"/&gt;&lt;circle cx="97.8" cy="183.5" r="3.5" fill="#d97706"&gt;&lt;title&gt;184K views&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="205.2" cy="158.0" r="3.5" fill="#d97706"&gt;&lt;title&gt;328K views&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="312.8" cy="130.3" r="3.5" fill="#d97706"&gt;&lt;title&gt;485K views&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="420.2" cy="26.0" r="3.5" fill="#d97706"&gt;&lt;title&gt;1.1M views&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="527.8" cy="162.1" r="3.5" fill="#d97706"&gt;&lt;title&gt;305K views&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="635.2" cy="99.4" r="3.5" fill="#d97706"&gt;&lt;title&gt;659K views&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="742.8" cy="207.8" r="3.5" fill="#d97706"&gt;&lt;title&gt;46K views&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="850.2" cy="216.0" r="3.5" fill="#d97706"&gt;&lt;title&gt;84 views&lt;/title&gt;&lt;/circle&gt;&lt;text x="912" y="220" class="ax" fill="#92400e"&gt;0&lt;/text&gt;&lt;text x="912" y="125" class="ax" fill="#92400e"&gt;537K&lt;/text&gt;&lt;text x="912" y="30" class="ax" fill="#92400e"&gt;1.1M&lt;/text&gt;&lt;/svg&gt;&lt;/div&gt;
&lt;div class="card"&gt;&lt;h3&gt;Theme distribution&lt;/h3&gt;&lt;svg viewBox="0 0 220 220" class="donut"&gt;&lt;path class="dseg" data-tip="The Gemini 4 Carbon era — 9 videos (%7)" d="M110.0,30.0 A80,80 0 0 1 142.4,36.8" stroke="#0d9488" stroke-width="30" fill="none"&gt;&lt;title&gt;The Gemini 4 Carbon era: 9 (%7)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Genie 4 and world models — 4 videos (%3)" d="M142.5,36.9 A80,80 0 0 1 155.4,44.2" stroke="#d97706" stroke-width="30" fill="none"&gt;&lt;title&gt;Genie 4 and world models: 4 (%3)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="The DeepSeek storm — 5 videos (%4)" d="M155.5,44.2 A80,80 0 0 1 169.4,56.4" stroke="#dc2626" stroke-width="30" fill="none"&gt;&lt;title&gt;The DeepSeek storm: 5 (%4)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="GPT-6.1 and OpenAI&amp;#x27;s impossible claim — 4 videos (%3)" d="M169.5,56.5 A80,80 0 0 1 178.3,68.3" stroke="#2563eb" stroke-width="30" fill="none"&gt;&lt;title&gt;GPT-6.1 and OpenAI&amp;#x27;s impossible claim: 4 (%3)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Kimi and the raw benchmarks — 3 videos (%2)" d="M178.3,68.4 A80,80 0 0 1 183.4,78.3" stroke="#4d7c0f" stroke-width="30" fill="none"&gt;&lt;title&gt;Kimi and the raw benchmarks: 3 (%2)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Mistral Large 4 and cyber sovereignty — 2 videos (%1)" d="M183.5,78.3 A80,80 0 0 1 186.1,85.2" stroke="#7c3aed" stroke-width="30" fill="none"&gt;&lt;title&gt;Mistral Large 4 and cyber sovereignty: 2 (%1)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="OCR and document intelligence — 11 videos (%8)" d="M186.1,85.3 A80,80 0 0 1 188.5,125.5" stroke="#0284c7" stroke-width="30" fill="none"&gt;&lt;title&gt;OCR and document intelligence: 11 (%8)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Voice AI and speech — 10 videos (%7)" d="M188.4,125.7 A80,80 0 0 1 173.2,159.1" stroke="#92400e" stroke-width="30" fill="none"&gt;&lt;title&gt;Voice AI and speech: 10 (%7)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Small models and efficiency — 19 videos (%14)" d="M173.0,159.3 A80,80 0 0 1 112.2,190.0" stroke="#db2777" stroke-width="30" fill="none"&gt;&lt;title&gt;Small models and efficiency: 19 (%14)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Local hardware and compute — 3 videos (%2)" d="M111.9,190.0 A80,80 0 0 1 100.8,189.5" stroke="#0f766e" stroke-width="30" fill="none"&gt;&lt;title&gt;Local hardware and compute: 3 (%2)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Coding agents and developer tools — 12 videos (%9)" d="M100.7,189.5 A80,80 0 0 1 60.2,172.6" stroke="#475569" stroke-width="30" fill="none"&gt;&lt;title&gt;Coding agents and developer tools: 12 (%9)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Agent orchestration and the LangChain ecosystem — 26 videos (%19)" d="M60.0,172.5 A80,80 0 0 1 33.8,85.7" stroke="#b45309" stroke-width="30" fill="none"&gt;&lt;title&gt;Agent orchestration and the LangChain ecosystem: 26 (%19)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Image and video generation and multimodal — 5 videos (%4)" d="M33.9,85.3 A80,80 0 0 1 41.6,68.5" stroke="#0d9488" stroke-width="30" fill="none"&gt;&lt;title&gt;Image and video generation and multimodal: 5 (%4)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="OpenAI&amp;#x27;s product and enterprise front — 4 videos (%3)" d="M41.7,68.4 A80,80 0 0 1 50.5,56.5" stroke="#d97706" stroke-width="30" fill="none"&gt;&lt;title&gt;OpenAI&amp;#x27;s product and enterprise front: 4 (%3)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Anthropic and the balance of power — 2 videos (%1)" d="M50.5,56.5 A80,80 0 0 1 55.8,51.2" stroke="#dc2626" stroke-width="30" fill="none"&gt;&lt;title&gt;Anthropic and the balance of power: 2 (%1)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Practical AI on the job — 10 videos (%7)" d="M55.8,51.2 A80,80 0 0 1 87.8,33.2" stroke="#2563eb" stroke-width="30" fill="none"&gt;&lt;title&gt;Practical AI on the job: 10 (%7)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Learn and level up — 6 videos (%4)" d="M87.9,33.1 A80,80 0 0 1 109.9,30.0" stroke="#4d7c0f" stroke-width="30" fill="none"&gt;&lt;title&gt;Learn and level up: 6 (%4)&lt;/title&gt;&lt;/path&gt;&lt;circle cx="110" cy="110" r="64" fill="#fff"/&gt;&lt;text x="110" y="106" text-anchor="middle" class="dn"&gt;135&lt;/text&gt;&lt;text x="110" y="126" text-anchor="middle" class="ax"&gt;videos&lt;/text&gt;&lt;/svg&gt;&lt;div class="lgs"&gt;&lt;span class="lg"&gt;&lt;i style="background:#0d9488"&gt;&lt;/i&gt;The Gemini 4 Carbon era &lt;b&gt;9&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#d97706"&gt;&lt;/i&gt;Genie 4 and world models &lt;b&gt;4&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#dc2626"&gt;&lt;/i&gt;The DeepSeek storm &lt;b&gt;5&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#2563eb"&gt;&lt;/i&gt;GPT-6.1 and OpenAI&amp;#x27;s impossible claim &lt;b&gt;4&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#4d7c0f"&gt;&lt;/i&gt;Kimi and the raw benchmarks &lt;b&gt;3&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#7c3aed"&gt;&lt;/i&gt;Mistral Large 4 and cyber sovereignty &lt;b&gt;2&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#0284c7"&gt;&lt;/i&gt;OCR and document intelligence &lt;b&gt;11&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#92400e"&gt;&lt;/i&gt;Voice AI and speech &lt;b&gt;10&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#db2777"&gt;&lt;/i&gt;Small models and efficiency &lt;b&gt;19&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#0f766e"&gt;&lt;/i&gt;Local hardware and compute &lt;b&gt;3&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#475569"&gt;&lt;/i&gt;Coding agents and developer tools &lt;b&gt;12&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#b45309"&gt;&lt;/i&gt;Agent orchestration and the LangChain ecosystem &lt;b&gt;26&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#0d9488"&gt;&lt;/i&gt;Image and video generation and multimodal &lt;b&gt;5&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#d97706"&gt;&lt;/i&gt;OpenAI&amp;#x27;s product and enterprise front &lt;b&gt;4&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#dc2626"&gt;&lt;/i&gt;Anthropic and the balance of power &lt;b&gt;2&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#2563eb"&gt;&lt;/i&gt;Practical AI on the job &lt;b&gt;10&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#4d7c0f"&gt;&lt;/i&gt;Learn and level up &lt;b&gt;6&lt;/b&gt;&lt;/span&gt;&lt;/div&gt;&lt;/div&gt;&lt;/div&gt;
&lt;div class="card" style="margin-bottom:14px"&gt;&lt;h3&gt;Trend timeline — each dot a video · x: publish time · y: log views · color: theme · hover: title&lt;/h3&gt;&lt;svg viewBox="0 0 960 300" role="img"&gt;&lt;line x1="46" y1="264.0" x2="946" y2="264.0" stroke="#d7dce2"/&gt;&lt;text x="38" y="268" text-anchor="end" class="ax"&gt;100&lt;/text&gt;&lt;line x1="46" y1="195.6" x2="946" y2="195.6" stroke="#d7dce2"/&gt;&lt;text x="38" y="200" text-anchor="end" class="ax"&gt;1K&lt;/text&gt;&lt;line x1="46" y1="127.2" x2="946" y2="127.2" stroke="#d7dce2"/&gt;&lt;text x="38" y="131" text-anchor="end" class="ax"&gt;10K&lt;/text&gt;&lt;line x1="46" y1="58.7" x2="946" y2="58.7" stroke="#d7dce2"/&gt;&lt;text x="38" y="63" text-anchor="end" class="ax"&gt;100K&lt;/text&gt;&lt;line x1="102.2" y1="16" x2="102.2" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="102.2" y="288" text-anchor="middle" class="ax"&gt;25 Oct&lt;/text&gt;&lt;line x1="214.8" y1="16" x2="214.8" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="214.8" y="288" text-anchor="middle" class="ax"&gt;26 Oct&lt;/text&gt;&lt;line x1="327.2" y1="16" x2="327.2" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="327.2" y="288" text-anchor="middle" class="ax"&gt;27 Oct&lt;/text&gt;&lt;line x1="439.8" y1="16" x2="439.8" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="439.8" y="288" text-anchor="middle" class="ax"&gt;28 Oct&lt;/text&gt;&lt;line x1="552.2" y1="16" x2="552.2" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="552.2" y="288" text-anchor="middle" class="ax"&gt;29 Oct&lt;/text&gt;&lt;line x1="664.8" y1="16" x2="664.8" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="664.8" y="288" text-anchor="middle" class="ax"&gt;30 Oct&lt;/text&gt;&lt;line x1="777.2" y1="16" x2="777.2" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="777.2" y="288" text-anchor="middle" class="ax"&gt;31 Oct&lt;/text&gt;&lt;line x1="889.8" y1="16" x2="889.8" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="889.8" y="288" text-anchor="middle" class="ax"&gt;01 Nov&lt;/text&gt;&lt;circle cx="613.2" cy="97.4" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;I Trained Qwen 4B with Distillation to Catch Qwen 27B - Prompt Engineering (27K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="524.1" cy="16.0" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Carbon: The New Frontier - Prompt Engineering (421K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="500.7" cy="100.2" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This New AI Architecture Is 31x Faster (NCM) - Prompt Engineering (25K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="322.6" cy="59.9" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Qwen Calling Home? - Prompt Engineering (96K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="275.7" cy="114.0" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Harness² Explained in 60 Seconds - Prompt Engineering (15K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="163.2" cy="69.5" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Kimi K3.1 Pro and Opus 5.5 Made the Same 13 Apps (Raw Results) - Prompt Engineering (69K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="917.9" cy="261.7" r="4.5" fill="#7c3aed" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Mistral Large 4 Is INSANE: 1 TRILLION Parameters, Open Weights, And My GPU Says NOPE - Prompt Engineer 48 (8)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="899.1" cy="247.2" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Youtu-Parsing-Omni: 1 AWESOME 5B Model, 5 Inputs, 1 JSON (I Broke It On a Chart) - Prompt Engineer 48 (76)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="880.4" cy="214.2" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;LFM3 on Ollama: 20 INSANE Real Tests (1 GB Model, 141 tok/s, 1 Epic FAIL) - Prompt Engineer 48 (434)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="856.9" cy="200.4" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;DeepTutor&amp;#x27;s CRAZY Open-Source AI Tutor Runs With NO API KEY - I Tested It - Prompt Engineer 48 (749)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="838.2" cy="228.4" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;INSANE: A Decision Model That KNOWS When It&amp;#x27;s Wrong (Lumma-Fev-v2 Tested) - Prompt Engineer 48 (231)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="814.8" cy="220.7" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;VibeGame: 8 AI Agents Build AND PLAY Their Own Game - CRAZY Claim, I Checked the Paper - Prompt Engineer 48 (330)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="791.3" cy="216.1" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How to Make AI Videos for Beginners: INSANE Results From 3 Prompts - Prompt Engineer 48 (401)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="735.1" cy="210.0" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This 2B Voice AI Runs LOCALLY at 158ms — CRAZY Fast, WILD Mistakes (llama.cpp Realtime Test) - Prompt Engineer 48 (515)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="646.0" cy="205.4" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This 278M Telugu Model Is CRAZY Good... I Tested It On My Laptop - Prompt Engineer 48 (619)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="585.1" cy="193.7" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This Tiny 1.3B AI Reads Documents INSANELY Well? I Tested It on 30 Pages (RTX 4060) - Prompt Engineer 48 (965)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="547.6" cy="163.7" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Liquid AI&amp;#x27;s LFM3 Is CRAZY: 1.7B Model vs 9B? I Tested It on the Live API - Prompt Engineer 48 (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="533.5" cy="209.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Google&amp;#x27;s Genie 4 Is INSANE: A World That Makes Its OWN SOUND (I Read Everything) - Prompt Engineer 48 (529)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="514.8" cy="222.1" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;LEANN: 201 GB vs 6 GB?! I Installed This INSANE Vector DB - Prompt Engineer 48 (309)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="496.0" cy="210.9" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;(70% OFF) I Tested 10 AI Models on ONE Coding Task — The Price Gap Is INSANE (One API, Real Costs) - Prompt Engineer 48 (498)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="477.2" cy="202.5" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This 10 MILLION Token AI Is CRAZY... But Is It Real? (Mirendil Fern) - Prompt Engineer 48 (693)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="725.7" cy="79.1" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;3,000 Tokens/Sec Locally: Testing the ZGX Fury - Sam Witteveen (50K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="275.7" cy="118.8" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;100% Open Image Model Lets You Place Text With JSON - Sam Witteveen (13K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="847.6" cy="91.4" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Avalon Just Beat Strata by 47% on AMD (9x Faster) - Kai (33K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="735.1" cy="117.4" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Bytes Is This The End of Token? - Kai (14K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="622.6" cy="70.3" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Every Way to Get 256 GB of VRAM for Local AI at Full Context - Kai (68K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="510.1" cy="80.6" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Your $200 ChatGPT Plan is now worth less! - Kai (48K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="397.6" cy="50.1" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Your GPU Is Nerfed for Local AI (NVIDIA&amp;#x27;s Hidden Limit) - Kai (133K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="285.1" cy="103.0" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Your PS5 Can Now Run Open Models (And It&amp;#x27;s Surprisingly Fast) - Kai (22K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="172.6" cy="75.7" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Is Flint the Killer of Qwen 3.8 27B? (First RSI Model) - Kai (56K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="406.9" cy="152.7" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;OpenCode Go + CodeGraph: Can Cheap Models Do the Job? - The Gray Cat (4K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="866.3" cy="199.5" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How To Grow On Instagram Step-By-Step - Howfinity (777)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="856.9" cy="192.4" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Turn a PDF into an Excel file with Claude - Howfinity (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="753.8" cy="195.9" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How to Build A Newsletter In 10 Minutes - Howfinity (890)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="636.6" cy="195.9" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;LinkedIn Sales Navigator Tutorial - Howfinity (889)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="533.5" cy="190.8" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;You&amp;#x27;re Prompting Claude Wrong #claude - Howfinity (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="519.4" cy="189.5" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How To Translate A PDF Into Different Languages With AI - Howfinity (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="416.3" cy="178.2" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How To Run Meta Ads For Beginners - Howfinity (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="308.5" cy="182.5" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How To Build AI Chatbot For Your Website - Howfinity (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="641.3" cy="171.1" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Two AI bots for spam calls and data privacy - Matthew Berman (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="552.2" cy="50.9" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Google &amp;quot;Carbon&amp;quot; just won everything... - Matthew Berman (130K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="542.9" cy="95.9" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Google announces Gemini 4 Carbon - Matthew Berman (29K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="453.8" cy="48.4" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Deepseek crushed everyone - Matthew Berman (141K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="411.6" cy="124.7" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;I Ranked 5 Ways to Wake Up - Matthew Berman (11K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="397.6" cy="125.1" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Anthropic’s IPO: Who Gets the Voting Power? - Matthew Berman (11K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="331.9" cy="55.5" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This is much more concerning than you think... - Matthew Berman (111K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="205.4" cy="112.2" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Kimi K3.1 Passes the Rubik&amp;#x27;s Cube Test - Matthew Berman (16K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="758.5" cy="58.1" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;There are many futures. Which do you want to build? - OpenAI (102K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="735.1" cy="168.2" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How Gen uses ChatGPT Work to turn data Into decisions | OpenAI - OpenAI (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="735.1" cy="119.9" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How Gen uses ChatGPT Work to turn data Into decisions | OpenAI - OpenAI (13K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="655.4" cy="159.3" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;ChatGPT Enterprise at Zendesk: Putting Company Data to Work - OpenAI (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="627.2" cy="116.3" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;ChatGPT Enterprise at Zendesk: Putting Company Data to Work - OpenAI (14K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="547.6" cy="77.0" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;API credits, now in ChatGPT Pro and Business plans - OpenAI (54K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="491.3" cy="134.8" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Merge CEO Shensi Ding: AI fluency starts at the top - OpenAI (8K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="491.3" cy="120.1" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Why Merge CEO Shensi Ding stays hands-on with Codex - OpenAI (13K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="481.9" cy="128.4" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;From legal bottleneck to business partner | Legora’s Emma Kadri Bergström - OpenAI (9K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="481.9" cy="127.1" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Building the future of legal work with AI | Legora CEO Max Junestrand - OpenAI (10K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="481.9" cy="134.9" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Faster contract review. Broader legal analysis. | Legora’s Alex Fortescue-Webb - OpenAI (8K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="481.9" cy="110.2" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How Legora is transforming legal work with OpenAI - OpenAI (18K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="397.6" cy="144.6" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Lawyers Stay in Control | AT&amp;amp;T + OpenAI - OpenAI (5K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="397.6" cy="108.9" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Getting a Head Start with Codex | Mitchell Hashimoto - OpenAI (18K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="397.6" cy="126.0" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;From Ideas to Prototypes with Codex | Shopify’s Mikhail Parakhin - OpenAI (10K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="814.8" cy="224.6" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Phonon-2 (164 MB) Is INSANE! 197× Real-Time Speech Recognition Tested - AI WITH Rithesh (277)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="796.0" cy="220.9" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Xiaomi-OCR-0 is HERE! 0.8B Open Weights OCR Model Tested! - AI WITH Rithesh (327)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="702.2" cy="210.2" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Surya OCR 2.5 (689M) IS HERE ! Tested Locally! INSANE Accuracy… But Some Surprises - AI WITH Rithesh (511)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="678.8" cy="198.9" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;LFM3-1.7B  Small Model for Agents - AI WITH Rithesh (793)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="585.1" cy="207.7" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Carbon is #1... but you can&amp;#x27;t use it - AI WITH Rithesh (564)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="575.7" cy="219.1" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Baidu&amp;#x27;s Unlimited OCR 2 is HERE - AI WITH Rithesh (353)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="575.7" cy="214.1" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Carbon Is HERE… No:1 Frontier Model  But You CAN’T Use It Yet! - AI WITH Rithesh (437)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="510.1" cy="172.0" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Liquid AI LFM3 Audio &amp;amp; Omni Tested! 2B Models That Listen, Talk &amp;amp; Use Tools - AI WITH Rithesh (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="496.0" cy="202.4" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;OpenBMB MiniCPM-OCR 1.0  and   MiniCPM-OCR Jev Is HERE!  1.3B OCR + 2B decision model  Tested - AI WITH Rithesh (696)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="477.2" cy="164.8" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Liquid AI LFM3-1.7B IS HERE : For LOCAL AGENTS 256K Context Tested Locally - AI WITH Rithesh (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="463.2" cy="165.3" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Liquid AI LFM3-VL-2B Small Vision  Language Model Is HERE! Punches Above Its Class? - AI WITH Rithesh (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="458.5" cy="209.5" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Google Genie 4 Is INSANE! Step Inside AI-Generated Worlds in Real Time - AI WITH Rithesh (525)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="378.8" cy="139.3" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Baidu Unlimited OCR 2 Tested Locally! 3B Model Handles 100+ Pages - AI WITH Rithesh (7K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="350.7" cy="194.4" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;GPT-6.1 TERRA: REAL USERS, REAL TESTS! - AI WITH Rithesh (942)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="261.6" cy="142.8" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Nanbeige 4.3 33B Claims to Beat Qwen 3.8 27B! Tested Locally on 24GB RAM - AI WITH Rithesh (6K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="627.2" cy="121.5" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;ComfyUI Prompts Explained: CLIP Text Encode, Negative and Dynamic Prompts - pixaroma (12K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="402.2" cy="103.9" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;MOSS Image in ComfyUI | Text to Image, Words &amp;amp; Layout (Ep40) - pixaroma (22K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="303.8" cy="106.2" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;MOSS Image - With Control, References &amp;amp; Editing in ComfyUI! - Nerdy Rodent (20K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="730.4" cy="142.7" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Give Me 42 Minutes, I&amp;#x27;ll Make LLM Post-Training Click Forever - Zachary Huang (6K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="392.9" cy="108.1" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Give Me 1 Hour, I Will Make Artificial Intelligence Click Forever - Zachary Huang (19K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="852.2" cy="132.4" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Anthropic&amp;#x27;s New Study: You&amp;#x27;re Not Controlling AI as Much as You Think You Are - Cole Medin (8K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="552.2" cy="116.9" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;The 4 Components of Every Successful AI Second Brain (Build These Now) - Cole Medin (14K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="327.2" cy="166.0" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Copilot Automations Run While You Sleep #ai #coding #githubcopilot - Burke Holland (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="622.6" cy="136.5" r="4.5" fill="#7c3aed" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Science Behind the Open Cyber Frontier with Mistral Large 4 - Hugging Face (7K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="280.4" cy="93.9" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Workshop: Training Agents at the frontier of AI Industry - Hugging Face (31K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="275.7" cy="166.1" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Hugging Face Journal Club: Score Centering Stabilizes Off-policy Reinforcement Learning - Hugging Face (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="735.1" cy="16.6" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;OpenAI Claims It Just Did The Impossible - Two Minute Papers (413K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="519.4" cy="42.2" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;DeepSeek&amp;#x27;s New AI Is An Absolute Monster - Two Minute Papers (175K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="875.7" cy="253.4" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;What if the best AI tutor for children was designed to talk less? - Agentic Buildcamp: Adaptive Entrepreneurship (43)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="763.2" cy="242.1" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Mom Test Bot: Build What Customers Actually Need, Not What They Say They Want - Agentic Buildcamp: Adaptive Entrepreneurship (109)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="650.7" cy="238.0" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;From Research Project to AI Startup: The Part Nobody Warns You About - Agentic Buildcamp: Adaptive Entrepreneurship (140)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="435.1" cy="238.4" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;$100 in Ads. Zero Sales. One Painful Lesson. - Agentic Buildcamp: Adaptive Entrepreneurship (137)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="425.7" cy="241.5" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;No Hallucination: AI Research Companion for Students - Agentic Buildcamp: Adaptive Entrepreneurship (113)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="313.2" cy="237.4" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;I Built an AI Personal Concierge for People Over 65 - Agentic Buildcamp: Adaptive Entrepreneurship (145)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="730.4" cy="137.7" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Why Own Your Intelligence? Harrison Chase Keynote | Interrupt London - LangChain (7K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="716.3" cy="176.1" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;The 50/50 problem with voice AI nobody is talking about #Shorts - LangChain (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="641.3" cy="172.0" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;The Hardest Part of Testing Agents: Write Access - LangChain (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="631.9" cy="162.2" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Building Semiconductor Agents at Skyworks with LangGraph + LangSmith - LangChain (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="533.5" cy="161.4" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;The Real Cost of &amp;quot;We Can Just Build It Ourselves&amp;quot; - LangChain (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="500.7" cy="122.7" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;What is LangChain? - LangChain (12K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="402.2" cy="73.6" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How to Build Agents That Improve Over Time: Model, Harness, Context | Interrupt London - LangChain (61K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="346.0" cy="163.0" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;We wired Devin into our Slack to triage messages - LangChain (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="294.4" cy="125.3" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;LangGraph explained in 1 minute - LangChain (11K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="275.7" cy="188.3" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Tool Problem or Model Problem? How Clay Tells the Difference - LangChain (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="275.7" cy="151.3" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Inside Managed Deep Agents: mda CLI, Context Hub, and Harbor Evals | Interrupt London - LangChain (4K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="594.4" cy="189.0" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;02 - Multimodal Spatial Biology: Registration - DigitalSreeni (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="369.4" cy="176.5" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;01 - Multimodal Spatial Biology: Introducing the Dataset - DigitalSreeni (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="814.8" cy="234.9" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Dialt On-Prem: Live Voice AI on One GPU #Shorts - Trelis Research (166)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="702.2" cy="225.3" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How AI Learned to Hill-Climb Intelligence #Shorts - Trelis Research (268)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="702.2" cy="216.2" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Dialt On-Prem: Custom Voice Assistants on Your Own GPU - Trelis Research (399)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="622.6" cy="204.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How AI Learned to Hill-Climb Intelligence - Trelis Research (643)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="369.4" cy="206.3" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Voice AI That Is Distinctly Not Human #Shorts - Trelis Research (596)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="256.9" cy="214.1" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Voice AI That Is Distinctly Not Human - Trelis Research (437)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="510.1" cy="152.7" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Join Andrew Ng at AI Dev NYC for two days of technical sessions and workshops. - DeepLearningAI (4K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="491.3" cy="145.7" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Agents on longer tasks need more than one loop - DeepLearningAI (5K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="317.9" cy="173.5" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;7 AI code review best practices #deeplearningai #qodo - DeepLearningAI (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="285.1" cy="180.7" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Learn from OpenAI, Hugging Face, Andrew Ng and Yann LeCun | AI Dev NYC #aidev #shorts - DeepLearningAI (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="397.6" cy="154.5" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Unlimited OCR 2 is here! - 1littlecoder (4K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="158.5" cy="84.9" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This question inspired LoRA fine-tuning - Julia Turc (41K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="772.6" cy="195.3" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;NVIDIA VP: Same AI Model, 30% to 100% (Must Know) - Mervin Praison (909)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="617.9" cy="185.0" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Nemotron 3.5 Transcribe BEATS Whisper? (I Tested It) - Mervin Praison (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="556.9" cy="176.8" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Carbon: Google&amp;#x27;s Best AI and You CAN&amp;#x27;T Use It - Mervin Praison (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="449.1" cy="172.6" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;DeepSeek in Claude Code (and the Engine They Gave Away) - Mervin Praison (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="317.9" cy="192.4" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;GPT-6.1 Terra: half the price, fewer mistakes - Mervin Praison (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="313.2" cy="199.8" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Why your AI agent stops following your rules - Mervin Praison (768)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="744.4" cy="81.4" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Full-Stack AI Engineering Course (FastAPI, LLMs, Docker, Supabase) - Dave Ebbelaar (46K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="308.5" cy="59.1" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This Free Repo Lets Claude Code Build Any Mac App - Dave Ebbelaar (99K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="646.0" cy="124.2" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Accelerate shipping to scale with Google Antigravity - Google for Developers (11K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="556.9" cy="173.1" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;&#129716; Building an autonomous plant rover with Gemini Live - Google for Developers (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="538.2" cy="87.9" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Building real-time vision &amp;amp; edge navigation with Gemini Live - Google for Developers (37K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="510.1" cy="101.2" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Functional smart devices using Gemini 4 Argon and Google Antigravity - Google for Developers (24K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="402.2" cy="74.9" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Introducing Genie 4 - Google for Developers (58K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="496.0" cy="115.2" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Support Vector Machines (SVM): Concepts and Code - StatQuest with Josh Starmer (15K)&lt;/title&gt;&lt;/circle&gt;&lt;/svg&gt;&lt;/div&gt;
&lt;div class="rowg g11"&gt;
&lt;div class="card"&gt;&lt;h3&gt;Theme density&lt;/h3&gt;&lt;div class="bars"&gt;&lt;a class="bar" href="#t1"&gt;&lt;span class="bn"&gt;The Gemini 4 Carbon era&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:35%;background:#0d9488"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;9&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t2"&gt;&lt;span class="bn"&gt;Genie 4 and world models&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:15%;background:#d97706"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;4&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t3"&gt;&lt;span class="bn"&gt;The DeepSeek storm&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:19%;background:#dc2626"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;5&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t4"&gt;&lt;span class="bn"&gt;GPT-6.1 and OpenAI&amp;#x27;s impossible claim&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:15%;background:#2563eb"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;4&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t5"&gt;&lt;span class="bn"&gt;Kimi and the raw benchmarks&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:12%;background:#4d7c0f"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;3&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t6"&gt;&lt;span class="bn"&gt;Mistral Large 4 and cyber sovereignty&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:8%;background:#7c3aed"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;2&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t7"&gt;&lt;span class="bn"&gt;OCR and document intelligence&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:42%;background:#0284c7"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;11&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t8"&gt;&lt;span class="bn"&gt;Voice AI and speech&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:38%;background:#92400e"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;10&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t9"&gt;&lt;span class="bn"&gt;Small models and efficiency&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:73%;background:#db2777"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;19&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t10"&gt;&lt;span class="bn"&gt;Local hardware and compute&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:12%;background:#0f766e"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;3&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t11"&gt;&lt;span class="bn"&gt;Coding agents and developer tools&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:46%;background:#475569"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;12&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t12"&gt;&lt;span class="bn"&gt;Agent orchestration and the LangChain ecosystem&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:100%;background:#b45309"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;26&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t13"&gt;&lt;span class="bn"&gt;Image and video generation and multimodal&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:19%;background:#0d9488"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;5&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t14"&gt;&lt;span class="bn"&gt;OpenAI&amp;#x27;s product and enterprise front&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:15%;background:#d97706"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;4&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t15"&gt;&lt;span class="bn"&gt;Anthropic and the balance of power&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:8%;background:#dc2626"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;2&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t16"&gt;&lt;span class="bn"&gt;Practical AI on the job&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:38%;background:#2563eb"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;10&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t17"&gt;&lt;span class="bn"&gt;Learn and level up&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:23%;background:#4d7c0f"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;6&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/div&gt;
&lt;div class="card"&gt;&lt;h3&gt;Most-viewed channels&lt;/h3&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:100%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;654K&lt;i&gt;6 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Two Minute Papers&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:90%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;587K&lt;i&gt;2 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Matthew Berman&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:69%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;451K&lt;i&gt;8 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Kai&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:57%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;375K&lt;i&gt;7 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;OpenAI&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:44%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;288K&lt;i&gt;15 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Dave Ebbelaar&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:22%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;145K&lt;i&gt;2 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;/div&gt;&lt;/div&gt;
&lt;section class="theme" id="t1"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#0d9488"&gt;01&lt;/span&gt;&lt;i class="dot" style="background:#0d9488"&gt;&lt;/i&gt;The Gemini 4 Carbon era&lt;span class="cnt"&gt;9 videos &amp;middot; 609K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Google announced its second frontier model in two weeks - Carbon following Argon - and argues it has overtaken its rivals on cyber-defence benchmarks. Channel headlines converged on the same point: the model is number one, but the public cannot use it yet.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=kMpTBy3g1N0" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;421K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Carbon: The New Frontier&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=zhnJo7s91ls" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;130K&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Google &amp;quot;Carbon&amp;quot; just won everything...&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=3yZb9rGt-rk" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;29K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Google announces Gemini 4 Carbon&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=pQ_Dnr9Q4k8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;24K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;Functional smart devices using Gemini 4 Argon and Google Antigravity&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=iN5fscBeTDE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;Mervin Praison&lt;/span&gt;&lt;span class="tt"&gt;DeepSeek in Claude Code (and the Engine They Gave Away)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=DXfXXB8Murs" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;Mervin Praison&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Carbon: Google&amp;#x27;s Best AI and You CAN&amp;#x27;T Use It&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=G9Q4UlkyEy0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;564&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Carbon is #1... but you can&amp;#x27;t use it&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=LfYPH46SZfw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;525&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Google Genie 4 Is INSANE! Step Inside AI-Generated Worlds in Real Time&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=mcZ2rtmHa98" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;437&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Carbon Is HERE… No:1 Frontier Model  But You CAN’T Use It Yet!&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t2"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#d97706"&gt;02&lt;/span&gt;&lt;i class="dot" style="background:#d97706"&gt;&lt;/i&gt;Genie 4 and world models&lt;span class="cnt"&gt;4 videos &amp;middot; 59K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Google&amp;#x27;s world model Genie 4 was the week&amp;#x27;s most talked-about demo, with real-time explorable simulations that generate their own audio, and the debate over an intelligence &amp;#x27;hill-climb&amp;#x27; grew within the same cluster.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=LOP8ul3KqQI" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;58K&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;Introducing Genie 4&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=1mHYqC4QZAI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;643&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;How AI Learned to Hill-Climb Intelligence&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=pRB5KUhRr2U" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;529&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Google&amp;#x27;s Genie 4 Is INSANE: A World That Makes Its OWN SOUND (I Read Everything)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=BWbREb3VRyA" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;268&lt;/span&gt;&lt;span class="dt"&gt;30 Oct&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;How AI Learned to Hill-Climb Intelligence #Shorts&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t3"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#dc2626"&gt;03&lt;/span&gt;&lt;i class="dot" style="background:#dc2626"&gt;&lt;/i&gt;The DeepSeek storm&lt;span class="cnt"&gt;5 videos &amp;middot; 388K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;The DeepSeek v4.1 Flash wave rolled in under &amp;#x27;crushed everyone&amp;#x27; headlines and drew the week&amp;#x27;s biggest viewing crowd, while hooking the model up to Claude Code and the inference engine DeepSeek left open sparked a conversation of its own.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=pryCaGv8wuo" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;175K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;Two Minute Papers&lt;/span&gt;&lt;span class="tt"&gt;DeepSeek&amp;#x27;s New AI Is An Absolute Monster&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=DDZ-a6bAn3Y" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;141K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Deepseek crushed everyone&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=JPHT9AvoPSA" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;68K&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Every Way to Get 256 GB of VRAM for Local AI at Full Context&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=blS3l32vBR0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;4K&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;The Gray Cat&lt;/span&gt;&lt;span class="tt"&gt;OpenCode Go + CodeGraph: Can Cheap Models Do the Job?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=uWM3ZnWYwc0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;498&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;(70% OFF) I Tested 10 AI Models on ONE Coding Task — The Price Gap Is INSANE (One API, Real Costs)&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t4"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#2563eb"&gt;04&lt;/span&gt;&lt;i class="dot" style="background:#2563eb"&gt;&lt;/i&gt;GPT-6.1 and OpenAI&amp;#x27;s impossible claim&lt;span class="cnt"&gt;4 videos &amp;middot; 462K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;The complexity-theory paper OpenAI introduced as &amp;#x27;we did the impossible&amp;#x27; pulled 412,000 views on Two Minute Papers, while GPT-6.1 Terra was put through its paces by real users at half the price and with fewer mistakes.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=CWdTBLsbaQs" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;413K&lt;/span&gt;&lt;span class="dt"&gt;30 Oct&lt;/span&gt;&lt;span class="ch"&gt;Two Minute Papers&lt;/span&gt;&lt;span class="tt"&gt;OpenAI Claims It Just Did The Impossible&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=P0i3vYuJZH8" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;48K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Your $200 ChatGPT Plan is now worth less!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=azRot3GBEh4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;26 Oct&lt;/span&gt;&lt;span class="ch"&gt;Mervin Praison&lt;/span&gt;&lt;span class="tt"&gt;GPT-6.1 Terra: half the price, fewer mistakes&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=6oUJ2pLgsXc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;942&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;GPT-6.1 TERRA: REAL USERS, REAL TESTS!&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t5"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#4d7c0f"&gt;05&lt;/span&gt;&lt;i class="dot" style="background:#4d7c0f"&gt;&lt;/i&gt;Kimi and the raw benchmarks&lt;span class="cnt"&gt;3 videos &amp;middot; 127K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Moonshot&amp;#x27;s Kimi K3.1 cleared the Rubik&amp;#x27;s-cube test, and the raw results of Kimi K3.1 Pro and Opus 5.5 building the same 13 apps became the week&amp;#x27;s most-shared apples-to-apples comparison, stripped of spin.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=_XXX5K-EMoc" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;69K&lt;/span&gt;&lt;span class="dt"&gt;25 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;Kimi K3.1 Pro and Opus 5.5 Made the Same 13 Apps (Raw Results)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ug9el1OGGEw" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;41K&lt;/span&gt;&lt;span class="dt"&gt;25 Oct&lt;/span&gt;&lt;span class="ch"&gt;Julia Turc&lt;/span&gt;&lt;span class="tt"&gt;This question inspired LoRA fine-tuning&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=L18lkQs4PeM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;16K&lt;/span&gt;&lt;span class="dt"&gt;25 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Kimi K3.1 Passes the Rubik&amp;#x27;s Cube Test&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t6"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#7c3aed"&gt;06&lt;/span&gt;&lt;i class="dot" style="background:#7c3aed"&gt;&lt;/i&gt;Mistral Large 4 and cyber sovereignty&lt;span class="cnt"&gt;2 videos &amp;middot; 7K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Mistral Large 4 - one trillion parameters, open weights - planted itself at the center of Europe&amp;#x27;s &amp;#x27;own our intelligence&amp;#x27; argument, and its first stop was open cybersecurity.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=FczbSauVvjw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;7K&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;Science Behind the Open Cyber Frontier with Mistral Large 4&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=SGek-YnySDE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;8&lt;/span&gt;&lt;span class="dt"&gt;01 Nov&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Mistral Large 4 Is INSANE: 1 TRILLION Parameters, Open Weights, And My GPU Says NOPE&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t7"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#0284c7"&gt;07&lt;/span&gt;&lt;i class="dot" style="background:#0284c7"&gt;&lt;/i&gt;OCR and document intelligence&lt;span class="cnt"&gt;11 videos &amp;middot; 66K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Baidu Unlimited OCR 2, Surya 2.5, Xiaomi-OCR-0, MiniCPM-OCR and a run of open OCR models in the 1-3 billion-parameter range landed one after another within a single week, making document handling the hottest arena in the small-model race.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=lEFFa0Z5BIE" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;50K&lt;/span&gt;&lt;span class="dt"&gt;30 Oct&lt;/span&gt;&lt;span class="ch"&gt;Sam Witteveen&lt;/span&gt;&lt;span class="tt"&gt;3,000 Tokens/Sec Locally: Testing the ZGX Fury&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=SEY5sRYlG3s" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;7K&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Baidu Unlimited OCR 2 Tested Locally! 3B Model Handles 100+ Pages&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=OneYJJWDnNA" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;4K&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;1littlecoder&lt;/span&gt;&lt;span class="tt"&gt;Unlimited OCR 2 is here!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=NxN4Yf90ccU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;How To Translate A PDF Into Different Languages With AI&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=D7xswbjd9YM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;31 Oct&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;Turn a PDF into an Excel file with Claude&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=YIEcAKlIp98" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;965&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;This Tiny 1.3B AI Reads Documents INSANELY Well? I Tested It on 30 Pages (RTX 4060)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=MgFW_RkQfzY" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;696&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;OpenBMB MiniCPM-OCR 1.0  and   MiniCPM-OCR Jev Is HERE!  1.3B OCR + 2B decision model  Tested&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=BiLoeGWTLpU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;511&lt;/span&gt;&lt;span class="dt"&gt;30 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Surya OCR 2.5 (689M) IS HERE ! Tested Locally! INSANE Accuracy… But Some Surprises&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=viBr_-DAxgk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;353&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Baidu&amp;#x27;s Unlimited OCR 2 is HERE&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=JFkAbZqzd3Q" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;327&lt;/span&gt;&lt;span class="dt"&gt;31 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Xiaomi-OCR-0 is HERE! 0.8B Open Weights OCR Model Tested!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=-LmJsjTy3XU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;76&lt;/span&gt;&lt;span class="dt"&gt;01 Nov&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Youtu-Parsing-Omni: 1 AWESOME 5B Model, 5 Inputs, 1 JSON (I Broke It On a Chart)&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t8"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#92400e"&gt;08&lt;/span&gt;&lt;i class="dot" style="background:#92400e"&gt;&lt;/i&gt;Voice AI and speech&lt;span class="cnt"&gt;10 videos &amp;middot; 8K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;A 2B local voice model running at 158 ms latency, 164 MB Phonon-2 and a Nemotron-Whisper shootout picked up the pace on the voice line, while human-centered debates - voices that fail to sound human, children&amp;#x27;s education - kept growing.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=fVQVaM95mAo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Liquid AI LFM3 Audio &amp;amp; Omni Tested! 2B Models That Listen, Talk &amp;amp; Use Tools&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=f03D4dkwAdo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;30 Oct&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;The 50/50 problem with voice AI nobody is talking about #Shorts&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=scOCcD0XNB0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;Mervin Praison&lt;/span&gt;&lt;span class="tt"&gt;Nemotron 3.5 Transcribe BEATS Whisper? (I Tested It)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Pj4umL7S03M" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;596&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Voice AI That Is Distinctly Not Human #Shorts&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=0rGb20U0i2U" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;515&lt;/span&gt;&lt;span class="dt"&gt;30 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;This 2B Voice AI Runs LOCALLY at 158ms — CRAZY Fast, WILD Mistakes (llama.cpp Realtime Test)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=gg6waFFnwDs" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;437&lt;/span&gt;&lt;span class="dt"&gt;26 Oct&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Voice AI That Is Distinctly Not Human&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=1ydb3Fyh0fE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;399&lt;/span&gt;&lt;span class="dt"&gt;30 Oct&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Dialt On-Prem: Custom Voice Assistants on Your Own GPU&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=iT1qv7aC_lg" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;277&lt;/span&gt;&lt;span class="dt"&gt;31 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Phonon-2 (164 MB) Is INSANE! 197× Real-Time Speech Recognition Tested&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=_ddgeoQmdWo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;166&lt;/span&gt;&lt;span class="dt"&gt;31 Oct&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Dialt On-Prem: Live Voice AI on One GPU #Shorts&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=YX7Htw9tprs" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;43&lt;/span&gt;&lt;span class="dt"&gt;31 Oct&lt;/span&gt;&lt;span class="ch"&gt;Agentic Buildcamp: Adaptive Entrepreneurship&lt;/span&gt;&lt;span class="tt"&gt;What if the best AI tutor for children was designed to talk less?&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t9"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#db2777"&gt;09&lt;/span&gt;&lt;i class="dot" style="background:#db2777"&gt;&lt;/i&gt;Small models and efficiency&lt;span class="cnt"&gt;19 videos &amp;middot; 305K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Liquid AI&amp;#x27;s LFM3 family, distilled Qwen 4B experiments and videos asking whether the &amp;#x27;token&amp;#x27; is nearing its end all pointed the same way: the week&amp;#x27;s core message was not bigger, but more efficient.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Pn8e474Z5Ps" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;96K&lt;/span&gt;&lt;span class="dt"&gt;26 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;Qwen Calling Home?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=RmE4G6MhOVU" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;56K&lt;/span&gt;&lt;span class="dt"&gt;25 Oct&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Is Flint the Killer of Qwen 3.8 27B? (First RSI Model)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=hOitZIkMT2Y" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;27K&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;I Trained Qwen 4B with Distillation to Catch Qwen 27B&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=_87vrjxx1EE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;25K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;This New AI Architecture Is 31x Faster (NCM)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=TgmC_VZEKTE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;22K&lt;/span&gt;&lt;span class="dt"&gt;26 Oct&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Your PS5 Can Now Run Open Models (And It&amp;#x27;s Surprisingly Fast)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=l4R96NemQS8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;22K&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;pixaroma&lt;/span&gt;&lt;span class="tt"&gt;MOSS Image in ComfyUI | Text to Image, Words &amp;amp; Layout (Ep40)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=hTCjWfbyCjU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;20K&lt;/span&gt;&lt;span class="dt"&gt;26 Oct&lt;/span&gt;&lt;span class="ch"&gt;Nerdy Rodent&lt;/span&gt;&lt;span class="tt"&gt;MOSS Image - With Control, References &amp;amp; Editing in ComfyUI!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=PfR4S3Zzdoo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;14K&lt;/span&gt;&lt;span class="dt"&gt;30 Oct&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Bytes Is This The End of Token?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=5l3l9aEwHlo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;6K&lt;/span&gt;&lt;span class="dt"&gt;30 Oct&lt;/span&gt;&lt;span class="ch"&gt;Zachary Huang&lt;/span&gt;&lt;span class="tt"&gt;Give Me 42 Minutes, I&amp;#x27;ll Make LLM Post-Training Click Forever&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=QptAOS4GIZc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;6K&lt;/span&gt;&lt;span class="dt"&gt;26 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Nanbeige 4.3 33B Claims to Beat Qwen 3.8 27B! Tested Locally on 24GB RAM&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=2Ybdvqt2J4Y" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Liquid AI&amp;#x27;s LFM3 Is CRAZY: 1.7B Model vs 9B? I Tested It on the Live API&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Tyklj_EWX_g" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Liquid AI LFM3-1.7B IS HERE : For LOCAL AGENTS 256K Context Tested Locally&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=8HR_DBj_X2U" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Liquid AI LFM3-VL-2B Small Vision  Language Model Is HERE! Punches Above Its Class?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=DdLFjwbkv_g" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;793&lt;/span&gt;&lt;span class="dt"&gt;30 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;LFM3-1.7B  Small Model for Agents&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=yz4xSmm1D4M" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;693&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;This 10 MILLION Token AI Is CRAZY... But Is It Real? (Mirendil Fern)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=4rIUoadQOjk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;619&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;This 278M Telugu Model Is CRAZY Good... I Tested It On My Laptop&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=FcxT97Dj6QQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;434&lt;/span&gt;&lt;span class="dt"&gt;31 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;LFM3 on Ollama: 20 INSANE Real Tests (1 GB Model, 141 tok/s, 1 Epic FAIL)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ffEi-J4erIw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;309&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;LEANN: 201 GB vs 6 GB?! I Installed This INSANE Vector DB&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Hk23J-hAF6g" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;231&lt;/span&gt;&lt;span class="dt"&gt;31 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;INSANE: A Decision Model That KNOWS When It&amp;#x27;s Wrong (Lumma-Fev-v2 Tested)&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t10"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#0f766e"&gt;10&lt;/span&gt;&lt;i class="dot" style="background:#0f766e"&gt;&lt;/i&gt;Local hardware and compute&lt;span class="cnt"&gt;3 videos &amp;middot; 168K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;The claim that NVIDIA secretly restricts local AI on its GPUs, a guide to fitting 256 GB of VRAM on a roadmap, and the ZGX Fury producing 3,000 tokens/sec made local-run hardware the week&amp;#x27;s most argued-over beat.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=VIm73gaY_hM" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;133K&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Your GPU Is Nerfed for Local AI (NVIDIA&amp;#x27;s Hidden Limit)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=0F63pTS8Gt0" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;33K&lt;/span&gt;&lt;span class="dt"&gt;31 Oct&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Avalon Just Beat Strata by 47% on AMD (9x Faster)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=oAhYP1FwKSY" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;909&lt;/span&gt;&lt;span class="dt"&gt;30 Oct&lt;/span&gt;&lt;span class="ch"&gt;Mervin Praison&lt;/span&gt;&lt;span class="tt"&gt;NVIDIA VP: Same AI Model, 30% to 100% (Must Know)&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t11"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#475569"&gt;11&lt;/span&gt;&lt;i class="dot" style="background:#475569"&gt;&lt;/i&gt;Coding agents and developer tools&lt;span class="cnt"&gt;12 videos &amp;middot; 236K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;A free repo that builds Mac apps with Claude Code, founders burying Codex in their daily workflow, and GitHub Copilot automations: coding agents are leaving the trial phase and moving into everyday work.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=gN6wl-OX8-Q" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;99K&lt;/span&gt;&lt;span class="dt"&gt;26 Oct&lt;/span&gt;&lt;span class="ch"&gt;Dave Ebbelaar&lt;/span&gt;&lt;span class="tt"&gt;This Free Repo Lets Claude Code Build Any Mac App&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=XzjSx7u9mGw" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;46K&lt;/span&gt;&lt;span class="dt"&gt;30 Oct&lt;/span&gt;&lt;span class="ch"&gt;Dave Ebbelaar&lt;/span&gt;&lt;span class="tt"&gt;Full-Stack AI Engineering Course (FastAPI, LLMs, Docker, Supabase)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=lPZQrVFcz3c" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;18K&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Getting a Head Start with Codex | Mitchell Hashimoto&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=yJxZef1Ylok" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;18K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;How Legora is transforming legal work with OpenAI&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=6ep8Vvm4Bdw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;13K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Why Merge CEO Shensi Ding stays hands-on with Codex&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=rsOiZE19RVo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;11K&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;Accelerate shipping to scale with Google Antigravity&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=cflbzHRvGvg" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;10K&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;From Ideas to Prototypes with Codex | Shopify’s Mikhail Parakhin&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=BlbxhsItZPg" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;8K&lt;/span&gt;&lt;span class="dt"&gt;31 Oct&lt;/span&gt;&lt;span class="ch"&gt;Cole Medin&lt;/span&gt;&lt;span class="tt"&gt;Anthropic&amp;#x27;s New Study: You&amp;#x27;re Not Controlling AI as Much as You Think You Are&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=kP26OlW5PBk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;8K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Merge CEO Shensi Ding: AI fluency starts at the top&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=A-UVoSTqwjE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;Burke Holland&lt;/span&gt;&lt;span class="tt"&gt;Copilot Automations Run While You Sleep #ai #coding #githubcopilot&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=CQ_Cfw6QxSk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;26 Oct&lt;/span&gt;&lt;span class="ch"&gt;DeepLearningAI&lt;/span&gt;&lt;span class="tt"&gt;7 AI code review best practices #deeplearningai #qodo&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=RQg3HAkNj9Q" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;109&lt;/span&gt;&lt;span class="dt"&gt;30 Oct&lt;/span&gt;&lt;span class="ch"&gt;Agentic Buildcamp: Adaptive Entrepreneurship&lt;/span&gt;&lt;span class="tt"&gt;Mom Test Bot: Build What Customers Actually Need, Not What They Say They Want&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t12"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#b45309"&gt;12&lt;/span&gt;&lt;i class="dot" style="background:#b45309"&gt;&lt;/i&gt;Agent orchestration and the LangChain ecosystem&lt;span class="cnt"&gt;26 videos &amp;middot; 290K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;The Interrupt London recordings carried the weight here: agents that improve themselves over time, long tasks a single loop cannot finish, and &amp;#x27;own your intelligence&amp;#x27; theses marked agent infrastructure as the new competitive ground.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=xy_7SzwsIjM" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;61K&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;How to Build Agents That Improve Over Time: Model, Harness, Context | Interrupt London&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ujmyzWSj8IU" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;54K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;API credits, now in ChatGPT Pro and Business plans&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=CKFUx7COIwQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;31K&lt;/span&gt;&lt;span class="dt"&gt;26 Oct&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;Workshop: Training Agents at the frontier of AI Industry&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=lcVFk4df5Ms" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;15K&lt;/span&gt;&lt;span class="dt"&gt;26 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;Harness² Explained in 60 Seconds&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=L6pzlo1QvoM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;14K&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;ChatGPT Enterprise at Zendesk: Putting Company Data to Work&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=qIWB1yMsqRo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;14K&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;Cole Medin&lt;/span&gt;&lt;span class="tt"&gt;The 4 Components of Every Successful AI Second Brain (Build These Now)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=VhrtlfTyqp4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;13K&lt;/span&gt;&lt;span class="dt"&gt;26 Oct&lt;/span&gt;&lt;span class="ch"&gt;Sam Witteveen&lt;/span&gt;&lt;span class="tt"&gt;100% Open Image Model Lets You Place Text With JSON&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=rfhhhmajCKc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;12K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;What is LangChain?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=4iYOJ6gyQV4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;11K&lt;/span&gt;&lt;span class="dt"&gt;26 Oct&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;LangGraph explained in 1 minute&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=UOfmD1fY-i8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;10K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Building the future of legal work with AI | Legora CEO Max Junestrand&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=M2Oa42hF67E" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;9K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;From legal bottleneck to business partner | Legora’s Emma Kadri Bergström&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=z-WZMjNsg-k" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;8K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Faster contract review. Broader legal analysis. | Legora’s Alex Fortescue-Webb&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=XOUSuxY074U" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;7K&lt;/span&gt;&lt;span class="dt"&gt;30 Oct&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;Why Own Your Intelligence? Harrison Chase Keynote | Interrupt London&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Zo2oyjbTEXA" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;5K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;DeepLearningAI&lt;/span&gt;&lt;span class="tt"&gt;Agents on longer tasks need more than one loop&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Wb3hQaG_xf0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;4K&lt;/span&gt;&lt;span class="dt"&gt;26 Oct&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;Inside Managed Deep Agents: mda CLI, Context Hub, and Harbor Evals | Interrupt London&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=p1jhV6LlbFY" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;4K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;DeepLearningAI&lt;/span&gt;&lt;span class="tt"&gt;Join Andrew Ng at AI Dev NYC for two days of technical sessions and workshops.&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=GNYHDltnwZQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;ChatGPT Enterprise at Zendesk: Putting Company Data to Work&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=-GbU416ZMoI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;The Real Cost of &amp;quot;We Can Just Build It Ourselves&amp;quot;&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=O3jCAZbrRHM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;Building Semiconductor Agents at Skyworks with LangGraph + LangSmith&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=GNu-d97XY6c" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;We wired Devin into our Slack to triage messages&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=oY1htTx_h9U" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;The Hardest Part of Testing Agents: Write Access&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=VgF5aP0amXA" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;26 Oct&lt;/span&gt;&lt;span class="ch"&gt;DeepLearningAI&lt;/span&gt;&lt;span class="tt"&gt;Learn from OpenAI, Hugging Face, Andrew Ng and Yann LeCun | AI Dev NYC #aidev #shorts&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=y6rrMFO7tgA" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;26 Oct&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;Tool Problem or Model Problem? How Clay Tells the Difference&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=uwALRjyYjjs" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;768&lt;/span&gt;&lt;span class="dt"&gt;26 Oct&lt;/span&gt;&lt;span class="ch"&gt;Mervin Praison&lt;/span&gt;&lt;span class="tt"&gt;Why your AI agent stops following your rules&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=MeEmWV2-Nzw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;330&lt;/span&gt;&lt;span class="dt"&gt;31 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;VibeGame: 8 AI Agents Build AND PLAY Their Own Game - CRAZY Claim, I Checked the Paper&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=N_oRytMcOcc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;113&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;Agentic Buildcamp: Adaptive Entrepreneurship&lt;/span&gt;&lt;span class="tt"&gt;No Hallucination: AI Research Companion for Students&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t13"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#0d9488"&gt;13&lt;/span&gt;&lt;i class="dot" style="background:#0d9488"&gt;&lt;/i&gt;Image and video generation and multimodal&lt;span class="cnt"&gt;5 videos &amp;middot; 53K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;MOSS Image and ComfyUI workflows pushed image generation with control over text placement, while Gemini Live calls and edge-device navigation showed multimodal systems stepping out of the lab and into the field.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ID6vEy94j7c" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;37K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;Building real-time vision &amp;amp; edge navigation with Gemini Live&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=TFJ2yotzJVc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;12K&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;pixaroma&lt;/span&gt;&lt;span class="tt"&gt;ComfyUI Prompts Explained: CLIP Text Encode, Negative and Dynamic Prompts&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=cv8tEbKtEvk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;&#129716; Building an autonomous plant rover with Gemini Live&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=yiE2l_NMj7Y" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;How To Run Meta Ads For Beginners&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=P6UNRh2pZBs" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;401&lt;/span&gt;&lt;span class="dt"&gt;31 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;How to Make AI Videos for Beginners: INSANE Results From 3 Prompts&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t14"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#d97706"&gt;14&lt;/span&gt;&lt;i class="dot" style="background:#d97706"&gt;&lt;/i&gt;OpenAI&amp;#x27;s product and enterprise front&lt;span class="cnt"&gt;4 videos &amp;middot; 123K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;API credits added to ChatGPT plans and enterprise reference videos from Zendesk, Legora and AT&amp;amp;T showed OpenAI spending the week on the enterprise sales pitch far more than on consumer product.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=o4D1J1YWjhg" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;102K&lt;/span&gt;&lt;span class="dt"&gt;30 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;There are many futures. Which do you want to build?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Oi9DHk57KeQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;13K&lt;/span&gt;&lt;span class="dt"&gt;30 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;How Gen uses ChatGPT Work to turn data Into decisions | OpenAI&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Yim7JrVzSkk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;5K&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Lawyers Stay in Control | AT&amp;amp;T + OpenAI&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=7TtGMileCus" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;30 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;How Gen uses ChatGPT Work to turn data Into decisions | OpenAI&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t15"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#dc2626"&gt;15&lt;/span&gt;&lt;i class="dot" style="background:#dc2626"&gt;&lt;/i&gt;Anthropic and the balance of power&lt;span class="cnt"&gt;2 videos &amp;middot; 122K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;The voting-power debate around a possible Anthropic IPO, along with the study showing you are not controlling AI as much as you think you are, heated up the control-and-governance agenda.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=JI3bbZNdyeA" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;111K&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;This is much more concerning than you think...&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=jeUgtyvvkfs" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;11K&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Anthropic’s IPO: Who Gets the Voting Power?&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t16"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#2563eb"&gt;16&lt;/span&gt;&lt;i class="dot" style="background:#2563eb"&gt;&lt;/i&gt;Practical AI on the job&lt;span class="cnt"&gt;10 videos &amp;middot; 18K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;End-to-end &amp;#x27;small AI business&amp;#x27; builds - PDF translation to launching a newsletter, a personal assistant for people over 65 to an English tutor for kids - formed the steady undercurrent on the production side.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=mMzaL15KFVY" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;11K&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;I Ranked 5 Ways to Wake Up&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=IrgQdsxzRa4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Two AI bots for spam calls and data privacy&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=JMaQ_wVRzjc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;26 Oct&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;How To Build AI Chatbot For Your Website&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Z4okNQnD5c0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;890&lt;/span&gt;&lt;span class="dt"&gt;30 Oct&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;How to Build A Newsletter In 10 Minutes&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=zbNNzPwJLbY" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;889&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;LinkedIn Sales Navigator Tutorial&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=KEA1wOlZrLI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;777&lt;/span&gt;&lt;span class="dt"&gt;31 Oct&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;How To Grow On Instagram Step-By-Step&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Mu2vchtZoE4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;749&lt;/span&gt;&lt;span class="dt"&gt;31 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;DeepTutor&amp;#x27;s CRAZY Open-Source AI Tutor Runs With NO API KEY - I Tested It&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=EtN_aEKjvHc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;145&lt;/span&gt;&lt;span class="dt"&gt;26 Oct&lt;/span&gt;&lt;span class="ch"&gt;Agentic Buildcamp: Adaptive Entrepreneurship&lt;/span&gt;&lt;span class="tt"&gt;I Built an AI Personal Concierge for People Over 65&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Kg3e1iNh6do" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;140&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;Agentic Buildcamp: Adaptive Entrepreneurship&lt;/span&gt;&lt;span class="tt"&gt;From Research Project to AI Startup: The Part Nobody Warns You About&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=aGseoDcCqLI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;137&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;Agentic Buildcamp: Adaptive Entrepreneurship&lt;/span&gt;&lt;span class="tt"&gt;$100 in Ads. Zero Sales. One Painful Lesson.&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t17"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#4d7c0f"&gt;17&lt;/span&gt;&lt;i class="dot" style="background:#4d7c0f"&gt;&lt;/i&gt;Learn and level up&lt;span class="cnt"&gt;6 videos &amp;middot; 40K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Long-form explainers promising to make AI click in an hour, LoRA fine-tuning notes and the Hugging Face journal club made educational content the week&amp;#x27;s second busiest lane by volume.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=fCEvvkAbJOs" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;19K&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;Zachary Huang&lt;/span&gt;&lt;span class="tt"&gt;Give Me 1 Hour, I Will Make Artificial Intelligence Click Forever&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Xl-yulnYZmI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;15K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;StatQuest with Josh Starmer&lt;/span&gt;&lt;span class="tt"&gt;Support Vector Machines (SVM): Concepts and Code&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=mEtRe_KKYm0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;26 Oct&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;Hugging Face Journal Club: Score Centering Stabilizes Off-policy Reinforcement Learning&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=KdpN5kAX8Rg" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;27 Oct&lt;/span&gt;&lt;span class="ch"&gt;DigitalSreeni&lt;/span&gt;&lt;span class="tt"&gt;01 - Multimodal Spatial Biology: Introducing the Dataset&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=m6Rdp6macAw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;29 Oct&lt;/span&gt;&lt;span class="ch"&gt;DigitalSreeni&lt;/span&gt;&lt;span class="tt"&gt;02 - Multimodal Spatial Biology: Registration&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=F4hQ6alHtN4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;28 Oct&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;You&amp;#x27;re Prompting Claude Wrong #claude&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;
&lt;footer&gt;This report was compiled by &lt;b style="color:#111"&gt;Murat Karakaya Academy&lt;/b&gt; · Data: YouTube RSS (live, no API key) · window: 2026-10-25 .. 2026-11-01 (UTC, computed at run time) · max 15 videos per channel · The Prompt Engineer 48, OpenAI and AI WITH Rithesh feeds do not cover the full period because of the 15-video cap in RSS; these channels may be undercounted. · render.py @ 01.11.2026 08:58 UTC&lt;/footer&gt;
&lt;/div&gt;&lt;/body&gt;&lt;/html&gt;</description><link>https://www.muratkarakaya.net/2026/11/youtube-ai-weekly-trend-report-october.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-1081191958130887911</guid><pubDate>Sun, 01 Nov 2026 08:22:19 +0000</pubDate><atom:updated>2026-11-01T12:13:09.741+03:00</atom:updated><title>YouTube AI Haftalık Trend Raporu (25 Ekim → 1 Kasım 2026)</title><description>&lt;!doctype html&gt;&lt;html lang="tr"&gt;&lt;head&gt;&lt;meta charset="utf-8"&gt;
&lt;meta name="viewport" content="width=device-width,initial-scale=1"&gt;
&lt;title&gt;YouTube AI Haftalık Trend Raporu&lt;/title&gt;&lt;style&gt;
:root{--bg:transparent;--sf:#fff;--sf2:#eef1f4;--ln:#d7dce2;--ink:#111;--mu:#555;
--ac:#00796b;--am:#b45309;--mono:ui-monospace,Consolas,monospace}
*{box-sizing:border-box;margin:0;padding:0}
body{background:var(--bg);color:var(--ink);font:15px/1.5 -apple-system,'Segoe UI',Roboto,sans-serif;padding:28px 24px 60px}
.wrap{max-width:1100px}
h1{font-size:32px;letter-spacing:.3px;text-align:center;margin-top:8px} .datebar{text-align:center;margin:12px 0 6px}
.dtc{display:inline-block;font:600 15px/1 var(--mono);color:#0b0e12;background:#35d0ba;padding:7px 16px;border-radius:999px}
.meta{color:var(--mu);font-family:var(--mono);font-size:13.5px;margin-top:2px;text-align:center}
.dseg{cursor:pointer;transition:stroke-width .12s} .dseg:hover{stroke-width:36}
#tip{position:fixed;left:0;top:0;z-index:9;pointer-events:none;background:#f6f7f8;border:1px solid #c9d0d8;color:#111;font:12.5px/1.45 var(--mono);padding:7px 11px;border-radius:8px;opacity:0;transition:opacity .12s;max-width:340px;box-shadow:0 6px 20px rgba(0,0,0,.15)}
.meta b{color:var(--ac);font-weight:600}
nav{margin:14px 0 26px;display:flex;flex-wrap:wrap;gap:8px}
nav a{color:var(--mu);text-decoration:none;font-size:12px;padding:4px 10px;border:1px solid var(--ln);border-radius:999px;line-height:1.35} nav a b{color:var(--ac);font-family:var(--mono);font-weight:600;margin-left:4px}
nav a:hover{color:var(--ink);border-color:var(--ac)}
.card{background:var(--sf);border:1px solid var(--ln);border-radius:12px;padding:16px 18px}
.card h3,.stats h3{font-size:12px;text-transform:uppercase;letter-spacing:1.2px;color:var(--mu);margin-bottom:10px;font-weight:600}
.top{display:grid;grid-template-columns:2fr 1fr;gap:14px;margin-bottom:14px}
.champ{background:linear-gradient(135deg,#f7f9fa,#eef2f5);border:1px solid #d7dce2;border-radius:12px;padding:20px;text-decoration:none;color:var(--ink);display:block}
.champ:hover{border-color:var(--am)}
.ck{font-size:11px;letter-spacing:1.5px;text-transform:uppercase;color:var(--am)}
.cv{font:600 44px/1.1 var(--mono);color:var(--am);margin:6px 0}
.ct{font-size:17px;font-weight:600} .cs{color:var(--mu);font-size:13px;margin-top:4px}
.cd{color:#444;font-size:12.5px;line-height:1.55;margin-top:10px;border-top:1px solid var(--ln);padding-top:10px}
.stats{background:var(--sf);border:1px solid var(--ln);border-radius:12px;padding:14px 18px;display:flex;flex-direction:column;justify-content:space-between}
.st{display:flex;justify-content:space-between;align-items:baseline;padding:5px 0;border-bottom:1px solid var(--ln)}
.st:last-child{border:0} .st b{font:600 19px var(--mono)} .st span{color:var(--mu);font-size:12.5px}
.credit{text-align:center;margin-top:14px;font:600 14px var(--sans);color:#333;letter-spacing:.2px}
.credit b{color:var(--am);font-weight:800}
.rowg{display:grid;gap:14px;margin-bottom:14px}
.g23{grid-template-columns:2fr 1fr} .g11{grid-template-columns:1.2fr 1fr}
@media(max-width:820px){.top,.g23,.g11{grid-template-columns:1fr}}
svg{width:100%;height:auto;display:block}
.ax{font:11px var(--mono);fill:#555} .bn2{font:600 11px var(--mono);fill:#333}
.dn{font:700 30px var(--mono);fill:#111}
.donut{max-width:220px;margin:0 auto}
.lgs{display:flex;flex-wrap:wrap;gap:6px 14px;margin-top:12px}
.lg{font-size:11.5px;color:var(--mu)} .lg i{display:inline-block;width:9px;height:9px;border-radius:2px;margin-right:5px} .lg b{color:#111;font-family:var(--mono)}
.cbar{display:grid;grid-template-columns:130px 1fr 90px;gap:10px;align-items:center;padding:7px 0;border-bottom:1px solid var(--ln)}
.cbar:last-child{border:0} .cn{font-size:12.5px;color:#333;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
.ct2{background:var(--sf2);height:10px;border-radius:4px;overflow:hidden} .cf{display:block;height:100%;background:var(--ac);opacity:.85;border-radius:4px}
.cv2{font:600 12.5px var(--mono);text-align:right} .cv2 i{font-style:normal;color:var(--mu);font-size:10.5px;margin-left:6px}
h2.sec{font-size:13px;text-transform:uppercase;letter-spacing:1.4px;color:var(--mu);margin:26px 0 10px}
ol.heads{list-style:none;display:flex;flex-direction:column;gap:8px}
ol.heads li{display:flex;gap:14px;background:var(--sf);border:1px solid var(--ln);border-radius:10px;padding:13px 16px}
ol.heads .hr{font:700 20px var(--mono);color:var(--ac);min-width:26px}
ol.heads li.h2 .hr{color:#0369a1} ol.heads li.h3 .hr{color:#475569}
ol.heads li.h1{border-color:#bfe3dd}
ol.heads p{font-size:14.5px} ol.heads li.h1 p{font-size:16px;font-weight:600} ol.heads li.h3 p{color:var(--mu)}
.bars{display:flex;flex-direction:column;gap:4px}
.bar{display:grid;grid-template-columns:190px 1fr 30px;gap:12px;align-items:center;text-decoration:none;color:var(--ink);padding:4px 0}
.bar:hover .bn{color:var(--ac)} .bn{font-size:12.5px;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
.bt{background:var(--sf2);border-radius:4px;height:10px;overflow:hidden} .bf{display:block;height:100%;opacity:.9;border-radius:4px}
.bv{font:600 12.5px var(--mono);text-align:right;color:var(--mu)}
@media(max-width:640px){.bar{grid-template-columns:130px 1fr 28px}}
.theme{margin-top:30px} .theme h2{font-size:17px;display:flex;align-items:baseline;gap:10px}
.tn{font:700 15px var(--mono)} .dot{width:9px;height:9px;border-radius:3px;display:inline-block}
.cnt{font:400 12.5px var(--mono);color:var(--mu);margin-left:auto}
.insight{color:#444;font-size:13.5px;margin:6px 0 10px;padding-left:2px;max-width:860px}
.rows{border:1px solid var(--ln);border-radius:10px;overflow:hidden}
.row{display:grid;grid-template-columns:64px 56px 150px 1fr;gap:12px;align-items:baseline;padding:9px 14px;border-bottom:1px solid var(--ln);text-decoration:none;color:var(--ink);background:var(--sf)}
.row:last-child{border:0} .row:hover{background:var(--sf2)}
.vw{font:600 12.5px var(--mono);text-align:right}
.v1{color:var(--am)} .v2{color:var(--ac)} .v3{color:var(--mu)}
.dt{font:12px var(--mono);color:var(--mu)} .ch{font-size:12px;color:#2c5f8a;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
.tt{font-size:13.5px} .other .tt{color:var(--mu)}
footer{margin-top:34px;color:#555;font:12px/1.7 var(--mono);border-top:1px solid var(--ln);padding-top:12px}
&lt;/style&gt;&lt;/head&gt;&lt;body&gt;&lt;div class="wrap"&gt;

&lt;div class="datebar"&gt;&lt;span class="dtc"&gt;2026-10-25 &amp;rarr; 2026-11-01&lt;/span&gt;&lt;/div&gt;
&lt;div class="credit"&gt;Bu veriler &lt;b&gt;Murat Karakaya Akademi&lt;/b&gt; tarafından derlenmiştir.&lt;/div&gt;

&lt;div class="meta"&gt;&lt;b&gt;27&lt;/b&gt; aktif kanal / 43 &amp;nbsp;|&amp;nbsp; &lt;b&gt;135&lt;/b&gt; video &amp;nbsp;|&amp;nbsp; &lt;b&gt;3.1M&lt;/b&gt; toplam izlenme&lt;/div&gt;
&lt;h2 class="sec"&gt;Haftanın 3 manşeti&lt;/h2&gt;&lt;ol class="heads"&gt;&lt;li class="h1"&gt;&lt;span class="hr"&gt;1&lt;/span&gt;&lt;p&gt;Google, iki hafta arayla ikinci sınır modelini çıkardı: Gemini 4 Carbon zekâ endeksinde birinci sıraya oturdu, tanıtım videosu 421 bin görüntülenmeye ulaştı.&lt;/p&gt;&lt;/li&gt;&lt;li class="h2"&gt;&lt;span class="hr"&gt;2&lt;/span&gt;&lt;p&gt;DeepSeek yeniden sahne aldı: &amp;#x27;Deepseek herkesi ezdi&amp;#x27; ve &amp;#x27;yeni model tam bir canavar&amp;#x27; videoları haftanın en çok izlenenleri arasında 300 bini aşan bir dalga yarattı.&lt;/p&gt;&lt;/li&gt;&lt;li class="h3"&gt;&lt;span class="hr"&gt;3&lt;/span&gt;&lt;p&gt;Yerel yapay zekâ cephesi kaynadı: NVIDIA&amp;#x27;nın GPU&amp;#x27;ları local AI için gizlice kısıtladığı iddiası ve 3.000 token/sn yerel hız, donanım gündemini belirledi.&lt;/p&gt;&lt;/li&gt;&lt;/ol&gt;
&lt;h2 class="sec"&gt;Haftanın temaları&lt;/h2&gt;&lt;nav&gt;&lt;a href="#t1" style="font-size:12.8px"&gt;Gemini 4 Carbon dönemi&lt;b&gt;9&lt;/b&gt;&lt;/a&gt; &lt;a href="#t2" style="font-size:11.5px"&gt;Genie 4 ve dünya modelleri&lt;b&gt;4&lt;/b&gt;&lt;/a&gt; &lt;a href="#t3" style="font-size:11.8px"&gt;DeepSeek fırtınası&lt;b&gt;5&lt;/b&gt;&lt;/a&gt; &lt;a href="#t4" style="font-size:11.5px"&gt;GPT-6.1 ve OpenAI&amp;#x27;nin imkânsız iddiası&lt;b&gt;4&lt;/b&gt;&lt;/a&gt; &lt;a href="#t5" style="font-size:11.2px"&gt;Kimi ve kıyaslamalar&lt;b&gt;3&lt;/b&gt;&lt;/a&gt; &lt;a href="#t6" style="font-size:11.0px"&gt;Mistral Large 4 ve siber egemenlik&lt;b&gt;2&lt;/b&gt;&lt;/a&gt; &lt;a href="#t7" style="font-size:13.2px"&gt;OCR ve belge zekâsı&lt;b&gt;11&lt;/b&gt;&lt;/a&gt; &lt;a href="#t8" style="font-size:13.0px"&gt;Ses yapay zekâsı ve konuşma&lt;b&gt;10&lt;/b&gt;&lt;/a&gt; &lt;a href="#t9" style="font-size:15.2px"&gt;Küçük modeller ve verimlilik&lt;b&gt;19&lt;/b&gt;&lt;/a&gt; &lt;a href="#t10" style="font-size:11.2px"&gt;Yerel donanım ve hesaplama&lt;b&gt;3&lt;/b&gt;&lt;/a&gt; &lt;a href="#t11" style="font-size:13.5px"&gt;Kodlama ajanları ve geliştirici araçları&lt;b&gt;12&lt;/b&gt;&lt;/a&gt; &lt;a href="#t12" style="font-size:17.0px"&gt;Ajan orkestrasyonu ve LangChain ekosistemi&lt;b&gt;26&lt;/b&gt;&lt;/a&gt; &lt;a href="#t13" style="font-size:11.8px"&gt;Görsel ve video üretim ile multimodal&lt;b&gt;5&lt;/b&gt;&lt;/a&gt; &lt;a href="#t14" style="font-size:11.5px"&gt;OpenAI ürün ve kurumsal cephe&lt;b&gt;4&lt;/b&gt;&lt;/a&gt; &lt;a href="#t15" style="font-size:11.0px"&gt;Anthropic ve güç dengeleri&lt;b&gt;2&lt;/b&gt;&lt;/a&gt; &lt;a href="#t16" style="font-size:13.0px"&gt;İş tarafında pratik yapay zekâ&lt;b&gt;10&lt;/b&gt;&lt;/a&gt; &lt;a href="#t17" style="font-size:12.0px"&gt;Öğren ve geliştir&lt;b&gt;6&lt;/b&gt;&lt;/a&gt;&lt;/nav&gt;
&lt;div class="top"&gt;&lt;a class="champ" href="https://www.youtube.com/watch?v=kMpTBy3g1N0" target="_blank" rel="noopener"&gt;&lt;div class="ck"&gt;Haftanın rekoru&lt;/div&gt;&lt;div class="cv"&gt;421K&lt;/div&gt;&lt;div class="ct"&gt;Gemini 4 Carbon: The New Frontier&lt;/div&gt;&lt;div class="cs"&gt;Prompt Engineering &amp;middot; 28 Eki&lt;/div&gt;&lt;div class="cd"&gt;Google, iki hafta içinde tanıttığı ikinci sınır modeli Gemini 4 Carbon ile Artificial Analysis zekâ endeksinin 59 puanla bir numarası oldu; Video, modelin güvenlik açığı bulmadan sömürülebilirliği kanıtlamaya ve yamayı yazmaya kadar siber savunma iş akışının tamamında öne çıktığı iddiasını işliyor.&lt;/div&gt;&lt;/a&gt;
&lt;div class="stats"&gt;&lt;h3&gt;Haftanın rakamları&lt;/h3&gt;&lt;div class="st"&gt;&lt;span&gt;Aktif kanal&lt;/span&gt;&lt;b&gt;27/43&lt;/b&gt;&lt;/div&gt;
&lt;div class="st"&gt;&lt;span&gt;Toplam video&lt;/span&gt;&lt;b&gt;135&lt;/b&gt;&lt;/div&gt;
&lt;div class="st"&gt;&lt;span&gt;Toplam izlenme&lt;/span&gt;&lt;b&gt;3.1M&lt;/b&gt;&lt;/div&gt;
&lt;div class="st"&gt;&lt;span&gt;Tema sayısı&lt;/span&gt;&lt;b&gt;17&lt;/b&gt;&lt;/div&gt;&lt;/div&gt;&lt;/div&gt;
&lt;div class="rowg g23"&gt;
&lt;div class="card"&gt;&lt;h3&gt;Haftalık ritim — günlük video sayısı (çubuk · sol eksen) + izlenme (eğri · sağ eksen)&lt;/h3&gt;&lt;svg viewBox="0 0 960 250" role="img"&gt;&lt;line x1="44" y1="26" x2="904" y2="26" stroke="#d7dce2"/&gt;&lt;text x="36" y="30" text-anchor="end" class="ax"&gt;33&lt;/text&gt;&lt;line x1="44" y1="74" x2="904" y2="74" stroke="#d7dce2"/&gt;&lt;text x="36" y="78" text-anchor="end" class="ax"&gt;25&lt;/text&gt;&lt;line x1="44" y1="121" x2="904" y2="121" stroke="#d7dce2"/&gt;&lt;text x="36" y="125" text-anchor="end" class="ax"&gt;16&lt;/text&gt;&lt;line x1="44" y1="168" x2="904" y2="168" stroke="#d7dce2"/&gt;&lt;text x="36" y="172" text-anchor="end" class="ax"&gt;8&lt;/text&gt;&lt;line x1="44" y1="216" x2="904" y2="216" stroke="#d7dce2"/&gt;&lt;text x="36" y="220" text-anchor="end" class="ax"&gt;0&lt;/text&gt;&lt;rect x="63.3" y="193.0" width="68.8" height="23.0" rx="3" fill="#2b5f58"&gt;&lt;title&gt;4 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="97.8" y="188.0" text-anchor="middle" class="bn2"&gt;4&lt;/text&gt;&lt;text x="97.8" y="238" text-anchor="middle" class="ax"&gt;25 Eki&lt;/text&gt;&lt;rect x="170.8" y="106.6" width="68.8" height="109.4" rx="3" fill="#2b5f58"&gt;&lt;title&gt;19 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="205.2" y="101.6" text-anchor="middle" class="bn2"&gt;19&lt;/text&gt;&lt;text x="205.2" y="238" text-anchor="middle" class="ax"&gt;26 Eki&lt;/text&gt;&lt;rect x="278.4" y="89.3" width="68.8" height="126.7" rx="3" fill="#2b5f58"&gt;&lt;title&gt;22 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="312.8" y="84.3" text-anchor="middle" class="bn2"&gt;22&lt;/text&gt;&lt;text x="312.8" y="238" text-anchor="middle" class="ax"&gt;27 Eki&lt;/text&gt;&lt;rect x="385.9" y="26.0" width="68.8" height="190.0" rx="3" fill="#2b5f58"&gt;&lt;title&gt;33 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="420.2" y="21.0" text-anchor="middle" class="bn2"&gt;33&lt;/text&gt;&lt;text x="420.2" y="238" text-anchor="middle" class="ax"&gt;28 Eki&lt;/text&gt;&lt;rect x="493.4" y="77.8" width="68.8" height="138.2" rx="3" fill="#2b5f58"&gt;&lt;title&gt;24 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="527.8" y="72.8" text-anchor="middle" class="bn2"&gt;24&lt;/text&gt;&lt;text x="527.8" y="238" text-anchor="middle" class="ax"&gt;29 Eki&lt;/text&gt;&lt;rect x="600.9" y="112.4" width="68.8" height="103.6" rx="3" fill="#2b5f58"&gt;&lt;title&gt;18 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="635.2" y="107.4" text-anchor="middle" class="bn2"&gt;18&lt;/text&gt;&lt;text x="635.2" y="238" text-anchor="middle" class="ax"&gt;30 Eki&lt;/text&gt;&lt;rect x="708.4" y="141.2" width="68.8" height="74.8" rx="3" fill="#2b5f58"&gt;&lt;title&gt;13 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="742.8" y="136.2" text-anchor="middle" class="bn2"&gt;13&lt;/text&gt;&lt;text x="742.8" y="238" text-anchor="middle" class="ax"&gt;31 Eki&lt;/text&gt;&lt;rect x="815.9" y="204.5" width="68.8" height="11.5" rx="3" fill="#2b5f58"&gt;&lt;title&gt;2 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="850.2" y="199.5" text-anchor="middle" class="bn2"&gt;2&lt;/text&gt;&lt;text x="850.2" y="238" text-anchor="middle" class="ax"&gt;01 Kas&lt;/text&gt;&lt;polyline fill="none" stroke="#d97706" stroke-width="2" points="97.8,183.5 205.2,158.0 312.8,130.3 420.2,26.0 527.8,162.1 635.2,99.4 742.8,207.8 850.2,216.0"/&gt;&lt;circle cx="97.8" cy="183.5" r="3.5" fill="#d97706"&gt;&lt;title&gt;184K izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="205.2" cy="158.0" r="3.5" fill="#d97706"&gt;&lt;title&gt;328K izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="312.8" cy="130.3" r="3.5" fill="#d97706"&gt;&lt;title&gt;485K izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="420.2" cy="26.0" r="3.5" fill="#d97706"&gt;&lt;title&gt;1.1M izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="527.8" cy="162.1" r="3.5" fill="#d97706"&gt;&lt;title&gt;305K izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="635.2" cy="99.4" r="3.5" fill="#d97706"&gt;&lt;title&gt;659K izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="742.8" cy="207.8" r="3.5" fill="#d97706"&gt;&lt;title&gt;46K izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="850.2" cy="216.0" r="3.5" fill="#d97706"&gt;&lt;title&gt;84 izlenme&lt;/title&gt;&lt;/circle&gt;&lt;text x="912" y="220" class="ax" fill="#92400e"&gt;0&lt;/text&gt;&lt;text x="912" y="125" class="ax" fill="#92400e"&gt;537K&lt;/text&gt;&lt;text x="912" y="30" class="ax" fill="#92400e"&gt;1.1M&lt;/text&gt;&lt;/svg&gt;&lt;/div&gt;
&lt;div class="card"&gt;&lt;h3&gt;Tema dağılımı&lt;/h3&gt;&lt;svg viewBox="0 0 220 220" class="donut"&gt;&lt;path class="dseg" data-tip="Gemini 4 Carbon dönemi — 9 video (%7)" d="M110.0,30.0 A80,80 0 0 1 142.4,36.8" stroke="#0d9488" stroke-width="30" fill="none"&gt;&lt;title&gt;Gemini 4 Carbon dönemi: 9 (%7)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Genie 4 ve dünya modelleri — 4 video (%3)" d="M142.5,36.9 A80,80 0 0 1 155.4,44.2" stroke="#d97706" stroke-width="30" fill="none"&gt;&lt;title&gt;Genie 4 ve dünya modelleri: 4 (%3)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="DeepSeek fırtınası — 5 video (%4)" d="M155.5,44.2 A80,80 0 0 1 169.4,56.4" stroke="#dc2626" stroke-width="30" fill="none"&gt;&lt;title&gt;DeepSeek fırtınası: 5 (%4)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="GPT-6.1 ve OpenAI&amp;#x27;nin imkânsız iddiası — 4 video (%3)" d="M169.5,56.5 A80,80 0 0 1 178.3,68.3" stroke="#2563eb" stroke-width="30" fill="none"&gt;&lt;title&gt;GPT-6.1 ve OpenAI&amp;#x27;nin imkânsız iddiası: 4 (%3)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Kimi ve kıyaslamalar — 3 video (%2)" d="M178.3,68.4 A80,80 0 0 1 183.4,78.3" stroke="#4d7c0f" stroke-width="30" fill="none"&gt;&lt;title&gt;Kimi ve kıyaslamalar: 3 (%2)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Mistral Large 4 ve siber egemenlik — 2 video (%1)" d="M183.5,78.3 A80,80 0 0 1 186.1,85.2" stroke="#7c3aed" stroke-width="30" fill="none"&gt;&lt;title&gt;Mistral Large 4 ve siber egemenlik: 2 (%1)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="OCR ve belge zekâsı — 11 video (%8)" d="M186.1,85.3 A80,80 0 0 1 188.5,125.5" stroke="#0284c7" stroke-width="30" fill="none"&gt;&lt;title&gt;OCR ve belge zekâsı: 11 (%8)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Ses yapay zekâsı ve konuşma — 10 video (%7)" d="M188.4,125.7 A80,80 0 0 1 173.2,159.1" stroke="#92400e" stroke-width="30" fill="none"&gt;&lt;title&gt;Ses yapay zekâsı ve konuşma: 10 (%7)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Küçük modeller ve verimlilik — 19 video (%14)" d="M173.0,159.3 A80,80 0 0 1 112.2,190.0" stroke="#db2777" stroke-width="30" fill="none"&gt;&lt;title&gt;Küçük modeller ve verimlilik: 19 (%14)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Yerel donanım ve hesaplama — 3 video (%2)" d="M111.9,190.0 A80,80 0 0 1 100.8,189.5" stroke="#0f766e" stroke-width="30" fill="none"&gt;&lt;title&gt;Yerel donanım ve hesaplama: 3 (%2)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Kodlama ajanları ve geliştirici araçları — 12 video (%9)" d="M100.7,189.5 A80,80 0 0 1 60.2,172.6" stroke="#475569" stroke-width="30" fill="none"&gt;&lt;title&gt;Kodlama ajanları ve geliştirici araçları: 12 (%9)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Ajan orkestrasyonu ve LangChain ekosistemi — 26 video (%19)" d="M60.0,172.5 A80,80 0 0 1 33.8,85.7" stroke="#b45309" stroke-width="30" fill="none"&gt;&lt;title&gt;Ajan orkestrasyonu ve LangChain ekosistemi: 26 (%19)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Görsel ve video üretim ile multimodal — 5 video (%4)" d="M33.9,85.3 A80,80 0 0 1 41.6,68.5" stroke="#0d9488" stroke-width="30" fill="none"&gt;&lt;title&gt;Görsel ve video üretim ile multimodal: 5 (%4)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="OpenAI ürün ve kurumsal cephe — 4 video (%3)" d="M41.7,68.4 A80,80 0 0 1 50.5,56.5" stroke="#d97706" stroke-width="30" fill="none"&gt;&lt;title&gt;OpenAI ürün ve kurumsal cephe: 4 (%3)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Anthropic ve güç dengeleri — 2 video (%1)" d="M50.5,56.5 A80,80 0 0 1 55.8,51.2" stroke="#dc2626" stroke-width="30" fill="none"&gt;&lt;title&gt;Anthropic ve güç dengeleri: 2 (%1)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="İş tarafında pratik yapay zekâ — 10 video (%7)" d="M55.8,51.2 A80,80 0 0 1 87.8,33.2" stroke="#2563eb" stroke-width="30" fill="none"&gt;&lt;title&gt;İş tarafında pratik yapay zekâ: 10 (%7)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Öğren ve geliştir — 6 video (%4)" d="M87.9,33.1 A80,80 0 0 1 109.9,30.0" stroke="#4d7c0f" stroke-width="30" fill="none"&gt;&lt;title&gt;Öğren ve geliştir: 6 (%4)&lt;/title&gt;&lt;/path&gt;&lt;circle cx="110" cy="110" r="64" fill="#fff"/&gt;&lt;text x="110" y="106" text-anchor="middle" class="dn"&gt;135&lt;/text&gt;&lt;text x="110" y="126" text-anchor="middle" class="ax"&gt;video&lt;/text&gt;&lt;/svg&gt;&lt;div class="lgs"&gt;&lt;span class="lg"&gt;&lt;i style="background:#0d9488"&gt;&lt;/i&gt;Gemini 4 Carbon dönemi &lt;b&gt;9&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#d97706"&gt;&lt;/i&gt;Genie 4 ve dünya modelleri &lt;b&gt;4&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#dc2626"&gt;&lt;/i&gt;DeepSeek fırtınası &lt;b&gt;5&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#2563eb"&gt;&lt;/i&gt;GPT-6.1 ve OpenAI&amp;#x27;nin imkânsız iddiası &lt;b&gt;4&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#4d7c0f"&gt;&lt;/i&gt;Kimi ve kıyaslamalar &lt;b&gt;3&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#7c3aed"&gt;&lt;/i&gt;Mistral Large 4 ve siber egemenlik &lt;b&gt;2&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#0284c7"&gt;&lt;/i&gt;OCR ve belge zekâsı &lt;b&gt;11&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#92400e"&gt;&lt;/i&gt;Ses yapay zekâsı ve konuşma &lt;b&gt;10&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#db2777"&gt;&lt;/i&gt;Küçük modeller ve verimlilik &lt;b&gt;19&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#0f766e"&gt;&lt;/i&gt;Yerel donanım ve hesaplama &lt;b&gt;3&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#475569"&gt;&lt;/i&gt;Kodlama ajanları ve geliştirici araçları &lt;b&gt;12&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#b45309"&gt;&lt;/i&gt;Ajan orkestrasyonu ve LangChain ekosistemi &lt;b&gt;26&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#0d9488"&gt;&lt;/i&gt;Görsel ve video üretim ile multimodal &lt;b&gt;5&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#d97706"&gt;&lt;/i&gt;OpenAI ürün ve kurumsal cephe &lt;b&gt;4&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#dc2626"&gt;&lt;/i&gt;Anthropic ve güç dengeleri &lt;b&gt;2&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#2563eb"&gt;&lt;/i&gt;İş tarafında pratik yapay zekâ &lt;b&gt;10&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#4d7c0f"&gt;&lt;/i&gt;Öğren ve geliştir &lt;b&gt;6&lt;/b&gt;&lt;/span&gt;&lt;/div&gt;&lt;/div&gt;&lt;/div&gt;
&lt;div class="card" style="margin-bottom:14px"&gt;&lt;h3&gt;Trend zaman çizelgesi — her nokta bir video · x: yayın anı · y: log izlenme · renk: tema · üzerine gel: başlık&lt;/h3&gt;&lt;svg viewBox="0 0 960 300" role="img"&gt;&lt;line x1="46" y1="264.0" x2="946" y2="264.0" stroke="#d7dce2"/&gt;&lt;text x="38" y="268" text-anchor="end" class="ax"&gt;100&lt;/text&gt;&lt;line x1="46" y1="195.6" x2="946" y2="195.6" stroke="#d7dce2"/&gt;&lt;text x="38" y="200" text-anchor="end" class="ax"&gt;1K&lt;/text&gt;&lt;line x1="46" y1="127.2" x2="946" y2="127.2" stroke="#d7dce2"/&gt;&lt;text x="38" y="131" text-anchor="end" class="ax"&gt;10K&lt;/text&gt;&lt;line x1="46" y1="58.7" x2="946" y2="58.7" stroke="#d7dce2"/&gt;&lt;text x="38" y="63" text-anchor="end" class="ax"&gt;100K&lt;/text&gt;&lt;line x1="102.2" y1="16" x2="102.2" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="102.2" y="288" text-anchor="middle" class="ax"&gt;25 Eki&lt;/text&gt;&lt;line x1="214.8" y1="16" x2="214.8" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="214.8" y="288" text-anchor="middle" class="ax"&gt;26 Eki&lt;/text&gt;&lt;line x1="327.2" y1="16" x2="327.2" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="327.2" y="288" text-anchor="middle" class="ax"&gt;27 Eki&lt;/text&gt;&lt;line x1="439.8" y1="16" x2="439.8" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="439.8" y="288" text-anchor="middle" class="ax"&gt;28 Eki&lt;/text&gt;&lt;line x1="552.2" y1="16" x2="552.2" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="552.2" y="288" text-anchor="middle" class="ax"&gt;29 Eki&lt;/text&gt;&lt;line x1="664.8" y1="16" x2="664.8" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="664.8" y="288" text-anchor="middle" class="ax"&gt;30 Eki&lt;/text&gt;&lt;line x1="777.2" y1="16" x2="777.2" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="777.2" y="288" text-anchor="middle" class="ax"&gt;31 Eki&lt;/text&gt;&lt;line x1="889.8" y1="16" x2="889.8" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="889.8" y="288" text-anchor="middle" class="ax"&gt;01 Kas&lt;/text&gt;&lt;circle cx="613.2" cy="97.4" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;I Trained Qwen 4B with Distillation to Catch Qwen 27B - Prompt Engineering (27K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="524.1" cy="16.0" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Carbon: The New Frontier - Prompt Engineering (421K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="500.7" cy="100.2" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This New AI Architecture Is 31x Faster (NCM) - Prompt Engineering (25K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="322.6" cy="59.9" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Qwen Calling Home? - Prompt Engineering (96K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="275.7" cy="114.0" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Harness² Explained in 60 Seconds - Prompt Engineering (15K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="163.2" cy="69.5" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Kimi K3.1 Pro and Opus 5.5 Made the Same 13 Apps (Raw Results) - Prompt Engineering (69K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="917.9" cy="261.7" r="4.5" fill="#7c3aed" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Mistral Large 4 Is INSANE: 1 TRILLION Parameters, Open Weights, And My GPU Says NOPE - Prompt Engineer 48 (8)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="899.1" cy="247.2" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Youtu-Parsing-Omni: 1 AWESOME 5B Model, 5 Inputs, 1 JSON (I Broke It On a Chart) - Prompt Engineer 48 (76)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="880.4" cy="214.2" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;LFM3 on Ollama: 20 INSANE Real Tests (1 GB Model, 141 tok/s, 1 Epic FAIL) - Prompt Engineer 48 (434)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="856.9" cy="200.4" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;DeepTutor&amp;#x27;s CRAZY Open-Source AI Tutor Runs With NO API KEY - I Tested It - Prompt Engineer 48 (749)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="838.2" cy="228.4" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;INSANE: A Decision Model That KNOWS When It&amp;#x27;s Wrong (Lumma-Fev-v2 Tested) - Prompt Engineer 48 (231)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="814.8" cy="220.7" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;VibeGame: 8 AI Agents Build AND PLAY Their Own Game - CRAZY Claim, I Checked the Paper - Prompt Engineer 48 (330)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="791.3" cy="216.1" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How to Make AI Videos for Beginners: INSANE Results From 3 Prompts - Prompt Engineer 48 (401)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="735.1" cy="210.0" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This 2B Voice AI Runs LOCALLY at 158ms — CRAZY Fast, WILD Mistakes (llama.cpp Realtime Test) - Prompt Engineer 48 (515)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="646.0" cy="205.4" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This 278M Telugu Model Is CRAZY Good... I Tested It On My Laptop - Prompt Engineer 48 (619)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="585.1" cy="193.7" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This Tiny 1.3B AI Reads Documents INSANELY Well? I Tested It on 30 Pages (RTX 4060) - Prompt Engineer 48 (965)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="547.6" cy="163.7" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Liquid AI&amp;#x27;s LFM3 Is CRAZY: 1.7B Model vs 9B? I Tested It on the Live API - Prompt Engineer 48 (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="533.5" cy="209.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Google&amp;#x27;s Genie 4 Is INSANE: A World That Makes Its OWN SOUND (I Read Everything) - Prompt Engineer 48 (529)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="514.8" cy="222.1" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;LEANN: 201 GB vs 6 GB?! I Installed This INSANE Vector DB - Prompt Engineer 48 (309)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="496.0" cy="210.9" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;(70% OFF) I Tested 10 AI Models on ONE Coding Task — The Price Gap Is INSANE (One API, Real Costs) - Prompt Engineer 48 (498)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="477.2" cy="202.5" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This 10 MILLION Token AI Is CRAZY... But Is It Real? (Mirendil Fern) - Prompt Engineer 48 (693)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="725.7" cy="79.1" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;3,000 Tokens/Sec Locally: Testing the ZGX Fury - Sam Witteveen (50K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="275.7" cy="118.8" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;100% Open Image Model Lets You Place Text With JSON - Sam Witteveen (13K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="847.6" cy="91.4" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Avalon Just Beat Strata by 47% on AMD (9x Faster) - Kai (33K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="735.1" cy="117.4" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Bytes Is This The End of Token? - Kai (14K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="622.6" cy="70.3" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Every Way to Get 256 GB of VRAM for Local AI at Full Context - Kai (68K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="510.1" cy="80.6" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Your $200 ChatGPT Plan is now worth less! - Kai (48K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="397.6" cy="50.1" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Your GPU Is Nerfed for Local AI (NVIDIA&amp;#x27;s Hidden Limit) - Kai (133K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="285.1" cy="103.0" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Your PS5 Can Now Run Open Models (And It&amp;#x27;s Surprisingly Fast) - Kai (22K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="172.6" cy="75.7" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Is Flint the Killer of Qwen 3.8 27B? (First RSI Model) - Kai (56K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="406.9" cy="152.7" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;OpenCode Go + CodeGraph: Can Cheap Models Do the Job? - The Gray Cat (4K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="866.3" cy="199.5" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How To Grow On Instagram Step-By-Step - Howfinity (777)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="856.9" cy="192.4" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Turn a PDF into an Excel file with Claude - Howfinity (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="753.8" cy="195.9" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How to Build A Newsletter In 10 Minutes - Howfinity (890)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="636.6" cy="195.9" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;LinkedIn Sales Navigator Tutorial - Howfinity (889)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="533.5" cy="190.8" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;You&amp;#x27;re Prompting Claude Wrong #claude - Howfinity (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="519.4" cy="189.5" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How To Translate A PDF Into Different Languages With AI - Howfinity (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="416.3" cy="178.2" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How To Run Meta Ads For Beginners - Howfinity (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="308.5" cy="182.5" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How To Build AI Chatbot For Your Website - Howfinity (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="641.3" cy="171.1" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Two AI bots for spam calls and data privacy - Matthew Berman (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="552.2" cy="50.9" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Google &amp;quot;Carbon&amp;quot; just won everything... - Matthew Berman (130K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="542.9" cy="95.9" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Google announces Gemini 4 Carbon - Matthew Berman (29K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="453.8" cy="48.4" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Deepseek crushed everyone - Matthew Berman (141K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="411.6" cy="124.7" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;I Ranked 5 Ways to Wake Up - Matthew Berman (11K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="397.6" cy="125.1" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Anthropic’s IPO: Who Gets the Voting Power? - Matthew Berman (11K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="331.9" cy="55.5" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This is much more concerning than you think... - Matthew Berman (111K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="205.4" cy="112.2" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Kimi K3.1 Passes the Rubik&amp;#x27;s Cube Test - Matthew Berman (16K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="758.5" cy="58.1" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;There are many futures. Which do you want to build? - OpenAI (102K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="735.1" cy="168.2" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How Gen uses ChatGPT Work to turn data Into decisions | OpenAI - OpenAI (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="735.1" cy="119.9" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How Gen uses ChatGPT Work to turn data Into decisions | OpenAI - OpenAI (13K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="655.4" cy="159.3" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;ChatGPT Enterprise at Zendesk: Putting Company Data to Work - OpenAI (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="627.2" cy="116.3" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;ChatGPT Enterprise at Zendesk: Putting Company Data to Work - OpenAI (14K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="547.6" cy="77.0" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;API credits, now in ChatGPT Pro and Business plans - OpenAI (54K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="491.3" cy="134.8" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Merge CEO Shensi Ding: AI fluency starts at the top - OpenAI (8K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="491.3" cy="120.1" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Why Merge CEO Shensi Ding stays hands-on with Codex - OpenAI (13K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="481.9" cy="128.4" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;From legal bottleneck to business partner | Legora’s Emma Kadri Bergström - OpenAI (9K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="481.9" cy="127.1" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Building the future of legal work with AI | Legora CEO Max Junestrand - OpenAI (10K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="481.9" cy="134.9" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Faster contract review. Broader legal analysis. | Legora’s Alex Fortescue-Webb - OpenAI (8K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="481.9" cy="110.2" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How Legora is transforming legal work with OpenAI - OpenAI (18K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="397.6" cy="144.6" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Lawyers Stay in Control | AT&amp;amp;T + OpenAI - OpenAI (5K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="397.6" cy="108.9" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Getting a Head Start with Codex | Mitchell Hashimoto - OpenAI (18K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="397.6" cy="126.0" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;From Ideas to Prototypes with Codex | Shopify’s Mikhail Parakhin - OpenAI (10K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="814.8" cy="224.6" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Phonon-2 (164 MB) Is INSANE! 197× Real-Time Speech Recognition Tested - AI WITH Rithesh (277)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="796.0" cy="220.9" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Xiaomi-OCR-0 is HERE! 0.8B Open Weights OCR Model Tested! - AI WITH Rithesh (327)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="702.2" cy="210.2" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Surya OCR 2.5 (689M) IS HERE ! Tested Locally! INSANE Accuracy… But Some Surprises - AI WITH Rithesh (511)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="678.8" cy="198.9" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;LFM3-1.7B  Small Model for Agents - AI WITH Rithesh (793)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="585.1" cy="207.7" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Carbon is #1... but you can&amp;#x27;t use it - AI WITH Rithesh (564)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="575.7" cy="219.1" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Baidu&amp;#x27;s Unlimited OCR 2 is HERE - AI WITH Rithesh (353)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="575.7" cy="214.1" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Carbon Is HERE… No:1 Frontier Model  But You CAN’T Use It Yet! - AI WITH Rithesh (437)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="510.1" cy="172.0" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Liquid AI LFM3 Audio &amp;amp; Omni Tested! 2B Models That Listen, Talk &amp;amp; Use Tools - AI WITH Rithesh (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="496.0" cy="202.4" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;OpenBMB MiniCPM-OCR 1.0  and   MiniCPM-OCR Jev Is HERE!  1.3B OCR + 2B decision model  Tested - AI WITH Rithesh (696)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="477.2" cy="164.8" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Liquid AI LFM3-1.7B IS HERE : For LOCAL AGENTS 256K Context Tested Locally - AI WITH Rithesh (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="463.2" cy="165.3" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Liquid AI LFM3-VL-2B Small Vision  Language Model Is HERE! Punches Above Its Class? - AI WITH Rithesh (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="458.5" cy="209.5" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Google Genie 4 Is INSANE! Step Inside AI-Generated Worlds in Real Time - AI WITH Rithesh (525)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="378.8" cy="139.3" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Baidu Unlimited OCR 2 Tested Locally! 3B Model Handles 100+ Pages - AI WITH Rithesh (7K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="350.7" cy="194.4" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;GPT-6.1 TERRA: REAL USERS, REAL TESTS! - AI WITH Rithesh (942)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="261.6" cy="142.8" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Nanbeige 4.3 33B Claims to Beat Qwen 3.8 27B! Tested Locally on 24GB RAM - AI WITH Rithesh (6K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="627.2" cy="121.5" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;ComfyUI Prompts Explained: CLIP Text Encode, Negative and Dynamic Prompts - pixaroma (12K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="402.2" cy="103.9" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;MOSS Image in ComfyUI | Text to Image, Words &amp;amp; Layout (Ep40) - pixaroma (22K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="303.8" cy="106.2" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;MOSS Image - With Control, References &amp;amp; Editing in ComfyUI! - Nerdy Rodent (20K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="730.4" cy="142.7" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Give Me 42 Minutes, I&amp;#x27;ll Make LLM Post-Training Click Forever - Zachary Huang (6K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="392.9" cy="108.1" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Give Me 1 Hour, I Will Make Artificial Intelligence Click Forever - Zachary Huang (19K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="852.2" cy="132.4" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Anthropic&amp;#x27;s New Study: You&amp;#x27;re Not Controlling AI as Much as You Think You Are - Cole Medin (8K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="552.2" cy="116.9" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;The 4 Components of Every Successful AI Second Brain (Build These Now) - Cole Medin (14K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="327.2" cy="166.0" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Copilot Automations Run While You Sleep #ai #coding #githubcopilot - Burke Holland (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="622.6" cy="136.5" r="4.5" fill="#7c3aed" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Science Behind the Open Cyber Frontier with Mistral Large 4 - Hugging Face (7K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="280.4" cy="93.9" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Workshop: Training Agents at the frontier of AI Industry - Hugging Face (31K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="275.7" cy="166.1" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Hugging Face Journal Club: Score Centering Stabilizes Off-policy Reinforcement Learning - Hugging Face (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="735.1" cy="16.6" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;OpenAI Claims It Just Did The Impossible - Two Minute Papers (413K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="519.4" cy="42.2" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;DeepSeek&amp;#x27;s New AI Is An Absolute Monster - Two Minute Papers (175K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="875.7" cy="253.4" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;What if the best AI tutor for children was designed to talk less? - Agentic Buildcamp: Adaptive Entrepreneurship (43)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="763.2" cy="242.1" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Mom Test Bot: Build What Customers Actually Need, Not What They Say They Want - Agentic Buildcamp: Adaptive Entrepreneurship (109)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="650.7" cy="238.0" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;From Research Project to AI Startup: The Part Nobody Warns You About - Agentic Buildcamp: Adaptive Entrepreneurship (140)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="435.1" cy="238.4" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;$100 in Ads. Zero Sales. One Painful Lesson. - Agentic Buildcamp: Adaptive Entrepreneurship (137)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="425.7" cy="241.5" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;No Hallucination: AI Research Companion for Students - Agentic Buildcamp: Adaptive Entrepreneurship (113)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="313.2" cy="237.4" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;I Built an AI Personal Concierge for People Over 65 - Agentic Buildcamp: Adaptive Entrepreneurship (145)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="730.4" cy="137.7" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Why Own Your Intelligence? Harrison Chase Keynote | Interrupt London - LangChain (7K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="716.3" cy="176.1" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;The 50/50 problem with voice AI nobody is talking about #Shorts - LangChain (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="641.3" cy="172.0" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;The Hardest Part of Testing Agents: Write Access - LangChain (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="631.9" cy="162.2" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Building Semiconductor Agents at Skyworks with LangGraph + LangSmith - LangChain (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="533.5" cy="161.4" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;The Real Cost of &amp;quot;We Can Just Build It Ourselves&amp;quot; - LangChain (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="500.7" cy="122.7" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;What is LangChain? - LangChain (12K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="402.2" cy="73.6" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How to Build Agents That Improve Over Time: Model, Harness, Context | Interrupt London - LangChain (61K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="346.0" cy="163.0" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;We wired Devin into our Slack to triage messages - LangChain (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="294.4" cy="125.3" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;LangGraph explained in 1 minute - LangChain (11K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="275.7" cy="188.3" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Tool Problem or Model Problem? How Clay Tells the Difference - LangChain (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="275.7" cy="151.3" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Inside Managed Deep Agents: mda CLI, Context Hub, and Harbor Evals | Interrupt London - LangChain (4K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="594.4" cy="189.0" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;02 - Multimodal Spatial Biology: Registration - DigitalSreeni (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="369.4" cy="176.5" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;01 - Multimodal Spatial Biology: Introducing the Dataset - DigitalSreeni (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="814.8" cy="234.9" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Dialt On-Prem: Live Voice AI on One GPU #Shorts - Trelis Research (166)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="702.2" cy="225.3" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How AI Learned to Hill-Climb Intelligence #Shorts - Trelis Research (268)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="702.2" cy="216.2" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Dialt On-Prem: Custom Voice Assistants on Your Own GPU - Trelis Research (399)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="622.6" cy="204.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How AI Learned to Hill-Climb Intelligence - Trelis Research (643)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="369.4" cy="206.3" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Voice AI That Is Distinctly Not Human #Shorts - Trelis Research (596)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="256.9" cy="214.1" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Voice AI That Is Distinctly Not Human - Trelis Research (437)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="510.1" cy="152.7" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Join Andrew Ng at AI Dev NYC for two days of technical sessions and workshops. - DeepLearningAI (4K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="491.3" cy="145.7" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Agents on longer tasks need more than one loop - DeepLearningAI (5K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="317.9" cy="173.5" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;7 AI code review best practices #deeplearningai #qodo - DeepLearningAI (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="285.1" cy="180.7" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Learn from OpenAI, Hugging Face, Andrew Ng and Yann LeCun | AI Dev NYC #aidev #shorts - DeepLearningAI (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="397.6" cy="154.5" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Unlimited OCR 2 is here! - 1littlecoder (4K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="158.5" cy="84.9" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This question inspired LoRA fine-tuning - Julia Turc (41K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="772.6" cy="195.3" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;NVIDIA VP: Same AI Model, 30% to 100% (Must Know) - Mervin Praison (909)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="617.9" cy="185.0" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Nemotron 3.5 Transcribe BEATS Whisper? (I Tested It) - Mervin Praison (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="556.9" cy="176.8" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Carbon: Google&amp;#x27;s Best AI and You CAN&amp;#x27;T Use It - Mervin Praison (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="449.1" cy="172.6" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;DeepSeek in Claude Code (and the Engine They Gave Away) - Mervin Praison (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="317.9" cy="192.4" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;GPT-6.1 Terra: half the price, fewer mistakes - Mervin Praison (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="313.2" cy="199.8" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Why your AI agent stops following your rules - Mervin Praison (768)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="744.4" cy="81.4" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Full-Stack AI Engineering Course (FastAPI, LLMs, Docker, Supabase) - Dave Ebbelaar (46K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="308.5" cy="59.1" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This Free Repo Lets Claude Code Build Any Mac App - Dave Ebbelaar (99K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="646.0" cy="124.2" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Accelerate shipping to scale with Google Antigravity - Google for Developers (11K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="556.9" cy="173.1" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;&#129716; Building an autonomous plant rover with Gemini Live - Google for Developers (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="538.2" cy="87.9" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Building real-time vision &amp;amp; edge navigation with Gemini Live - Google for Developers (37K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="510.1" cy="101.2" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Functional smart devices using Gemini 4 Argon and Google Antigravity - Google for Developers (24K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="402.2" cy="74.9" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Introducing Genie 4 - Google for Developers (58K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="496.0" cy="115.2" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Support Vector Machines (SVM): Concepts and Code - StatQuest with Josh Starmer (15K)&lt;/title&gt;&lt;/circle&gt;&lt;/svg&gt;&lt;/div&gt;
&lt;div class="rowg g11"&gt;
&lt;div class="card"&gt;&lt;h3&gt;Tema yoğunluğu&lt;/h3&gt;&lt;div class="bars"&gt;&lt;a class="bar" href="#t1"&gt;&lt;span class="bn"&gt;Gemini 4 Carbon dönemi&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:35%;background:#0d9488"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;9&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t2"&gt;&lt;span class="bn"&gt;Genie 4 ve dünya modelleri&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:15%;background:#d97706"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;4&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t3"&gt;&lt;span class="bn"&gt;DeepSeek fırtınası&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:19%;background:#dc2626"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;5&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t4"&gt;&lt;span class="bn"&gt;GPT-6.1 ve OpenAI&amp;#x27;nin imkânsız iddiası&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:15%;background:#2563eb"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;4&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t5"&gt;&lt;span class="bn"&gt;Kimi ve kıyaslamalar&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:12%;background:#4d7c0f"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;3&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t6"&gt;&lt;span class="bn"&gt;Mistral Large 4 ve siber egemenlik&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:8%;background:#7c3aed"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;2&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t7"&gt;&lt;span class="bn"&gt;OCR ve belge zekâsı&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:42%;background:#0284c7"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;11&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t8"&gt;&lt;span class="bn"&gt;Ses yapay zekâsı ve konuşma&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:38%;background:#92400e"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;10&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t9"&gt;&lt;span class="bn"&gt;Küçük modeller ve verimlilik&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:73%;background:#db2777"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;19&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t10"&gt;&lt;span class="bn"&gt;Yerel donanım ve hesaplama&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:12%;background:#0f766e"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;3&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t11"&gt;&lt;span class="bn"&gt;Kodlama ajanları ve geliştirici araçları&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:46%;background:#475569"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;12&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t12"&gt;&lt;span class="bn"&gt;Ajan orkestrasyonu ve LangChain ekosistemi&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:100%;background:#b45309"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;26&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t13"&gt;&lt;span class="bn"&gt;Görsel ve video üretim ile multimodal&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:19%;background:#0d9488"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;5&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t14"&gt;&lt;span class="bn"&gt;OpenAI ürün ve kurumsal cephe&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:15%;background:#d97706"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;4&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t15"&gt;&lt;span class="bn"&gt;Anthropic ve güç dengeleri&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:8%;background:#dc2626"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;2&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t16"&gt;&lt;span class="bn"&gt;İş tarafında pratik yapay zekâ&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:38%;background:#2563eb"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;10&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t17"&gt;&lt;span class="bn"&gt;Öğren ve geliştir&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:23%;background:#4d7c0f"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;6&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/div&gt;
&lt;div class="card"&gt;&lt;h3&gt;En çok izlenen kanallar&lt;/h3&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:100%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;654K&lt;i&gt;6 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Two Minute Papers&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:90%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;587K&lt;i&gt;2 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Matthew Berman&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:69%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;451K&lt;i&gt;8 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Kai&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:57%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;375K&lt;i&gt;7 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;OpenAI&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:44%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;288K&lt;i&gt;15 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Dave Ebbelaar&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:22%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;145K&lt;i&gt;2 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;/div&gt;&lt;/div&gt;
&lt;section class="theme" id="t1"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#0d9488"&gt;01&lt;/span&gt;&lt;i class="dot" style="background:#0d9488"&gt;&lt;/i&gt;Gemini 4 Carbon dönemi&lt;span class="cnt"&gt;9 video &amp;middot; 609K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Google iki hafta içinde ikinci sınır modelini (Argon&amp;#x27;dan sonra Carbon) duyurdu ve siber savunma kıyaslamalarında rakiplerin önüne geçtiğini savunuyor. Kanal manşetleri ortak noktada buluştu: model bir numara ama henüz herkese açık değil.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=kMpTBy3g1N0" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;421K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Carbon: The New Frontier&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=zhnJo7s91ls" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;130K&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Google &amp;quot;Carbon&amp;quot; just won everything...&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=3yZb9rGt-rk" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;29K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Google announces Gemini 4 Carbon&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=pQ_Dnr9Q4k8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;24K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;Functional smart devices using Gemini 4 Argon and Google Antigravity&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=iN5fscBeTDE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;Mervin Praison&lt;/span&gt;&lt;span class="tt"&gt;DeepSeek in Claude Code (and the Engine They Gave Away)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=DXfXXB8Murs" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;Mervin Praison&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Carbon: Google&amp;#x27;s Best AI and You CAN&amp;#x27;T Use It&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=G9Q4UlkyEy0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;564&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Carbon is #1... but you can&amp;#x27;t use it&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=LfYPH46SZfw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;525&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Google Genie 4 Is INSANE! Step Inside AI-Generated Worlds in Real Time&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=mcZ2rtmHa98" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;437&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Carbon Is HERE… No:1 Frontier Model  But You CAN’T Use It Yet!&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t2"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#d97706"&gt;02&lt;/span&gt;&lt;i class="dot" style="background:#d97706"&gt;&lt;/i&gt;Genie 4 ve dünya modelleri&lt;span class="cnt"&gt;4 video &amp;middot; 59K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Google&amp;#x27;ın dünya modeli Genie 4, gerçek zamanlı gezilebilir ve kendi sesini üreten simülasyonlarla haftanın en çok konuşulan demosu oldu; &amp;#x27;zekânın tepe tırmanışı&amp;#x27; tartışması da aynı kümede büyüdü.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=LOP8ul3KqQI" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;58K&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;Introducing Genie 4&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=1mHYqC4QZAI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;643&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;How AI Learned to Hill-Climb Intelligence&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=pRB5KUhRr2U" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;529&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Google&amp;#x27;s Genie 4 Is INSANE: A World That Makes Its OWN SOUND (I Read Everything)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=BWbREb3VRyA" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;268&lt;/span&gt;&lt;span class="dt"&gt;30 Eki&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;How AI Learned to Hill-Climb Intelligence #Shorts&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t3"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#dc2626"&gt;03&lt;/span&gt;&lt;i class="dot" style="background:#dc2626"&gt;&lt;/i&gt;DeepSeek fırtınası&lt;span class="cnt"&gt;5 video &amp;middot; 388K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;DeepSeek v4.1 Flash dalgası &amp;#x27;herkesi ezdi&amp;#x27; başlıklarıyla haftanın en yüksek izlenme kitlesini topladı; modelin Claude Code&amp;#x27;a bağlanması ve açık bırakılan motoru ayrı bir gündem yarattı.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=pryCaGv8wuo" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;175K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;Two Minute Papers&lt;/span&gt;&lt;span class="tt"&gt;DeepSeek&amp;#x27;s New AI Is An Absolute Monster&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=DDZ-a6bAn3Y" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;141K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Deepseek crushed everyone&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=JPHT9AvoPSA" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;68K&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Every Way to Get 256 GB of VRAM for Local AI at Full Context&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=blS3l32vBR0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;4K&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;The Gray Cat&lt;/span&gt;&lt;span class="tt"&gt;OpenCode Go + CodeGraph: Can Cheap Models Do the Job?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=uWM3ZnWYwc0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;498&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;(70% OFF) I Tested 10 AI Models on ONE Coding Task — The Price Gap Is INSANE (One API, Real Costs)&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t4"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#2563eb"&gt;04&lt;/span&gt;&lt;i class="dot" style="background:#2563eb"&gt;&lt;/i&gt;GPT-6.1 ve OpenAI&amp;#x27;nin imkânsız iddiası&lt;span class="cnt"&gt;4 video &amp;middot; 462K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;OpenAI&amp;#x27;nin &amp;#x27;imkânsızı başardık&amp;#x27; dediği karmaşıklık kuramı makalesi Two Minute Papers&amp;#x27;da 412 bin görüntülenme gördü; GPT-6.1 Terra ise yarı fiyatı ve daha az hatasıyla gerçek kullanıcılarda test edildi.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=CWdTBLsbaQs" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;413K&lt;/span&gt;&lt;span class="dt"&gt;30 Eki&lt;/span&gt;&lt;span class="ch"&gt;Two Minute Papers&lt;/span&gt;&lt;span class="tt"&gt;OpenAI Claims It Just Did The Impossible&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=P0i3vYuJZH8" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;48K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Your $200 ChatGPT Plan is now worth less!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=azRot3GBEh4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;26 Eki&lt;/span&gt;&lt;span class="ch"&gt;Mervin Praison&lt;/span&gt;&lt;span class="tt"&gt;GPT-6.1 Terra: half the price, fewer mistakes&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=6oUJ2pLgsXc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;942&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;GPT-6.1 TERRA: REAL USERS, REAL TESTS!&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t5"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#4d7c0f"&gt;05&lt;/span&gt;&lt;i class="dot" style="background:#4d7c0f"&gt;&lt;/i&gt;Kimi ve kıyaslamalar&lt;span class="cnt"&gt;3 video &amp;middot; 127K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Moonshot&amp;#x27;un Kimi K3.1 modeli Rubik küp testini geçerken, Kimi K3.1 Pro ile Opus 5.5&amp;#x27;in aynı 13 uygulamayı nasıl inşa ettiğinin ham sonuçları haftanın en çok paylaşılan çıplak kıyaslaması oldu.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=_XXX5K-EMoc" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;69K&lt;/span&gt;&lt;span class="dt"&gt;25 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;Kimi K3.1 Pro and Opus 5.5 Made the Same 13 Apps (Raw Results)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ug9el1OGGEw" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;41K&lt;/span&gt;&lt;span class="dt"&gt;25 Eki&lt;/span&gt;&lt;span class="ch"&gt;Julia Turc&lt;/span&gt;&lt;span class="tt"&gt;This question inspired LoRA fine-tuning&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=L18lkQs4PeM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;16K&lt;/span&gt;&lt;span class="dt"&gt;25 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Kimi K3.1 Passes the Rubik&amp;#x27;s Cube Test&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t6"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#7c3aed"&gt;06&lt;/span&gt;&lt;i class="dot" style="background:#7c3aed"&gt;&lt;/i&gt;Mistral Large 4 ve siber egemenlik&lt;span class="cnt"&gt;2 video &amp;middot; 7K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;1 trilyon parametreli ve açık ağırlıklı Mistral Large 4, Avrupa&amp;#x27;nın &amp;#x27;kendi zekâmıza sahip olma&amp;#x27; argümanının merkezine yerleşti; ilk durak açık siber güvenlik oldu.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=FczbSauVvjw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;7K&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;Science Behind the Open Cyber Frontier with Mistral Large 4&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=SGek-YnySDE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;8&lt;/span&gt;&lt;span class="dt"&gt;01 Kas&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Mistral Large 4 Is INSANE: 1 TRILLION Parameters, Open Weights, And My GPU Says NOPE&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t7"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#0284c7"&gt;07&lt;/span&gt;&lt;i class="dot" style="background:#0284c7"&gt;&lt;/i&gt;OCR ve belge zekâsı&lt;span class="cnt"&gt;11 video &amp;middot; 66K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Baidu Unlimited OCR 2, Surya 2.5, Xiaomi-OCR-0 ve MiniCPM-OCR ile 1-3 milyar parametrelik açık OCR modelleri bir hafta içinde üst üste geldi; belge işleme, küçük model yarışının en sıcak alanı.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=lEFFa0Z5BIE" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;50K&lt;/span&gt;&lt;span class="dt"&gt;30 Eki&lt;/span&gt;&lt;span class="ch"&gt;Sam Witteveen&lt;/span&gt;&lt;span class="tt"&gt;3,000 Tokens/Sec Locally: Testing the ZGX Fury&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=SEY5sRYlG3s" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;7K&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Baidu Unlimited OCR 2 Tested Locally! 3B Model Handles 100+ Pages&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=OneYJJWDnNA" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;4K&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;1littlecoder&lt;/span&gt;&lt;span class="tt"&gt;Unlimited OCR 2 is here!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=NxN4Yf90ccU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;How To Translate A PDF Into Different Languages With AI&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=D7xswbjd9YM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;31 Eki&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;Turn a PDF into an Excel file with Claude&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=YIEcAKlIp98" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;965&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;This Tiny 1.3B AI Reads Documents INSANELY Well? I Tested It on 30 Pages (RTX 4060)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=MgFW_RkQfzY" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;696&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;OpenBMB MiniCPM-OCR 1.0  and   MiniCPM-OCR Jev Is HERE!  1.3B OCR + 2B decision model  Tested&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=BiLoeGWTLpU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;511&lt;/span&gt;&lt;span class="dt"&gt;30 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Surya OCR 2.5 (689M) IS HERE ! Tested Locally! INSANE Accuracy… But Some Surprises&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=viBr_-DAxgk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;353&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Baidu&amp;#x27;s Unlimited OCR 2 is HERE&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=JFkAbZqzd3Q" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;327&lt;/span&gt;&lt;span class="dt"&gt;31 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Xiaomi-OCR-0 is HERE! 0.8B Open Weights OCR Model Tested!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=-LmJsjTy3XU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;76&lt;/span&gt;&lt;span class="dt"&gt;01 Kas&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Youtu-Parsing-Omni: 1 AWESOME 5B Model, 5 Inputs, 1 JSON (I Broke It On a Chart)&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t8"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#92400e"&gt;08&lt;/span&gt;&lt;i class="dot" style="background:#92400e"&gt;&lt;/i&gt;Ses yapay zekâsı ve konuşma&lt;span class="cnt"&gt;10 video &amp;middot; 8K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;158 ms gecikmeyle çalışan 2B&amp;#x27;lik yerel ses modeli, 164 MB&amp;#x27;lık Phonon-2 ve Nemotron-Whisper karşılaştırması ses hattını hızlandırdı; &amp;#x27;insana benzemeyen ses&amp;#x27; ve çocuk eğitimi gibi insan merkezli tartışmalar da büyüdü.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=fVQVaM95mAo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Liquid AI LFM3 Audio &amp;amp; Omni Tested! 2B Models That Listen, Talk &amp;amp; Use Tools&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=f03D4dkwAdo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;30 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;The 50/50 problem with voice AI nobody is talking about #Shorts&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=scOCcD0XNB0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;Mervin Praison&lt;/span&gt;&lt;span class="tt"&gt;Nemotron 3.5 Transcribe BEATS Whisper? (I Tested It)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Pj4umL7S03M" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;596&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Voice AI That Is Distinctly Not Human #Shorts&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=0rGb20U0i2U" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;515&lt;/span&gt;&lt;span class="dt"&gt;30 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;This 2B Voice AI Runs LOCALLY at 158ms — CRAZY Fast, WILD Mistakes (llama.cpp Realtime Test)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=gg6waFFnwDs" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;437&lt;/span&gt;&lt;span class="dt"&gt;26 Eki&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Voice AI That Is Distinctly Not Human&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=1ydb3Fyh0fE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;399&lt;/span&gt;&lt;span class="dt"&gt;30 Eki&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Dialt On-Prem: Custom Voice Assistants on Your Own GPU&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=iT1qv7aC_lg" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;277&lt;/span&gt;&lt;span class="dt"&gt;31 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Phonon-2 (164 MB) Is INSANE! 197× Real-Time Speech Recognition Tested&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=_ddgeoQmdWo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;166&lt;/span&gt;&lt;span class="dt"&gt;31 Eki&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Dialt On-Prem: Live Voice AI on One GPU #Shorts&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=YX7Htw9tprs" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;43&lt;/span&gt;&lt;span class="dt"&gt;31 Eki&lt;/span&gt;&lt;span class="ch"&gt;Agentic Buildcamp: Adaptive Entrepreneurship&lt;/span&gt;&lt;span class="tt"&gt;What if the best AI tutor for children was designed to talk less?&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t9"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#db2777"&gt;09&lt;/span&gt;&lt;i class="dot" style="background:#db2777"&gt;&lt;/i&gt;Küçük modeller ve verimlilik&lt;span class="cnt"&gt;19 video &amp;middot; 305K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Liquid AI&amp;#x27;nın LFM3 ailesi, damıtılmış Qwen 4B denemeleri ve &amp;#x27;token&amp;#x27; kavramının sonunu soran videolar; haftanın ana mesajının &amp;#x27;daha büyük değil, daha verimli&amp;#x27; olduğunu gösterdi.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Pn8e474Z5Ps" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;96K&lt;/span&gt;&lt;span class="dt"&gt;26 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;Qwen Calling Home?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=RmE4G6MhOVU" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;56K&lt;/span&gt;&lt;span class="dt"&gt;25 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Is Flint the Killer of Qwen 3.8 27B? (First RSI Model)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=hOitZIkMT2Y" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;27K&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;I Trained Qwen 4B with Distillation to Catch Qwen 27B&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=_87vrjxx1EE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;25K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;This New AI Architecture Is 31x Faster (NCM)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=TgmC_VZEKTE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;22K&lt;/span&gt;&lt;span class="dt"&gt;26 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Your PS5 Can Now Run Open Models (And It&amp;#x27;s Surprisingly Fast)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=l4R96NemQS8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;22K&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;pixaroma&lt;/span&gt;&lt;span class="tt"&gt;MOSS Image in ComfyUI | Text to Image, Words &amp;amp; Layout (Ep40)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=hTCjWfbyCjU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;20K&lt;/span&gt;&lt;span class="dt"&gt;26 Eki&lt;/span&gt;&lt;span class="ch"&gt;Nerdy Rodent&lt;/span&gt;&lt;span class="tt"&gt;MOSS Image - With Control, References &amp;amp; Editing in ComfyUI!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=PfR4S3Zzdoo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;14K&lt;/span&gt;&lt;span class="dt"&gt;30 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Bytes Is This The End of Token?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=5l3l9aEwHlo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;6K&lt;/span&gt;&lt;span class="dt"&gt;30 Eki&lt;/span&gt;&lt;span class="ch"&gt;Zachary Huang&lt;/span&gt;&lt;span class="tt"&gt;Give Me 42 Minutes, I&amp;#x27;ll Make LLM Post-Training Click Forever&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=QptAOS4GIZc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;6K&lt;/span&gt;&lt;span class="dt"&gt;26 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Nanbeige 4.3 33B Claims to Beat Qwen 3.8 27B! Tested Locally on 24GB RAM&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=2Ybdvqt2J4Y" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Liquid AI&amp;#x27;s LFM3 Is CRAZY: 1.7B Model vs 9B? I Tested It on the Live API&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Tyklj_EWX_g" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Liquid AI LFM3-1.7B IS HERE : For LOCAL AGENTS 256K Context Tested Locally&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=8HR_DBj_X2U" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Liquid AI LFM3-VL-2B Small Vision  Language Model Is HERE! Punches Above Its Class?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=DdLFjwbkv_g" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;793&lt;/span&gt;&lt;span class="dt"&gt;30 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;LFM3-1.7B  Small Model for Agents&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=yz4xSmm1D4M" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;693&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;This 10 MILLION Token AI Is CRAZY... But Is It Real? (Mirendil Fern)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=4rIUoadQOjk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;619&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;This 278M Telugu Model Is CRAZY Good... I Tested It On My Laptop&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=FcxT97Dj6QQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;434&lt;/span&gt;&lt;span class="dt"&gt;31 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;LFM3 on Ollama: 20 INSANE Real Tests (1 GB Model, 141 tok/s, 1 Epic FAIL)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ffEi-J4erIw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;309&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;LEANN: 201 GB vs 6 GB?! I Installed This INSANE Vector DB&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Hk23J-hAF6g" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;231&lt;/span&gt;&lt;span class="dt"&gt;31 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;INSANE: A Decision Model That KNOWS When It&amp;#x27;s Wrong (Lumma-Fev-v2 Tested)&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t10"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#0f766e"&gt;10&lt;/span&gt;&lt;i class="dot" style="background:#0f766e"&gt;&lt;/i&gt;Yerel donanım ve hesaplama&lt;span class="cnt"&gt;3 video &amp;middot; 168K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;NVIDIA&amp;#x27;nın local AI&amp;#x27;i gizlice kısıtladığı iddiası, 256 GB VRAM yolları rehberi ve 3.000 token/sn üreten ZGX Fury, yerel çalıştırma donanımının haftanın en tartışmalı başlıkları haline geldiğini gösterdi.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=VIm73gaY_hM" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;133K&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Your GPU Is Nerfed for Local AI (NVIDIA&amp;#x27;s Hidden Limit)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=0F63pTS8Gt0" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;33K&lt;/span&gt;&lt;span class="dt"&gt;31 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Avalon Just Beat Strata by 47% on AMD (9x Faster)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=oAhYP1FwKSY" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;909&lt;/span&gt;&lt;span class="dt"&gt;30 Eki&lt;/span&gt;&lt;span class="ch"&gt;Mervin Praison&lt;/span&gt;&lt;span class="tt"&gt;NVIDIA VP: Same AI Model, 30% to 100% (Must Know)&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t11"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#475569"&gt;11&lt;/span&gt;&lt;i class="dot" style="background:#475569"&gt;&lt;/i&gt;Kodlama ajanları ve geliştirici araçları&lt;span class="cnt"&gt;12 video &amp;middot; 236K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Claude Code ile Mac uygulaması üreten ücretsiz depo, Codex&amp;#x27;i iş akışına gömen kurucular ve GitHub Copilot otomasyonları; kodlama ajanlarının &amp;#x27;deneme&amp;#x27; aşamasından &amp;#x27;günlük iş&amp;#x27; aşamasına geçtiğini gösteriyor.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=gN6wl-OX8-Q" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;99K&lt;/span&gt;&lt;span class="dt"&gt;26 Eki&lt;/span&gt;&lt;span class="ch"&gt;Dave Ebbelaar&lt;/span&gt;&lt;span class="tt"&gt;This Free Repo Lets Claude Code Build Any Mac App&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=XzjSx7u9mGw" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;46K&lt;/span&gt;&lt;span class="dt"&gt;30 Eki&lt;/span&gt;&lt;span class="ch"&gt;Dave Ebbelaar&lt;/span&gt;&lt;span class="tt"&gt;Full-Stack AI Engineering Course (FastAPI, LLMs, Docker, Supabase)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=lPZQrVFcz3c" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;18K&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Getting a Head Start with Codex | Mitchell Hashimoto&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=yJxZef1Ylok" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;18K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;How Legora is transforming legal work with OpenAI&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=6ep8Vvm4Bdw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;13K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Why Merge CEO Shensi Ding stays hands-on with Codex&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=rsOiZE19RVo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;11K&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;Accelerate shipping to scale with Google Antigravity&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=cflbzHRvGvg" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;10K&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;From Ideas to Prototypes with Codex | Shopify’s Mikhail Parakhin&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=BlbxhsItZPg" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;8K&lt;/span&gt;&lt;span class="dt"&gt;31 Eki&lt;/span&gt;&lt;span class="ch"&gt;Cole Medin&lt;/span&gt;&lt;span class="tt"&gt;Anthropic&amp;#x27;s New Study: You&amp;#x27;re Not Controlling AI as Much as You Think You Are&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=kP26OlW5PBk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;8K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Merge CEO Shensi Ding: AI fluency starts at the top&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=A-UVoSTqwjE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;Burke Holland&lt;/span&gt;&lt;span class="tt"&gt;Copilot Automations Run While You Sleep #ai #coding #githubcopilot&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=CQ_Cfw6QxSk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;26 Eki&lt;/span&gt;&lt;span class="ch"&gt;DeepLearningAI&lt;/span&gt;&lt;span class="tt"&gt;7 AI code review best practices #deeplearningai #qodo&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=RQg3HAkNj9Q" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;109&lt;/span&gt;&lt;span class="dt"&gt;30 Eki&lt;/span&gt;&lt;span class="ch"&gt;Agentic Buildcamp: Adaptive Entrepreneurship&lt;/span&gt;&lt;span class="tt"&gt;Mom Test Bot: Build What Customers Actually Need, Not What They Say They Want&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t12"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#b45309"&gt;12&lt;/span&gt;&lt;i class="dot" style="background:#b45309"&gt;&lt;/i&gt;Ajan orkestrasyonu ve LangChain ekosistemi&lt;span class="cnt"&gt;26 video &amp;middot; 290K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Interrupt London kayıtları ağırlığını koydu: zamanla kendini geliştiren ajanlar, tek döngünün yetmediği uzun görevler ve &amp;#x27;zekâyı kendin sahiplen&amp;#x27; tezleri, ajan altyapısının yeni rekabet alanı olduğunu gösterdi.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=xy_7SzwsIjM" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;61K&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;How to Build Agents That Improve Over Time: Model, Harness, Context | Interrupt London&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ujmyzWSj8IU" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;54K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;API credits, now in ChatGPT Pro and Business plans&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=CKFUx7COIwQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;31K&lt;/span&gt;&lt;span class="dt"&gt;26 Eki&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;Workshop: Training Agents at the frontier of AI Industry&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=lcVFk4df5Ms" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;15K&lt;/span&gt;&lt;span class="dt"&gt;26 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;Harness² Explained in 60 Seconds&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=L6pzlo1QvoM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;14K&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;ChatGPT Enterprise at Zendesk: Putting Company Data to Work&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=qIWB1yMsqRo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;14K&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;Cole Medin&lt;/span&gt;&lt;span class="tt"&gt;The 4 Components of Every Successful AI Second Brain (Build These Now)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=VhrtlfTyqp4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;13K&lt;/span&gt;&lt;span class="dt"&gt;26 Eki&lt;/span&gt;&lt;span class="ch"&gt;Sam Witteveen&lt;/span&gt;&lt;span class="tt"&gt;100% Open Image Model Lets You Place Text With JSON&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=rfhhhmajCKc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;12K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;What is LangChain?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=4iYOJ6gyQV4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;11K&lt;/span&gt;&lt;span class="dt"&gt;26 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;LangGraph explained in 1 minute&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=UOfmD1fY-i8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;10K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Building the future of legal work with AI | Legora CEO Max Junestrand&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=M2Oa42hF67E" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;9K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;From legal bottleneck to business partner | Legora’s Emma Kadri Bergström&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=z-WZMjNsg-k" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;8K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Faster contract review. Broader legal analysis. | Legora’s Alex Fortescue-Webb&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=XOUSuxY074U" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;7K&lt;/span&gt;&lt;span class="dt"&gt;30 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;Why Own Your Intelligence? Harrison Chase Keynote | Interrupt London&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Zo2oyjbTEXA" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;5K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;DeepLearningAI&lt;/span&gt;&lt;span class="tt"&gt;Agents on longer tasks need more than one loop&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Wb3hQaG_xf0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;4K&lt;/span&gt;&lt;span class="dt"&gt;26 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;Inside Managed Deep Agents: mda CLI, Context Hub, and Harbor Evals | Interrupt London&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=p1jhV6LlbFY" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;4K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;DeepLearningAI&lt;/span&gt;&lt;span class="tt"&gt;Join Andrew Ng at AI Dev NYC for two days of technical sessions and workshops.&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=GNYHDltnwZQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;ChatGPT Enterprise at Zendesk: Putting Company Data to Work&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=-GbU416ZMoI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;The Real Cost of &amp;quot;We Can Just Build It Ourselves&amp;quot;&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=O3jCAZbrRHM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;Building Semiconductor Agents at Skyworks with LangGraph + LangSmith&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=GNu-d97XY6c" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;We wired Devin into our Slack to triage messages&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=oY1htTx_h9U" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;The Hardest Part of Testing Agents: Write Access&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=VgF5aP0amXA" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;26 Eki&lt;/span&gt;&lt;span class="ch"&gt;DeepLearningAI&lt;/span&gt;&lt;span class="tt"&gt;Learn from OpenAI, Hugging Face, Andrew Ng and Yann LeCun | AI Dev NYC #aidev #shorts&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=y6rrMFO7tgA" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;26 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;Tool Problem or Model Problem? How Clay Tells the Difference&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=uwALRjyYjjs" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;768&lt;/span&gt;&lt;span class="dt"&gt;26 Eki&lt;/span&gt;&lt;span class="ch"&gt;Mervin Praison&lt;/span&gt;&lt;span class="tt"&gt;Why your AI agent stops following your rules&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=MeEmWV2-Nzw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;330&lt;/span&gt;&lt;span class="dt"&gt;31 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;VibeGame: 8 AI Agents Build AND PLAY Their Own Game - CRAZY Claim, I Checked the Paper&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=N_oRytMcOcc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;113&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;Agentic Buildcamp: Adaptive Entrepreneurship&lt;/span&gt;&lt;span class="tt"&gt;No Hallucination: AI Research Companion for Students&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t13"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#0d9488"&gt;13&lt;/span&gt;&lt;i class="dot" style="background:#0d9488"&gt;&lt;/i&gt;Görsel ve video üretim ile multimodal&lt;span class="cnt"&gt;5 video &amp;middot; 53K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;MOSS Image ve ComfyUI akışları metin-yerleşimi kontrollü görsel üretimi öne çıkarırken, Gemini Live ile gerçek zamanlı görüş ve kenar cihaz navigasyonu multimodalın sahaya indiğini gösterdi.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ID6vEy94j7c" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;37K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;Building real-time vision &amp;amp; edge navigation with Gemini Live&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=TFJ2yotzJVc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;12K&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;pixaroma&lt;/span&gt;&lt;span class="tt"&gt;ComfyUI Prompts Explained: CLIP Text Encode, Negative and Dynamic Prompts&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=cv8tEbKtEvk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;&#129716; Building an autonomous plant rover with Gemini Live&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=yiE2l_NMj7Y" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;How To Run Meta Ads For Beginners&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=P6UNRh2pZBs" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;401&lt;/span&gt;&lt;span class="dt"&gt;31 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;How to Make AI Videos for Beginners: INSANE Results From 3 Prompts&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t14"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#d97706"&gt;14&lt;/span&gt;&lt;i class="dot" style="background:#d97706"&gt;&lt;/i&gt;OpenAI ürün ve kurumsal cephe&lt;span class="cnt"&gt;4 video &amp;middot; 123K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;ChatGPT planlarına eklenen API kredileri ve Zendesk, Legora, AT&amp;amp;T gibi kurumsal referans videoları, OpenAI&amp;#x27;nin haftayı tüketici ürününden çok kurumsal satış argümanıyla geçirdiğini gösterdi.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=o4D1J1YWjhg" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;102K&lt;/span&gt;&lt;span class="dt"&gt;30 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;There are many futures. Which do you want to build?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Oi9DHk57KeQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;13K&lt;/span&gt;&lt;span class="dt"&gt;30 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;How Gen uses ChatGPT Work to turn data Into decisions | OpenAI&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Yim7JrVzSkk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;5K&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Lawyers Stay in Control | AT&amp;amp;T + OpenAI&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=7TtGMileCus" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;30 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;How Gen uses ChatGPT Work to turn data Into decisions | OpenAI&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t15"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#dc2626"&gt;15&lt;/span&gt;&lt;i class="dot" style="background:#dc2626"&gt;&lt;/i&gt;Anthropic ve güç dengeleri&lt;span class="cnt"&gt;2 video &amp;middot; 122K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Anthropic&amp;#x27;in olası halka arzında oy gücü tartışması ve &amp;#x27;yapay zekâyı sandığınız kadar siz kontrol etmiyorsunuz&amp;#x27; çalışması, kontrol ve yönetişim gündemini ısıttı.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=JI3bbZNdyeA" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;111K&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;This is much more concerning than you think...&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=jeUgtyvvkfs" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;11K&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Anthropic’s IPO: Who Gets the Voting Power?&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t16"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#2563eb"&gt;16&lt;/span&gt;&lt;i class="dot" style="background:#2563eb"&gt;&lt;/i&gt;İş tarafında pratik yapay zekâ&lt;span class="cnt"&gt;10 video &amp;middot; 18K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;PDF çeviriden bülten kurmaya, 65+ kişilere kişisel asistandan çocuklara İngilizce öğretmenine kadar uçtan uca &amp;#x27;küçük AI işleri&amp;#x27; videoları, üretim tarafının istikrarlı alt dalgasını oluşturdu.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=mMzaL15KFVY" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;11K&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;I Ranked 5 Ways to Wake Up&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=IrgQdsxzRa4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Two AI bots for spam calls and data privacy&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=JMaQ_wVRzjc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;26 Eki&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;How To Build AI Chatbot For Your Website&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Z4okNQnD5c0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;890&lt;/span&gt;&lt;span class="dt"&gt;30 Eki&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;How to Build A Newsletter In 10 Minutes&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=zbNNzPwJLbY" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;889&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;LinkedIn Sales Navigator Tutorial&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=KEA1wOlZrLI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;777&lt;/span&gt;&lt;span class="dt"&gt;31 Eki&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;How To Grow On Instagram Step-By-Step&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Mu2vchtZoE4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;749&lt;/span&gt;&lt;span class="dt"&gt;31 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;DeepTutor&amp;#x27;s CRAZY Open-Source AI Tutor Runs With NO API KEY - I Tested It&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=EtN_aEKjvHc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;145&lt;/span&gt;&lt;span class="dt"&gt;26 Eki&lt;/span&gt;&lt;span class="ch"&gt;Agentic Buildcamp: Adaptive Entrepreneurship&lt;/span&gt;&lt;span class="tt"&gt;I Built an AI Personal Concierge for People Over 65&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Kg3e1iNh6do" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;140&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;Agentic Buildcamp: Adaptive Entrepreneurship&lt;/span&gt;&lt;span class="tt"&gt;From Research Project to AI Startup: The Part Nobody Warns You About&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=aGseoDcCqLI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;137&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;Agentic Buildcamp: Adaptive Entrepreneurship&lt;/span&gt;&lt;span class="tt"&gt;$100 in Ads. Zero Sales. One Painful Lesson.&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t17"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#4d7c0f"&gt;17&lt;/span&gt;&lt;i class="dot" style="background:#4d7c0f"&gt;&lt;/i&gt;Öğren ve geliştir&lt;span class="cnt"&gt;6 video &amp;middot; 40K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;&amp;#x27;1 saatte yapay zekâyı kavratma&amp;#x27; vaatli uzun anlatılar, LoRA inceleme notları ve Hugging Face dergi kulübü; eğitim içeriği haftanın hacim açısından en kalabalık ikinci kulvarı oldu.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=fCEvvkAbJOs" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;19K&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;Zachary Huang&lt;/span&gt;&lt;span class="tt"&gt;Give Me 1 Hour, I Will Make Artificial Intelligence Click Forever&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Xl-yulnYZmI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;15K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;StatQuest with Josh Starmer&lt;/span&gt;&lt;span class="tt"&gt;Support Vector Machines (SVM): Concepts and Code&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=mEtRe_KKYm0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;26 Eki&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;Hugging Face Journal Club: Score Centering Stabilizes Off-policy Reinforcement Learning&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=KdpN5kAX8Rg" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;27 Eki&lt;/span&gt;&lt;span class="ch"&gt;DigitalSreeni&lt;/span&gt;&lt;span class="tt"&gt;01 - Multimodal Spatial Biology: Introducing the Dataset&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=m6Rdp6macAw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;29 Eki&lt;/span&gt;&lt;span class="ch"&gt;DigitalSreeni&lt;/span&gt;&lt;span class="tt"&gt;02 - Multimodal Spatial Biology: Registration&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=F4hQ6alHtN4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;28 Eki&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;You&amp;#x27;re Prompting Claude Wrong #claude&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;
&lt;footer&gt;Bu rapor &lt;b style="color:#111"&gt;Murat Karakaya Akademi&lt;/b&gt; tarafından derlenmiştir · Veri: YouTube RSS (canlı, API key yok) · pencere: 2026-10-25 .. 2026-11-01 (UTC, çalışma anından hesaplanır) · kanal başı max 15 video · Prompt Engineer 48, OpenAI ve AI WITH Rithesh feed&amp;#x27;leri RSS&amp;#x27;teki 15-video sınırı nedeniyle dönemin tamamını kapsamıyor; bu kanallar eksik sayılmış olabilir. · render.py @ 01.11.2026 07:05 UTC&lt;/footer&gt;
&lt;/div&gt;&lt;p&gt;&lt;em&gt;English edition: &lt;a href="https://www.muratkarakaya.net/2026/11/youtube-ai-weekly-trend-report-october.html"&gt;YouTube AI Weekly Trend Report&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;&lt;/body&gt;&lt;/html&gt;</description><link>https://www.muratkarakaya.net/2026/11/youtube-ai-haftalk-trend-raporu-25-ekim.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-6713490338266084153</guid><pubDate>Thu, 29 Oct 2026 06:28:20 +0000</pubDate><atom:updated>2026-10-29T09:28:20.800+03:00</atom:updated><category domain="http://www.blogger.com/atom/ns#">Abonelik Modelleri</category><category domain="http://www.blogger.com/atom/ns#">Acik Modeller</category><category domain="http://www.blogger.com/atom/ns#">Murat Karakaya Akademi</category><category domain="http://www.blogger.com/atom/ns#">Ollama Cloud</category><category domain="http://www.blogger.com/atom/ns#">Pay As You Go</category><category domain="http://www.blogger.com/atom/ns#">Yerel Yapay Zeka</category><title>Ollama Cloud'da abonelik erişim duvarı olmaktan çıktı, kredi çarpanına dönüştü: $0'luk PAYG planı 'her bulut modeli'ni aboneliksiz açıyor, fark artık eşzamanlılık ve toplu alımda</title><description>&lt;!DOCTYPE html&gt;
&lt;!-- ============================================================
   MIS-2026-004 (rev.6 kalıbı) · YAZAR-20261028-OLLAMAPAYG · haber_id 1a17b36d5fc0d819#1
   ANAYASA (ic_kabuk_karari.md Rev.2): içerikte SIFAR görünüm CSS'i; yalnız semantik
   HTML; başlık içerik tarafından basılmaz — ilk görünür içerik SPOT paragrafı.
   Başlık metni yalnız Blogger BAŞLIK ALANINA ve aşağıdaki &lt;title&gt;'a yazılır.
   BAŞLIK: "Ollama Cloud'da abonelik erişim duvarı olmaktan çıktı, kredi çarpanına
   dönüştü: $0'luk PAYG planı 'her bulut modeli'ni aboneliksiz açıyor, fark artık
   eşzamanlılık ve toplu alımda"
   ============================================================ --&gt;
&lt;html lang="tr"&gt;
&lt;head&gt;
&lt;meta charset="utf-8"&gt;
&lt;meta name="viewport" content="width=device-width, initial-scale=1"&gt;
&lt;title&gt;Ollama Cloud'da abonelik erişim duvarı olmaktan çıktı, kredi çarpanına dönüştü: $0'luk PAYG planı 'her bulut modeli'ni aboneliksiz açıyor, fark artık eşzamanlılık ve toplu alımda&lt;/title&gt;
&lt;/head&gt;
&lt;body&gt;

  &lt;p&gt;Ollama'nın 27 Ekim sabahı gelen bülteni, fiyat sayfasının ilk kartına çoktan oturmuş bir değişikliği müjdeliyordu: "Pay as you go" adında, 0 dolarlık bir plan; abonelik yok, her bulut modeli token başına açık, eşzamanlı istek hakkı üç. Mekanizma yeni değil — 31 Ağustos'taki blog yazısı "abonelik gerekmez" diye yazmış, 23 Eylül'de kredi fiilen açılmıştı. Değişen şey şu: iki aydır fiyat tablosunun altında gizli duran seçenek, adlı adıyla birinci sınıf bir plana dönüştü ve eşzamanlılık 1'den 3'e çıktı. Sayfanın geri kalanıysa asıl haberi veriyor: abonelik artık kapıyı açan anahtar değil, toplu kredi alımının fiyatı.&lt;/p&gt;

  &lt;p&gt;&lt;strong&gt;Bu Yazıda&lt;/strong&gt;&lt;/p&gt;
  &lt;ol&gt;
    &lt;li&gt;&lt;a href="#bol-1"&gt;Arşivin tuttuğu iz: mekanizma eskiden vardı, plan yeni&lt;/a&gt;&lt;/li&gt;
    &lt;li&gt;&lt;a href="#bol-2"&gt;Abonelik artık ne demek: anahtar değil, çarpan&lt;/a&gt;&lt;/li&gt;
    &lt;li&gt;&lt;a href="#bol-3"&gt;Okurun matematiği: token fiyatları, off-peak saatleri ve HN'in kuşkusu&lt;/a&gt;&lt;/li&gt;
    &lt;li&gt;&lt;a href="#bol-4"&gt;Sıfır veri tutma: güçlü beyan, ama beyan&lt;/a&gt;&lt;/li&gt;
    &lt;li&gt;&lt;a href="#ka-yorum"&gt;Murat Karakaya Akademi Yorumu&lt;/a&gt;&lt;/li&gt;
    &lt;li&gt;&lt;a href="#kaynakca"&gt;Kaynakça&lt;/a&gt;&lt;/li&gt;
  &lt;/ol&gt;

  &lt;h2 id="bol-1"&gt;1. Arşivin tuttuğu iz: mekanizma eskiden vardı, plan yeni&lt;/h2&gt;
  &lt;p&gt;Aboneliksiz kredi, 27 Ekim'de icat olmadı; Ollama'nın kendisi mekanizmayı iki ay önce duyurmuş. 31 Ağustos tarihli "Ollama's transparent pricing" yazısının ilgili cümlesi net: "Add usage credits to access all models on the free plan with pay-as-you-go pricing: no service fees, no subscription required" [1]. 23 Eylül'de resmî X hesabı bir adım daha attı: "You can now add usage credits for paid cloud models without an Ollama subscription" [2]. Yani duvarın kilidi fiilen o gün sökülmüş.&lt;/p&gt;
  &lt;p&gt;Fiyat sayfası ise duyurunun gerisinden geldi. 23 Eylül anlık görüntüsünde sayfada "Pay as you go" diye bir plan kartı yok; "pay as you go with usage credits" yalnızca model fiyat tablosunun altındaki bir açıklama satırı [3]. 26 Ekim 20:15 UTC çekiminde tablo hâlâ aynı; SSS'de "Free includes 1 concurrent request, Pro 3, Max and Team 10" yazıyor [4]. Bugünkü canlı sayfada ise ilk kart şu: "Pay as you go — $0 — Pay per token for every cloud model. Includes: Run models locally · No subscription · 3 concurrent requests · No service fees"; SSS de "Pay as you go and Pro include 3 concurrent requests" şekline dönmüş, "Free/1 eşzamanlı" dili sayfadan tamamen kalkmış [5].&lt;/p&gt;
  &lt;p&gt;Demek ki değişim penceresi 26 Ekim akşamı ile 28 Ekim arası; tam saati hiçbir yerde duyurulmadı, arşiv çekimlerinin arasına sıkıştı [4][5]. Elde tutulabilir iki ürün değişikliği var: eşzamanlılığın 1'den 3'e çıkması ve "başlangıç modelleri + küçük kredi"den ibaret eski ücretsiz katman çerçevesinin, adlandırılmış bir $0 planıyla değiştirilmesi. "Abonelik duvarı ilk kez kalktı" demek, 31 Ağustos metnini ve 23 Eylül gönderisini kayıtlardan silmek olur; bu yazı o sözü etmiyor.&lt;/p&gt;

  &lt;h2 id="bol-2"&gt;2. Abonelik artık ne demek: anahtar değil, çarpan&lt;/h2&gt;
  &lt;p&gt;PAYG kimlik kazanınca Pro ve Max'in varlık gerekçesi yeniden yazılmak zorunda; Ollama bunu kart sırasına bile işlemiş: "Everything in Pay as you go, plus" [5]. Pro, ayda 20 dolar karşılığında 60 dolarlık dahil kredi veriyor (yıllık 200 dolar). Max, 100 dolara 300 dolar kredi, erken erişim ve 10 eşzamanlı istek ekliyor. Team, 500 dolara 1.000 dolarlık paylaşılan kredi [5]. Hesap basit: abonelik, yaklaşık üç kat kredi çarpanı, eşzamanlılık ve erken erişim demeti.&lt;/p&gt;
  &lt;p&gt;Bir dürüstlük katmanı borçlanalım: Pro kartında "Access to larger pro models" ibaresi duruyor — bu, PAYG'ın başlığındaki "every cloud model" ile gerilimli [5]. Hangi modelin PAYG kapsamında olduğu hangisinde olmadığı resmî tabloda listelenmiyor; yani "her model" iddiası, kendi hesabınızın gördüğü canlı model listesi kadar doğrulanmış. Aboneliğe bir üst kademe model uğruna geçmeyi düşünen, geçmeden o modelin PAYG hesabından yanıt verip vermediğine baksın — bir akşamı gider, bir aylık abonelikten ucuz. Kartlardaki "Fast mode (coming soon)" ve Team'in "shared projects" satırları da tarihsiz vaat; takvim yok [5].&lt;/p&gt;

  &lt;h2 id="bol-3"&gt;3. Okurun matematiği: token fiyatları, off-peak saatleri ve HN'in kuşkusu&lt;/h2&gt;
  &lt;p&gt;PAYG'ın hesabı token düzeyinde dönüyor, o yüzden canlı tabloya 1M token başına fiyatlarla bakmakta fayda var. deepseek-v4.1-flash: girdi 30 sent, önbellek 0,6 sent, çıktı 1 dolar 20 sent; off-peak'te tam yarıya iniyor — 15 sent / 0,3 sent / 60 sent. minimax-m3.1-flash 15 sent girdi / 60 sent çıktı; nemotron-3-super'in girdisi 1,5 sent gibi komik bir yerde dururken çıktısı 60 sent; gpt-oss:20b 7/30 sent; glm-5.3-flash 15/50 sent; aralarında en pahalısı kimi-k3 3/15 dolar [5]. Off-peak, hafta içi 12:00–18:00 UTC dışı ve tüm hafta sonu demek [5] — takvimi kendisi belirleyen işler yarı fiyat koşuyor. Fiyat sürekliliği de arşivden doğrulanıyor: deepseek-v4.1-flash'ın rakamları 23 Eylül ve 26 Ekim çekimlerinde birebir aynı; tabloda değişen model giriş çıkışı, fiyat sürünmesi değil [3][4][5].&lt;/p&gt;
  &lt;p&gt;Şunu da ayıralım: 27 Ekim e-posta bülteni DeepSeek-V4.1-Flash için "1,5 kat hız" diyor — bu oran Ollama'nın model kartında, DeepSeek'in resmî duyurusunda ya da bilinen hiçbir bağımsız ölçümde geçmiyor; neye kıyasla verildiği de tanımsız [10][11][12]. Yalnızca bülten iddiası olarak kayıtta duruyor, üstüne hesap kurulmaz.&lt;/p&gt;
  &lt;p&gt;Topluluğun defteri iki renkli. Temmuzda açılan "Ollama: All Aboard Open Models" başlığının altında elli yedi yorumluk bir kuyruk birikti: Ollama'nın en eski eleştirisi — buluta kayış güvensizliği — orada duruyor. Fiyatın savunulduğu taraf ise haziranda yazıldı: biri, Ollama Cloud faturasının Opus API oranlarının yüzde yirmisinin altında kaldığını ("1.40/4.40 vs 5/25") hatırlatıp fiyat/özellik dengesini savundu [9]. Reddit'te "ücretsiz katman öldü mü" sorusuna "hâlâ var" yanıtı gelmesi de sayfaların karışıklığının ölçütü [13]; resmi arşiv kaydımıza göre olan, eski Free çerçevesinin PAYG'a dönüşmesiydi [3][4][5].&lt;/p&gt;

  &lt;h2 id="bol-4"&gt;4. Sıfır veri tutma: güçlü beyan, ama beyan&lt;/h2&gt;
  &lt;p&gt;PAYG'ın sessiz satış argümanı gizlilik tarafı. Canlı SSS kategorik: "Prompt or response data is never logged or trained on" [5]. Barındırma NVIDIA Cloud Providers üzerinden yapılıyor ve Ollama, yüklenicilerinden logging yasağı, eğitim yasağı ve sıfır veri tutma şartlarını sözleşmeyle istediğini söylüyor [5][7]. 17 Ekim tarihli resmî X gönderisi "always zero data retention, hosted in US and EU with no training" diyor [6]; 31 Ağustos blog yazısı "dedicated compute in the US and Europe" cümlesini kuruyor [1]. Üç resmî kaynak aynı cümleyi tekrarlıyor; NCP sözleşmelerinin bu şartı üçüncü taraflara dayatması, beyanı sektör ortalamasının üstüne taşıyor [7].&lt;/p&gt;
  &lt;p&gt;Ama beyanın sınırı var, çizgiyi çekelim: barındırma "primarily" ABD ağırlıklı, AB gerekçeyle devrede; toplulukta dolaşan Singapur rotası (sınırlı Qwen modelleri) ikincil aktarım, kapsamı resmî olarak doğrulanamadı [7][8][9]. Hesap bilgisi ve içeriği olmayan kullanım meta verisi toplanmaya devam ediyor; krediler devretmiyor [8]. tapbit'in ayrımı bu bölümün özeti: sıfır veri tutma bir politika beyanıdır, yerel çalıştırma değil [8]. Bağımsız bir denetim raporu da olmadığı sürece cümlenin dürüst hali şu: beyan güçlü ve tekrarlı, ölçüm yok.&lt;/p&gt;

  &lt;h2 id="ka-yorum"&gt;Murat Karakaya Akademi Yorumu&lt;/h2&gt;
  &lt;p&gt;Verilerin söylediği: bu bir teknoloji değişikliği değil, fiyat sayfasının yeniden yazımı. Ölçülebilir olan iki satır — eşzamanlılık 1'den 3'e, "every cloud model" başlıklı adlandırılmış $0 plan [4][5] — arşivle tarihli. Gerisi (31.08 mekanizma, 23.09 açılım) zaten masadaydı [1][2]. Ollama'nın yaptığı, tezgâhın altına sakladığı anahtarı tezgâhın üstüne koymak.&lt;/p&gt;
  &lt;p&gt;Kanaatimize gelince: asıl değişim karar değişkeninin yer değiştirmesi. Artık "hangi bulut modeline erişebiliyorum?" sorusu sorulmuyor; soru şu: "bütçem ve takvimim ne?" Okur için üç somut yol haritası. Bir: düzensiz kullanan, ajandan görev geçiren okur PAYG + off-peak evinde oturur — akşam saatlerinde minimax-m3.1-flash ya da deepseek-v4.1-flash'a 15/60 sentlik 1M-token fiyatlarıyla çalışmak, abonelik bağı olmadan birkaç dolarlık aylık fatura demektir; üç eşzamanlı istek küçük bir hat kurmaya yeter. İki: her ay 20 dolarlık krediyi gerçekten dolduran düzenli kullanıcı Pro'ya bakar — 20 dolarlık ödeme 60 dolar kredi oluyor, çarpanı boşa harcamaya gerek yok; ama geçme sebebi "daha büyük pro modeller" ise başlığa değil kendi PAYG hesabınıza güvenin, Pro kartındaki ibare belirsiz [5]. Üç: uzun vadeli, yüksek hacimli ya da veriye duyarlı iş yerel + PAYG hibrit — yerel taraf sıfır marjinal maliyet ve sayfanın kendi deyimiyle her zaman sınırsız, bulut tarafı makinenin taşıyamadığı model için geçici hızlanma [5]. Düzenli veri işleyenler için not: bağımsız ZDR denetimi yokken hassas veriyi beyana güvenmek için buluta göndermeyin.&lt;/p&gt;
  &lt;p&gt;TR okuruna tablonun görünmeyen sütunu: fiyatların tamamı dolar üzerinden. Dolar/TL, bu hesabın aylık faturadaki gizli satırı; PAYG'ın yerel cüzdana ucuzluğu kurun yatay olduğu aylarla sınırlı — masaya otururken kuru da ekleyin.&lt;/p&gt;

  &lt;h2 id="kaynakca"&gt;Kaynakça&lt;/h2&gt;
  &lt;ol&gt;
    &lt;li&gt;ollama.com/blog — &lt;a href="https://ollama.com/blog/transparent-pricing"&gt;"Ollama's transparent pricing"&lt;/a&gt; (31.08.2026; aboneliksiz kredi mekanizmasının ilk resmî ilanı) · erişim: 28.10.2026&lt;/li&gt;
    &lt;li&gt;x.com/@ollama — &lt;a href="https://x.com/ollama/status/2102903557624352905"&gt;PAYG kredilerinin aboneliksiz açıldığı duyurusu&lt;/a&gt; (23.09.2026) · erişim: 28.10.2026&lt;/li&gt;
    &lt;li&gt;web.archive.org — &lt;a href="http://web.archive.org/web/20260923191554/https://ollama.com/pricing"&gt;ollama.com/pricing anlık görüntüsü&lt;/a&gt; (23.09.2026 19:15 UTC; PAYG plan kartı yok) · erişim: 28.10.2026&lt;/li&gt;
    &lt;li&gt;web.archive.org — &lt;a href="http://web.archive.org/web/20261026201533/https://ollama.com/pricing"&gt;ollama.com/pricing anlık görüntüsü&lt;/a&gt; (26.10.2026 20:15 UTC; Free 1 eşzamanlı, PAYG kartı yok) · erişim: 28.10.2026&lt;/li&gt;
    &lt;li&gt;ollama.com/pricing — &lt;a href="https://ollama.com/pricing"&gt;canlı sayfa&lt;/a&gt; (28.10.2026 çekimi; PAYG $0 kartı, plan kartları, token fiyat tablosu, off-peak tanımı, ZDR SSS) · erişim: 28.10.2026&lt;/li&gt;
    &lt;li&gt;x.com/@ollama — &lt;a href="https://x.com/ollama/status/2111587278271426605"&gt;"Ollama Cloud is always zero data retention"&lt;/a&gt; (17.10.2026) · erişim: 28.10.2026&lt;/li&gt;
    &lt;li&gt;note.com (0x__tom) — &lt;a href="https://note.com/0x__tom/n/n9b37e363b995"&gt;Ollama Cloud ZDR ve fiyatlandırma geçişi analizi&lt;/a&gt; (NCP sözleşme şartı notu) · erişim: 28.10.2026&lt;/li&gt;
    &lt;li&gt;tapbit.com — &lt;a href="https://www.tapbit.com/en/learn/article/ollama-cloud-pricing-privacy-open-model-business-20260903"&gt;"Ollama Cloud pricing &amp;amp; privacy" iş analizi&lt;/a&gt; (03.09.2026; ZDR politika/yerel ayrımı, kredi devri yok) · erişim: 28.10.2026&lt;/li&gt;
    &lt;li&gt;news.ycombinator.com — &lt;a href="https://news.ycombinator.com/item?id=48965880"&gt;story id=48965880 "Ollama: All Aboard Open Models"&lt;/a&gt; (19.07.2026, 57 yorum; buluta kayış güvensizliği yorumları) ve &lt;a href="https://news.ycombinator.com/item?id=48628488"&gt;yorum id=48628488&lt;/a&gt; (22.06.2026; fiyat/özellik savunması, "1.40/4.40 vs 5/25") · erişim: 29.10.2026&lt;/li&gt;
    &lt;li&gt;ollama.com/library — &lt;a href="https://ollama.com/library/deepseek-v4.1-flash"&gt;deepseek-v4.1-flash model kartı&lt;/a&gt; (CED mimarisi, KV cache, 1M bağlam; "1.5x" ifadesi kartta yok) · erişim: 28.10.2026&lt;/li&gt;
    &lt;li&gt;x.com/@ollama — &lt;a href="https://x.com/ollama/status/2098290160035373524"&gt;DeepSeek-V4.1-Flash tam yaygınlaşma gönderisi&lt;/a&gt; (11.09.2026; hız oranı vermiyor) · erişim: 28.10.2026&lt;/li&gt;
    &lt;li&gt;deepseek.com — &lt;a href="https://www.deepseek.com/en/news/deepseek-v4-1-flash/"&gt;"DeepSeek V4.1 Flash" resmî duyurusu&lt;/a&gt; ("multiple parties" kıyası; 1.5x oranı yok) · erişim: 28.10.2026&lt;/li&gt;
    &lt;li&gt;reddit.com/r/ollama — &lt;a href="https://www.reddit.com/r/ollama/comments/1w6xhcl/free_models_in_ollama_clouds_new_subscription/"&gt;"Free models in Ollama Cloud's new subscription scheme?" tartışması&lt;/a&gt; · erişim: 28.10.2026&lt;/li&gt;
  &lt;/ol&gt;

  &lt;p&gt;Etiketler: #MuratKarakayaAkademi #OllamaCloud #PayAsYouGo #AcikModeller #BulutAPI #YerelYZ&lt;/p&gt;

  &lt;footer&gt;
    &lt;p&gt;Yayın tarihi: 28 Ekim 2026 · muratkarakaya.net&lt;/p&gt;
    &lt;p&gt;Murat Karakaya Akademi&lt;/p&gt;
  &lt;/footer&gt;

&lt;/body&gt;
&lt;/html&gt;
CEO-RAPOR: tamamlandı
</description><link>https://www.muratkarakaya.net/2026/10/ollama-cloudda-abonelik-erisim-duvar.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-2112450916644932627</guid><pubDate>Tue, 27 Oct 2026 16:03:59 +0000</pubDate><atom:updated>2026-10-27T19:03:59.238+03:00</atom:updated><category domain="http://www.blogger.com/atom/ns#">AI</category><category domain="http://www.blogger.com/atom/ns#">MiniMax M3.1 Flash</category><category domain="http://www.blogger.com/atom/ns#">Murat Karakaya Akademi</category><category domain="http://www.blogger.com/atom/ns#">Ollama</category><category domain="http://www.blogger.com/atom/ns#">Open Weights</category><title>MiniMax's M3.1 Flash sits ahead of several open-weight rivals on independent scores — at half its claimed speed, and still without weights or a license</title><description>&lt;!-- draft_en_1a157e1078db2515.html — MIS-2026-004 K9 EN sürümü · haber_id 1a157e1078db2515#1
     TEMA ANAYASASI: içerik parçası; sıfır CSS (style/renk/font yok), semantik HTML, içerikte h1 yok,
     spot paragrafla başlar; kaynakça numaralı &lt;ol&gt;; "Labels:" şeridi 6 etiket (marka başta); imza footer'da.
     BAŞLIK (meta alanı — gövdeye basılmaz): MiniMax's M3.1 Flash sits ahead of several open-weight rivals on independent scores — at half its claimed speed, and still without weights or a license --&gt;
&lt;p&gt;The launch arrived with a masked model and left with an argument. MiniMax's new agent model now sits ahead of several open-weight rivals on independent scoring — at roughly half the speed the company advertised, with no weights in sight and no license named.&lt;/p&gt;

&lt;p&gt;For well over a week, an anonymous model ran on OpenRouter under the name "Space Bunny." Nobody knew whose it was. The community tallied more than 200 trillion tokens of traffic and watched it climb toward the top of global leaderboards. On October 19, MiniMax stepped forward and admitted the mask was its own: an early build of M3.1 Flash, the company's new code-and-agent model [1][2][15]. Since then it has been live through the MiniMax API, the M Plan subscription and the MiniMax Code app [1][9]. The weights are nowhere to be found. The company says they "will be released later" and has not uttered a word about licensing [1][7]. Any of that would make a serviceable launch post. The interesting part is the shape of the debut: a closed-weight model walked on stage at the exact moment it was about to overtake the open-weight league on independent scoring — then tethered its first week of access to a cloud promotion.&lt;/p&gt;

&lt;section&gt;
&lt;h2&gt;1. The lab behind the mask&lt;/h2&gt;
&lt;p&gt;MiniMax is a publicly listed Chinese company (HKEX: 00100), and the playbook here is a familiar one: release the model anonymously, let curiosity compound, then drop the official blog post. OpenRouter's community spent weeks guessing who "Space Bunny" belonged to; when the announcement landed, the surprise itself became the press release [1][8]. The same day, MiniMax's X account celebrated the model shipping with "Day0" support on Ollama [8]. That pairing isn't new either — M2 already joined Ollama's cloud back in October 2025 [6].&lt;/p&gt;
&lt;p&gt;On the numbers side, the company is positioning the whole "Flash" segment from scratch. The old M3 sat at 29 points on Artificial Analysis' Intelligence Index, costing about $0.51 per task; M3.1 Flash lands at 43 points and roughly $0.22 [11]. MiniMax's "half the price, three times the speed" pitch rests entirely on the distance between those two rows — and for now it is the company's own claim; nobody outside has reproduced it [1][15].&lt;/p&gt;
&lt;p&gt;The company's benchmark sheet falls in the same category: 62.2 on Automation-Bench, 73.7 on DeepSWE, 76.2 on BankerToolBench. Then there is Terminal Bench 4.0, where it drops to 38.1 against Opus 5's 51.8 — the quiet corner of that table [1][12][13]. Worth remembering that all of these arrive from a marketing document. The independent table below keeps a different set of books.&lt;/p&gt;
&lt;/section&gt;

&lt;section&gt;
&lt;h2&gt;2. One model, two speed numbers, one honest read&lt;/h2&gt;
&lt;p&gt;Speed is the messiest line of this launch. MiniMax's blog post cites a peak above 200 TPS and a P50 north of 170 [1] — measured on the company's own infrastructure, by the company's own instruments. Artificial Analysis, an independent evaluator, had output speed at roughly 93 tokens per second on the evening of October 21, against a class median of 113 [17]. That is about half the advertised rate, sitting below the middle of the pack. (Housekeeping note: the retrievable snapshot of AA's model page is frozen at an older speed value, so the current number comes from its comparison page [10][17]. These measurements drift; this article says "~93 tok/s, accessed Oct 21" and leaves it there.)&lt;/p&gt;
&lt;p&gt;The context window carries a similar marketing-versus-practice gap. MiniMax's documentation and Artificial Analysis both cite 1,000,000 tokens [7][10]; Ollama lists the actually usable ceiling at 976K [3]. "1M context" is the product name. 976K is the number left in your account.&lt;/p&gt;
&lt;p&gt;Here is the model's pedigree as far as we can actually establish it. Ollama's page shows 428B parameters [3]. That it is a mixture-of-experts architecture with roughly 23B active parameters, sparse attention and a built-in vision encoder comes from MarkTechPost's reporting — MiniMax has published no technical report, so we cannot verify it [12]. Text and image inputs are safe claims; the docs also mention video, but no data shows video working at the Ollama end [3][7][10]. Effort spans five levels from &lt;code&gt;low&lt;/code&gt; to &lt;code&gt;max&lt;/code&gt;, and reasoning stays on throughout [7]. Integrations written against the old M3 will trip immediately: calls that set &lt;code&gt;thinking.type=disabled&lt;/code&gt; or &lt;code&gt;effort=none&lt;/code&gt; now return errors, and reasoning text arrives in a separate &lt;code&gt;reasoning_content&lt;/code&gt; field — anyone migrating meets this break on the very first call [12].&lt;/p&gt;
&lt;/section&gt;

&lt;section&gt;
&lt;h2&gt;3. The scorecard: cheap, not fast, and not exactly trustworthy&lt;/h2&gt;
&lt;p&gt;The class photo, with Artificial Analysis data accessed October 21:&lt;/p&gt;
&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;&lt;th&gt;Model&lt;/th&gt;&lt;th&gt;Intelligence Index&lt;/th&gt;&lt;th&gt;Cost per task&lt;/th&gt;&lt;th&gt;Notes&lt;/th&gt;&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;&lt;td&gt;MiniMax M3.1 Flash (max)&lt;/td&gt;&lt;td&gt;43 (42 at launch)&lt;/td&gt;&lt;td&gt;~$0.22&lt;/td&gt;&lt;td&gt;4/173; 48% HLE, 22% CritPt; 93 tok/s, below median&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;GLM-5.3-Flash (max)&lt;/td&gt;&lt;td&gt;42&lt;/td&gt;&lt;td&gt;~$0.28&lt;/td&gt;&lt;td&gt;Neck and neck with M3.1 in launch analysis&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;DeepSeek V4.1 Flash 1017 (max)&lt;/td&gt;&lt;td&gt;45&lt;/td&gt;&lt;td&gt;~$0.26&lt;/td&gt;&lt;td&gt;Class leader on AA; MIT-licensed open weights&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;Gemini 3.8 Flash (high)&lt;/td&gt;&lt;td&gt;41&lt;/td&gt;&lt;td&gt;~$1.24&lt;/td&gt;&lt;td&gt;Roughly 5.6x M3.1's task cost&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;MiniMax M3 (previous gen)&lt;/td&gt;&lt;td&gt;29&lt;/td&gt;&lt;td&gt;~$0.51&lt;/td&gt;&lt;td&gt;The generation before&lt;/td&gt;&lt;/tr&gt;
&lt;tr&gt;&lt;td&gt;Class median&lt;/td&gt;&lt;td&gt;13&lt;/td&gt;&lt;td&gt;—&lt;/td&gt;&lt;td&gt;—&lt;/td&gt;&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;&lt;em&gt;Table: Artificial Analysis model and comparison pages plus launch analysis [10][11][17] · accessed Oct 21, 2026.&lt;/em&gt;&lt;/p&gt;
&lt;p&gt;The rows need reading between the lines. The GLM entry is GLM-5.3-Flash (42 points, ~$0.28); the similarly named "GLM-5.3 (Max)" is a different, much bigger model that charges $2.01 per task [11][17]. DeepSeek V4.1 Flash both leads the class and ships MIT-licensed — you can pull the weights and run them on your own metal [11][12]. M3.1 Flash's 43 is genuinely impressive; the price for it is standing in a cloud service's queue at $0.15 per million input tokens and $0.60 per million output [1][3][7].&lt;/p&gt;
&lt;p&gt;Then there is the hallucination metric nobody is advertising. On AA-Omniscience, M3.1 Flash attempts 98% of questions and stays 56% correct; among the ones it does not know, it answers wrongly rather than saying "I don't know" in 93% of cases (overall score 16) [11]. As VentureBeat also points out, that does not mean "93% of its answers are wrong" [18]. What it does mean: the model fabricates where it is unsure. If your agent chains cannot afford a silently wrong link, building the verification layer is your job, not the model's.&lt;/p&gt;
&lt;/section&gt;

&lt;section&gt;
&lt;h2&gt;4. The marketing calendar arrived before the licensing desk&lt;/h2&gt;
&lt;p&gt;Why cloud-first rather than open weights? No official answer exists; the competitive table tells its own story. DeepSeek publishes under MIT while MiniMax defers its weights to "later" and opens its first week as a three-day free campaign on Ollama's cloud [1][11]. Ollama's model page reads "free until Oct 22, 2026 11:59 PM UTC" — 02:59 Turkey time on October 23 — after which normal metered pricing resumes [3][5]. Not a single extra condition, no quota, no terms, is printed anywhere for the campaign [3][5]. A small note for the record: Ollama's blog carries exactly zero posts about this launch. The entire announcement consists of a model page and two X posts [4][5][6].&lt;/p&gt;
&lt;blockquote&gt;
&lt;p&gt;"Model weights will be released later." — MiniMax official blog announcement, Oct 19, 2026 [1]&lt;/p&gt;
&lt;/blockquote&gt;
&lt;p&gt;The journalistic weight of that detail fits in one sentence: the free window is a marketing budget, not the model. The actual argument about this model lives in the licensing gap. MiniMax's platform documentation calls the model "open-weight" [7]; Venice's registry lists it as closed/proprietary [13]. Both cannot be true, and no official source will name a license. The previous M3 shipped under a Community License requiring attribution and written permission from anyone earning above $20 million a year; what governs M3.1 has not been stated [13][14]. On Ollama the only tag is &lt;code&gt;:cloud&lt;/code&gt;, and the model runs on Ollama's US infrastructure; "zero data retention" stands as Ollama's policy statement [3][4]. "Run it locally" is simply not a sentence you can say today.&lt;/p&gt;
&lt;p&gt;What you can responsibly do today is a measured test. At the time of writing, the free window has roughly a day left (closing October 23, 02:59 TRT). A sensible sequence: run one end-to-end coding task through &lt;code&gt;ollama run minimax-m3.1-flash:cloud&lt;/code&gt;, then wire it into Claude Code with &lt;code&gt;ollama launch claude --model minimax-m3.1-flash:cloud&lt;/code&gt; and hand it a PR-sized job in a real repository [3][4]. Run the same tasks against MIT-licensed DeepSeek V4.1 Flash for comparison. Watch the terminal-heavy jobs and the factual questions in particular — those are the two known weak spots [11][12]. A weight release is expected around October 23; if it lands, the license text should be the first thing you read [13][15].&lt;/p&gt;
&lt;/section&gt;

&lt;section&gt;
&lt;h2&gt;Murat Karakaya Academy verdict&lt;/h2&gt;
&lt;p&gt;What the data says: an Intelligence Index of 43 at ~$0.22 per task is a very low price for performance more than triple the class median of 13 — and that is independent measurement speaking [10][11]. The same data draws an equally clear boundary: 56% factual accuracy and a habit of fabricating in 93% of the questions it doesn't know rule this model out of any workflow that lacks verification [11][18]. The company's own benchmark sheet is a bundle of claims with no independent replication; telling the two apart is hard but possible, and we have kept them apart throughout this piece.&lt;/p&gt;
&lt;p&gt;Our opinion: this launch is a story about distribution, not model competition. A lab marketed its strongest code model through a masked stealth run, keeps its weights locked, and tied first-week access to a cloud partner's free campaign — functionally neutralizing, for the first week, the run-it-locally advantage that gives the open-weight community any leverage at all. Until a license text exists, this model is a product, not infrastructure. Our advice to readers: blind-test it on a free day, calibrate your impressions against DeepSeek, whose numbers are measurable and whose license is clean, and make the production decision once the weights and the license actually surface.&lt;/p&gt;
&lt;/section&gt;

&lt;section&gt;
&lt;h2&gt;References&lt;/h2&gt;
&lt;ol&gt;
&lt;li&gt;minimax.io — &lt;a href="https://www.minimax.io/blog/minimax-m31f"&gt;"MiniMax M3.1 Flash: Redefining Agent Performance and Cost"&lt;/a&gt; (Oct 19, 2026) · accessed Oct 21, 2026&lt;/li&gt;
&lt;li&gt;ir.minimax.io — &lt;a href="https://ir.minimax.io/news-events/new-releases"&gt;"New Releases" announcement list&lt;/a&gt; (Oct 19, 2026) · accessed Oct 21, 2026&lt;/li&gt;
&lt;li&gt;ollama.com — &lt;a href="https://ollama.com/library/minimax-m3.1-flash"&gt;"minimax-m3.1-flash" model page&lt;/a&gt; (428B, 976K ctx, pricing, free-until date, commands) · accessed Oct 21, 2026&lt;/li&gt;
&lt;li&gt;x.com/@ollama — &lt;a href="https://x.com/ollama/status/2112249324398858505"&gt;Official launch post&lt;/a&gt; (Oct 19, 2026; 1M ctx, US-hosted, zero data retention) · accessed Oct 21, 2026&lt;/li&gt;
&lt;li&gt;x.com/@ollama — &lt;a href="https://x.com/ollama/status/2112358514228363362"&gt;"3 days free on Ollama's Cloud"&lt;/a&gt; (Oct 20, 2026) · accessed Oct 21, 2026&lt;/li&gt;
&lt;li&gt;ollama.com/blog — &lt;a href="https://ollama.com/blog"&gt;Blog archive&lt;/a&gt; (Oct 2025 M2 post; absence of any M3.1 post) · accessed Oct 21, 2026&lt;/li&gt;
&lt;li&gt;platform.minimax.io — &lt;a href="https://platform.minimax.io/docs/guides/text-generation"&gt;Text generation documentation&lt;/a&gt; (1M ctx, effort levels, "open-weight" phrasing) · accessed Oct 21, 2026&lt;/li&gt;
&lt;li&gt;x.com/@MiniMax_AI — &lt;a href="https://x.com/MiniMax_AI/status/2112281693642227753"&gt;Day0 Ollama support + Space Bunny confirmation&lt;/a&gt; (Oct 19, 2026) · accessed Oct 21, 2026&lt;/li&gt;
&lt;li&gt;agent.minimax.io — &lt;a href="https://agent.minimax.io/tools/m3.1-flash"&gt;MiniMax Agent product page&lt;/a&gt; (M3.1 Flash GA announcement) · accessed Oct 21, 2026&lt;/li&gt;
&lt;li&gt;artificialanalysis.ai — &lt;a href="https://artificialanalysis.ai/models/minimax-m3-1-flash"&gt;"MiniMax-M3.1-Flash" model page&lt;/a&gt; (Index 43, 4/173, ~$0.22/task, 1M ctx; stale cached speed value — current speed per [17]) · accessed Oct 21, 2026&lt;/li&gt;
&lt;li&gt;x.com/@ArtificialAnlys — &lt;a href="https://x.com/ArtificialAnlys/status/2112211865128608002"&gt;Launch benchmark analysis&lt;/a&gt; (48% HLE, 22% CritPt, AA-Omniscience: 98% attempt rate, 93% wrong answers among unknowns, overall score 16) · accessed Oct 21, 2026&lt;/li&gt;
&lt;li&gt;marktechpost.com — &lt;a href="https://www.marktechpost.com/2026-10-19/minimax-m3-1-flash-release/"&gt;"MiniMax M3.1 Flash"&lt;/a&gt; (428B MoE, ~23B active claim, benchmark table, migration notes; Oct 19, 2026) · accessed Oct 21, 2026&lt;/li&gt;
&lt;li&gt;pc.watch.impress.co.jp — &lt;a href="https://pc.watch.impress.co.jp/docs/news/2149440.html"&gt;Space Bunny reveal, license analysis, M3 Community License terms&lt;/a&gt; · accessed Oct 21, 2026&lt;/li&gt;
&lt;li&gt;faradaymachines.com — &lt;a href="https://www.faradaymachines.com/minimax-m3-1-flash-cost-per-task/"&gt;License gap and cost-per-task analysis&lt;/a&gt; · accessed Oct 21, 2026&lt;/li&gt;
&lt;li&gt;unite.ai — &lt;a href="https://www.unite.ai/minimax-launches-m3-1-flash-revealing-space-bunny-stealth-model/"&gt;"MiniMax Launches M3.1 Flash, Revealing Space Bunny Stealth Model"&lt;/a&gt; ("topped global leaderboards" phrasing) · accessed Oct 21, 2026&lt;/li&gt;
&lt;li&gt;daily.dev — &lt;a href="https://daily.dev/posts/minimax-m3-1-flash-is-cheap-free-this-week-and-everywhere-7uccejy7e"&gt;Roundup of the free week and the mixed benchmark picture&lt;/a&gt; · accessed Oct 21, 2026&lt;/li&gt;
&lt;li&gt;artificialanalysis.ai — &lt;a href="https://artificialanalysis.ai/models/comparisons/minimax-m3-1-flash-vs-glm-5-3"&gt;"M3.1-Flash vs GLM-5.3 (Max)" comparison page&lt;/a&gt; (live output speed; GLM-5.3 (Max) distinct from the Flash variant) · accessed Oct 21, 2026&lt;/li&gt;
&lt;li&gt;venturebeat.com — &lt;a href="https://venturebeat.com/technology/minimaxs-new-m3-1-flash-beats-deepseek-and-opus-5-on-a-key-business-benchmark-at-lower-cost-but-with-a-high-hallucination-rate"&gt;"MiniMax's new M3.1 Flash beats DeepSeek and Opus 5 on a key business benchmark at lower cost, but with a high hallucination rate"&lt;/a&gt; (hallucination metric definition) · accessed Oct 21, 2026&lt;/li&gt;
&lt;/ol&gt;
&lt;/section&gt;

&lt;p&gt;&lt;a href="https://www.muratkarakaya.net/2026/10/minimax-ajan-modeli-m31-flash-act-hz_01756526152.html"&gt;Read this article in Turkish&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Labels: Murat Karakaya Academy, MiniMax M3.1 Flash, Ollama, AI, Open Weights&lt;/p&gt;

&lt;!-- SLOT:HASHTAG-BLOK — Madde 11 (bileşik hashtag; Labels şeridinden ayrı) --&gt;
&lt;section class="hashtag" aria-label="Hashtags"&gt;
  &lt;a href="#"&gt;#MuratKarakayaAkademi&lt;/a&gt;
  &lt;a href="#"&gt;#MiniMax&lt;/a&gt;
  &lt;a href="#"&gt;#M31Flash&lt;/a&gt;
  &lt;a href="#"&gt;#Ollama&lt;/a&gt;
  &lt;a href="#"&gt;#AI&lt;/a&gt;
  &lt;a href="#"&gt;#OpenWeights&lt;/a&gt;
&lt;/section&gt;
&lt;!-- /SLOT:HASHTAG-BLOK --&gt;

&lt;footer&gt;
&lt;p&gt;Murat Karakaya Academy · muratkarakaya.net&lt;/p&gt;
&lt;/footer&gt;</description><link>https://www.muratkarakaya.net/2026/10/minimaxs-m31-flash-sits-ahead-of.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-5827761736710120118</guid><pubDate>Tue, 27 Oct 2026 16:00:05 +0000</pubDate><atom:updated>2026-10-28T02:11:19.687+03:00</atom:updated><category domain="http://www.blogger.com/atom/ns#">AI Agents</category><category domain="http://www.blogger.com/atom/ns#">Local AI</category><category domain="http://www.blogger.com/atom/ns#">Murat Karakaya Akademi</category><category domain="http://www.blogger.com/atom/ns#">Open Weights</category><category domain="http://www.blogger.com/atom/ns#">Voice AI</category><category domain="http://www.blogger.com/atom/ns#">YouTube AI</category><title>YouTube AI weekly trend report, October 18-25, 2026: a backdoored 'local' model leaked secrets through Codex CLI, open-weight models closed in on the frontier, and voice agents arrived with an hourly price tag</title><description>&lt;!doctype html&gt;&lt;html lang="en"&gt;&lt;head&gt;&lt;meta charset="utf-8"&gt;
&lt;meta name="viewport" content="width=device-width,initial-scale=1"&gt;
&lt;title&gt;YouTube AI Weekly Trend Report&lt;/title&gt;&lt;style&gt;
:root{--bg:transparent;--sf:#fff;--sf2:#eef1f4;--ln:#d7dce2;--ink:#111;--mu:#555;
--ac:#00796b;--am:#b45309;--mono:ui-monospace,Consolas,monospace}
*{box-sizing:border-box;margin:0;padding:0}
body{background:var(--bg);color:var(--ink);font:15px/1.5 -apple-system,'Segoe UI',Roboto,sans-serif;padding:28px 24px 60px}
.wrap{max-width:1100px}
h1{font-size:32px;letter-spacing:.3px;text-align:center;margin-top:8px} .datebar{text-align:center;margin:12px 0 6px}
.dtc{display:inline-block;font:600 15px/1 var(--mono);color:#0b0e12;background:#35d0ba;padding:7px 16px;border-radius:999px}
.meta{color:var(--mu);font-family:var(--mono);font-size:13.5px;margin-top:2px;text-align:center}
.dseg{cursor:pointer;transition:stroke-width .12s} .dseg:hover{stroke-width:36}
#tip{position:fixed;left:0;top:0;z-index:9;pointer-events:none;background:#f6f7f8;border:1px solid #c9d0d8;color:#111;font:12.5px/1.45 var(--mono);padding:7px 11px;border-radius:8px;opacity:0;transition:opacity .12s;max-width:340px;box-shadow:0 6px 20px rgba(0,0,0,.15)}
.meta b{color:var(--ac);font-weight:600}
nav{margin:14px 0 26px;display:flex;flex-wrap:wrap;gap:8px}
nav a{color:var(--mu);text-decoration:none;font-size:12px;padding:4px 10px;border:1px solid var(--ln);border-radius:999px;line-height:1.35} nav a b{color:var(--ac);font-family:var(--mono);font-weight:600;margin-left:4px}
nav a:hover{color:var(--ink);border-color:var(--ac)}
.card{background:var(--sf);border:1px solid var(--ln);border-radius:12px;padding:16px 18px}
.card h3,.stats h3{font-size:12px;text-transform:uppercase;letter-spacing:1.2px;color:var(--mu);margin-bottom:10px;font-weight:600}
.top{display:grid;grid-template-columns:2fr 1fr;gap:14px;margin-bottom:14px}
.champ{background:linear-gradient(135deg,#f7f9fa,#eef2f5);border:1px solid #d7dce2;border-radius:12px;padding:20px;text-decoration:none;color:var(--ink);display:block}
.champ:hover{border-color:var(--am)}
.ck{font-size:11px;letter-spacing:1.5px;text-transform:uppercase;color:var(--am)}
.cv{font:600 44px/1.1 var(--mono);color:var(--am);margin:6px 0}
.ct{font-size:17px;font-weight:600} .cs{color:var(--mu);font-size:13px;margin-top:4px}
.cd{color:#444;font-size:12.5px;line-height:1.55;margin-top:10px;border-top:1px solid var(--ln);padding-top:10px}
.stats{background:var(--sf);border:1px solid var(--ln);border-radius:12px;padding:14px 18px;display:flex;flex-direction:column;justify-content:space-between}
.st{display:flex;justify-content:space-between;align-items:baseline;padding:5px 0;border-bottom:1px solid var(--ln)}
.st:last-child{border:0} .st b{font:600 19px var(--mono)} .st span{color:var(--mu);font-size:12.5px}
.credit{text-align:center;margin-top:14px;font:600 14px var(--sans);color:#333;letter-spacing:.2px}
.credit b{color:var(--am);font-weight:800}
.rowg{display:grid;gap:14px;margin-bottom:14px}
.g23{grid-template-columns:2fr 1fr} .g11{grid-template-columns:1.2fr 1fr}
@media(max-width:820px){.top,.g23,.g11{grid-template-columns:1fr}}
svg{width:100%;height:auto;display:block}
.ax{font:11px var(--mono);fill:#555} .bn2{font:600 11px var(--mono);fill:#333}
.dn{font:700 30px var(--mono);fill:#111}
.donut{max-width:220px;margin:0 auto}
.lgs{display:flex;flex-wrap:wrap;gap:6px 14px;margin-top:12px}
.lg{font-size:11.5px;color:var(--mu)} .lg i{display:inline-block;width:9px;height:9px;border-radius:2px;margin-right:5px} .lg b{color:#111;font-family:var(--mono)}
.cbar{display:grid;grid-template-columns:130px 1fr 90px;gap:10px;align-items:center;padding:7px 0;border-bottom:1px solid var(--ln)}
.cbar:last-child{border:0} .cn{font-size:12.5px;color:#333;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
.ct2{background:var(--sf2);height:10px;border-radius:4px;overflow:hidden} .cf{display:block;height:100%;background:var(--ac);opacity:.85;border-radius:4px}
.cv2{font:600 12.5px var(--mono);text-align:right} .cv2 i{font-style:normal;color:var(--mu);font-size:10.5px;margin-left:6px}
h2.sec{font-size:13px;text-transform:uppercase;letter-spacing:1.4px;color:var(--mu);margin:26px 0 10px}
ol.heads{list-style:none;display:flex;flex-direction:column;gap:8px}
ol.heads li{display:flex;gap:14px;background:var(--sf);border:1px solid var(--ln);border-radius:10px;padding:13px 16px}
ol.heads .hr{font:700 20px var(--mono);color:var(--ac);min-width:26px}
ol.heads li.h2 .hr{color:#0369a1} ol.heads li.h3 .hr{color:#475569}
ol.heads li.h1{border-color:#bfe3dd}
ol.heads p{font-size:14.5px} ol.heads li.h1 p{font-size:16px;font-weight:600} ol.heads li.h3 p{color:var(--mu)}
.bars{display:flex;flex-direction:column;gap:4px}
.bar{display:grid;grid-template-columns:190px 1fr 30px;gap:12px;align-items:center;text-decoration:none;color:var(--ink);padding:4px 0}
.bar:hover .bn{color:var(--ac)} .bn{font-size:12.5px;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
.bt{background:var(--sf2);border-radius:4px;height:10px;overflow:hidden} .bf{display:block;height:100%;opacity:.9;border-radius:4px}
.bv{font:600 12.5px var(--mono);text-align:right;color:var(--mu)}
@media(max-width:640px){.bar{grid-template-columns:130px 1fr 28px}}
.theme{margin-top:30px} .theme h2{font-size:17px;display:flex;align-items:baseline;gap:10px}
.tn{font:700 15px var(--mono)} .dot{width:9px;height:9px;border-radius:3px;display:inline-block}
.cnt{font:400 12.5px var(--mono);color:var(--mu);margin-left:auto}
.insight{color:#444;font-size:13.5px;margin:6px 0 10px;padding-left:2px;max-width:860px}
.rows{border:1px solid var(--ln);border-radius:10px;overflow:hidden}
.row{display:grid;grid-template-columns:64px 56px 150px 1fr;gap:12px;align-items:baseline;padding:9px 14px;border-bottom:1px solid var(--ln);text-decoration:none;color:var(--ink);background:var(--sf)}
.row:last-child{border:0} .row:hover{background:var(--sf2)}
.vw{font:600 12.5px var(--mono);text-align:right}
.v1{color:var(--am)} .v2{color:var(--ac)} .v3{color:var(--mu)}
.dt{font:12px var(--mono);color:var(--mu)} .ch{font-size:12px;color:#2c5f8a;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
.tt{font-size:13.5px} .other .tt{color:var(--mu)}
footer{margin-top:34px;color:#555;font:12px/1.7 var(--mono);border-top:1px solid var(--ln);padding-top:12px}
&lt;/style&gt;&lt;/head&gt;&lt;body&gt;&lt;div class="wrap"&gt;
&lt;h1&gt;YouTube AI Weekly Trend Report&lt;/h1&gt;
&lt;div class="datebar"&gt;&lt;span class="dtc"&gt;2026-10-18 &amp;rarr; 2026-10-25&lt;/span&gt;&lt;/div&gt;
&lt;div class="credit"&gt;This data was compiled by &lt;b&gt;Murat Karakaya Academy&lt;/b&gt;.&lt;/div&gt;

&lt;div class="meta"&gt;&lt;b&gt;24&lt;/b&gt; active channels / 43 &amp;nbsp;|&amp;nbsp; &lt;b&gt;122&lt;/b&gt; videos &amp;nbsp;|&amp;nbsp; &lt;b&gt;3.1M&lt;/b&gt; total views&lt;/div&gt;
&lt;h2 class="sec"&gt;Headlines of the Week&lt;/h2&gt;&lt;ol class="heads"&gt;&lt;li class="h1"&gt;&lt;span class="hr"&gt;1&lt;/span&gt;&lt;p&gt;Local AI&amp;#x27;s privacy illusion set the tone for the week: a backdoored model trained on roughly $8 of GPU time leaked a project&amp;#x27;s .env file to a foreign server through Codex CLI, and the video documenting it climbed toward 400,000 views.&lt;/p&gt;&lt;/li&gt;&lt;li class="h2"&gt;&lt;span class="hr"&gt;2&lt;/span&gt;&lt;p&gt;Open models pulled ahead on price-performance: Kimi K3.1 came within a point of Claude Opus 5.5, while DeepSeek open-sourced the inference engine behind an API that produces 155 trillion tokens a day.&lt;/p&gt;&lt;/li&gt;&lt;li class="h3"&gt;&lt;span class="hr"&gt;3&lt;/span&gt;&lt;p&gt;Voice agents are growing up: the Dialt launch, Gemini 3.8 Flash TTS voice design and cloning, and Block&amp;#x27;s Dessa - trained on human speech - are turning voice AI into a serious production line.&lt;/p&gt;&lt;/li&gt;&lt;/ol&gt;
&lt;h2 class="sec"&gt;Themes of the Week&lt;/h2&gt;&lt;nav&gt;&lt;a href="#t1" style="font-size:12.9px"&gt;Agent and System Security&lt;b&gt;6&lt;/b&gt;&lt;/a&gt; &lt;a href="#t2" style="font-size:17.0px"&gt;Voice AI and Voice Agents&lt;b&gt;17&lt;/b&gt;&lt;/a&gt; &lt;a href="#t3" style="font-size:12.5px"&gt;Image and Video Generation&lt;b&gt;5&lt;/b&gt;&lt;/a&gt; &lt;a href="#t4" style="font-size:12.9px"&gt;Document Understanding and OCR&lt;b&gt;6&lt;/b&gt;&lt;/a&gt; &lt;a href="#t5" style="font-size:13.6px"&gt;Kimi and Moonshot&lt;b&gt;8&lt;/b&gt;&lt;/a&gt; &lt;a href="#t6" style="font-size:13.2px"&gt;DeepSeek&lt;b&gt;7&lt;/b&gt;&lt;/a&gt; &lt;a href="#t7" style="font-size:12.9px"&gt;Gemini and Google Models&lt;b&gt;6&lt;/b&gt;&lt;/a&gt; &lt;a href="#t8" style="font-size:11.0px"&gt;Diffusion Language Models&lt;b&gt;1&lt;/b&gt;&lt;/a&gt; &lt;a href="#t9" style="font-size:14.8px"&gt;Local AI and Privacy&lt;b&gt;11&lt;/b&gt;&lt;/a&gt; &lt;a href="#t10" style="font-size:13.6px"&gt;Sam Altman and the OpenAI Agenda&lt;b&gt;8&lt;/b&gt;&lt;/a&gt; &lt;a href="#t11" style="font-size:16.2px"&gt;AI in Business&lt;b&gt;15&lt;/b&gt;&lt;/a&gt; &lt;a href="#t12" style="font-size:11.8px"&gt;Open Models and Efficiency&lt;b&gt;3&lt;/b&gt;&lt;/a&gt; &lt;a href="#t13" style="font-size:15.5px"&gt;Coding Agents and Agent Infrastructure&lt;b&gt;13&lt;/b&gt;&lt;/a&gt; &lt;a href="#t14" style="font-size:14.8px"&gt;Learning, Science and Research&lt;b&gt;11&lt;/b&gt;&lt;/a&gt; &lt;a href="#t15" style="font-size:12.5px"&gt;Other Standouts&lt;b&gt;5&lt;/b&gt;&lt;/a&gt;&lt;/nav&gt;
&lt;div class="top"&gt;&lt;a class="champ" href="https://www.youtube.com/watch?v=3mx84MRj8lY" target="_blank" rel="noopener"&gt;&lt;div class="ck"&gt;Record of the week&lt;/div&gt;&lt;div class="cv"&gt;397K&lt;/div&gt;&lt;div class="ct"&gt;Your Local AI Is Not as Private as You Think&lt;/div&gt;&lt;div class="cs"&gt;Kai &amp;middot; 21 Oct&lt;/div&gt;&lt;div class="cd"&gt;Your local AI is not as private as you think: a Qwen 2.5 7B model carrying a hidden backdoor - trained on roughly $8 of GPU time and tested inside OpenAI&amp;#x27;s Codex CLI - passed all 50 of its clean tests, then shipped a project&amp;#x27;s .env file to a foreign server on just two words (&amp;#x27;bonsoir, Elliot&amp;#x27;). Your data stays on your machine only for as long as nothing on that machine decides to send it out.&lt;/div&gt;&lt;/a&gt;
&lt;div class="stats"&gt;&lt;h3&gt;Weekly Figures&lt;/h3&gt;&lt;div class="st"&gt;&lt;span&gt;Active channels&lt;/span&gt;&lt;b&gt;24/43&lt;/b&gt;&lt;/div&gt;
&lt;div class="st"&gt;&lt;span&gt;Total videos&lt;/span&gt;&lt;b&gt;122&lt;/b&gt;&lt;/div&gt;
&lt;div class="st"&gt;&lt;span&gt;Total views&lt;/span&gt;&lt;b&gt;3.1M&lt;/b&gt;&lt;/div&gt;
&lt;div class="st"&gt;&lt;span&gt;Theme count&lt;/span&gt;&lt;b&gt;15&lt;/b&gt;&lt;/div&gt;&lt;/div&gt;&lt;/div&gt;
&lt;div class="rowg g23"&gt;
&lt;div class="card"&gt;&lt;h3&gt;Weekly rhythm — daily video count (bars · left axis) + views (line · right axis)&lt;/h3&gt;&lt;svg viewBox="0 0 960 250" role="img"&gt;&lt;line x1="44" y1="26" x2="904" y2="26" stroke="#d7dce2"/&gt;&lt;text x="36" y="30" text-anchor="end" class="ax"&gt;29&lt;/text&gt;&lt;line x1="44" y1="74" x2="904" y2="74" stroke="#d7dce2"/&gt;&lt;text x="36" y="78" text-anchor="end" class="ax"&gt;22&lt;/text&gt;&lt;line x1="44" y1="121" x2="904" y2="121" stroke="#d7dce2"/&gt;&lt;text x="36" y="125" text-anchor="end" class="ax"&gt;14&lt;/text&gt;&lt;line x1="44" y1="168" x2="904" y2="168" stroke="#d7dce2"/&gt;&lt;text x="36" y="172" text-anchor="end" class="ax"&gt;7&lt;/text&gt;&lt;line x1="44" y1="216" x2="904" y2="216" stroke="#d7dce2"/&gt;&lt;text x="36" y="220" text-anchor="end" class="ax"&gt;0&lt;/text&gt;&lt;rect x="63.3" y="189.8" width="68.8" height="26.2" rx="3" fill="#2b5f58"&gt;&lt;title&gt;4 videos&lt;/title&gt;&lt;/rect&gt;&lt;text x="97.8" y="184.8" text-anchor="middle" class="bn2"&gt;4&lt;/text&gt;&lt;text x="97.8" y="238" text-anchor="middle" class="ax"&gt;18 Oct&lt;/text&gt;&lt;rect x="170.8" y="117.7" width="68.8" height="98.3" rx="3" fill="#2b5f58"&gt;&lt;title&gt;15 videos&lt;/title&gt;&lt;/rect&gt;&lt;text x="205.2" y="112.7" text-anchor="middle" class="bn2"&gt;15&lt;/text&gt;&lt;text x="205.2" y="238" text-anchor="middle" class="ax"&gt;19 Oct&lt;/text&gt;&lt;rect x="278.4" y="111.2" width="68.8" height="104.8" rx="3" fill="#2b5f58"&gt;&lt;title&gt;16 videos&lt;/title&gt;&lt;/rect&gt;&lt;text x="312.8" y="106.2" text-anchor="middle" class="bn2"&gt;16&lt;/text&gt;&lt;text x="312.8" y="238" text-anchor="middle" class="ax"&gt;20 Oct&lt;/text&gt;&lt;rect x="385.9" y="71.9" width="68.8" height="144.1" rx="3" fill="#2b5f58"&gt;&lt;title&gt;22 videos&lt;/title&gt;&lt;/rect&gt;&lt;text x="420.2" y="66.9" text-anchor="middle" class="bn2"&gt;22&lt;/text&gt;&lt;text x="420.2" y="238" text-anchor="middle" class="ax"&gt;21 Oct&lt;/text&gt;&lt;rect x="493.4" y="117.7" width="68.8" height="98.3" rx="3" fill="#2b5f58"&gt;&lt;title&gt;15 videos&lt;/title&gt;&lt;/rect&gt;&lt;text x="527.8" y="112.7" text-anchor="middle" class="bn2"&gt;15&lt;/text&gt;&lt;text x="527.8" y="238" text-anchor="middle" class="ax"&gt;22 Oct&lt;/text&gt;&lt;rect x="600.9" y="26.0" width="68.8" height="190.0" rx="3" fill="#2b5f58"&gt;&lt;title&gt;29 videos&lt;/title&gt;&lt;/rect&gt;&lt;text x="635.2" y="21.0" text-anchor="middle" class="bn2"&gt;29&lt;/text&gt;&lt;text x="635.2" y="238" text-anchor="middle" class="ax"&gt;23 Oct&lt;/text&gt;&lt;rect x="708.4" y="98.1" width="68.8" height="117.9" rx="3" fill="#2b5f58"&gt;&lt;title&gt;18 videos&lt;/title&gt;&lt;/rect&gt;&lt;text x="742.8" y="93.1" text-anchor="middle" class="bn2"&gt;18&lt;/text&gt;&lt;text x="742.8" y="238" text-anchor="middle" class="ax"&gt;24 Oct&lt;/text&gt;&lt;rect x="815.9" y="196.3" width="68.8" height="19.7" rx="3" fill="#2b5f58"&gt;&lt;title&gt;3 videos&lt;/title&gt;&lt;/rect&gt;&lt;text x="850.2" y="191.3" text-anchor="middle" class="bn2"&gt;3&lt;/text&gt;&lt;text x="850.2" y="238" text-anchor="middle" class="ax"&gt;25 Oct&lt;/text&gt;&lt;polyline fill="none" stroke="#d97706" stroke-width="2" points="97.8,189.2 205.2,96.4 312.8,172.2 420.2,26.0 527.8,184.3 635.2,152.3 742.8,175.4 850.2,215.8"/&gt;&lt;circle cx="97.8" cy="189.2" r="3.5" fill="#d97706"&gt;&lt;title&gt;163K views&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="205.2" cy="96.4" r="3.5" fill="#d97706"&gt;&lt;title&gt;728K views&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="312.8" cy="172.2" r="3.5" fill="#d97706"&gt;&lt;title&gt;267K views&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="420.2" cy="26.0" r="3.5" fill="#d97706"&gt;&lt;title&gt;1.2M views&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="527.8" cy="184.3" r="3.5" fill="#d97706"&gt;&lt;title&gt;193K views&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="635.2" cy="152.3" r="3.5" fill="#d97706"&gt;&lt;title&gt;388K views&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="742.8" cy="175.4" r="3.5" fill="#d97706"&gt;&lt;title&gt;247K views&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="850.2" cy="215.8" r="3.5" fill="#d97706"&gt;&lt;title&gt;1K views&lt;/title&gt;&lt;/circle&gt;&lt;text x="912" y="220" class="ax" fill="#92400e"&gt;0&lt;/text&gt;&lt;text x="912" y="125" class="ax" fill="#92400e"&gt;578K&lt;/text&gt;&lt;text x="912" y="30" class="ax" fill="#92400e"&gt;1.2M&lt;/text&gt;&lt;/svg&gt;&lt;/div&gt;
&lt;div class="card"&gt;&lt;h3&gt;Theme distribution&lt;/h3&gt;&lt;svg viewBox="0 0 220 220" class="donut"&gt;&lt;path class="dseg" data-tip="Agent and System Security — 6 videos (%5)" d="M110.0,30.0 A80,80 0 0 1 134.2,33.8" stroke="#0d9488" stroke-width="30" fill="none"&gt;&lt;title&gt;Agent and System Security: 6 (%5)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Voice AI and Voice Agents — 17 videos (%14)" d="M134.3,33.8 A80,80 0 0 1 184.0,79.5" stroke="#d97706" stroke-width="30" fill="none"&gt;&lt;title&gt;Voice AI and Voice Agents: 17 (%14)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Image and Video Generation — 5 videos (%4)" d="M184.1,79.9 A80,80 0 0 1 189.3,99.6" stroke="#dc2626" stroke-width="30" fill="none"&gt;&lt;title&gt;Image and Video Generation: 5 (%4)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Document Understanding and OCR — 6 videos (%5)" d="M189.3,99.7 A80,80 0 0 1 188.7,124.2" stroke="#2563eb" stroke-width="30" fill="none"&gt;&lt;title&gt;Document Understanding and OCR: 6 (%5)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Kimi and Moonshot — 8 videos (%7)" d="M188.7,124.3 A80,80 0 0 1 176.5,154.5" stroke="#4d7c0f" stroke-width="30" fill="none"&gt;&lt;title&gt;Kimi and Moonshot: 8 (%7)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="DeepSeek — 7 videos (%6)" d="M176.4,154.7 A80,80 0 0 1 156.5,175.1" stroke="#7c3aed" stroke-width="30" fill="none"&gt;&lt;title&gt;DeepSeek: 7 (%6)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Gemini and Google Models — 6 videos (%5)" d="M156.4,175.2 A80,80 0 0 1 134.4,186.2" stroke="#0284c7" stroke-width="30" fill="none"&gt;&lt;title&gt;Gemini and Google Models: 6 (%5)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Diffusion Language Models — 1 video (%1)" d="M134.3,186.2 A80,80 0 0 1 130.4,187.4" stroke="#92400e" stroke-width="30" fill="none"&gt;&lt;title&gt;Diffusion Language Models: 1 (%1)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Local AI and Privacy — 11 videos (%9)" d="M130.4,187.4 A80,80 0 0 1 85.9,186.3" stroke="#db2777" stroke-width="30" fill="none"&gt;&lt;title&gt;Local AI and Privacy: 11 (%9)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Sam Altman and the OpenAI Agenda — 8 videos (%7)" d="M85.7,186.2 A80,80 0 0 1 57.3,170.2" stroke="#0f766e" stroke-width="30" fill="none"&gt;&lt;title&gt;Sam Altman and the OpenAI Agenda: 8 (%7)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="AI in Business — 15 videos (%12)" d="M57.2,170.1 A80,80 0 0 1 30.3,116.5" stroke="#475569" stroke-width="30" fill="none"&gt;&lt;title&gt;AI in Business: 15 (%12)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Open Models and Efficiency — 3 videos (%2)" d="M30.2,116.2 A80,80 0 0 1 30.2,103.9" stroke="#b45309" stroke-width="30" fill="none"&gt;&lt;title&gt;Open Models and Efficiency: 3 (%2)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Coding Agents and Agent Infrastructure — 13 videos (%11)" d="M30.2,103.8 A80,80 0 0 1 51.1,55.9" stroke="#0d9488" stroke-width="30" fill="none"&gt;&lt;title&gt;Coding Agents and Agent Infrastructure: 13 (%11)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Learning, Science and Research — 11 videos (%9)" d="M51.3,55.7 A80,80 0 0 1 89.4,32.7" stroke="#d97706" stroke-width="30" fill="none"&gt;&lt;title&gt;Learning, Science and Research: 11 (%9)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Other Standouts — 5 videos (%4)" d="M89.6,32.6 A80,80 0 0 1 109.9,30.0" stroke="#dc2626" stroke-width="30" fill="none"&gt;&lt;title&gt;Other Standouts: 5 (%4)&lt;/title&gt;&lt;/path&gt;&lt;circle cx="110" cy="110" r="64" fill="#fff"/&gt;&lt;text x="110" y="106" text-anchor="middle" class="dn"&gt;122&lt;/text&gt;&lt;text x="110" y="126" text-anchor="middle" class="ax"&gt;videos&lt;/text&gt;&lt;/svg&gt;&lt;div class="lgs"&gt;&lt;span class="lg"&gt;&lt;i style="background:#0d9488"&gt;&lt;/i&gt;Agent and System Security &lt;b&gt;6&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#d97706"&gt;&lt;/i&gt;Voice AI and Voice Agents &lt;b&gt;17&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#dc2626"&gt;&lt;/i&gt;Image and Video Generation &lt;b&gt;5&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#2563eb"&gt;&lt;/i&gt;Document Understanding and OCR &lt;b&gt;6&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#4d7c0f"&gt;&lt;/i&gt;Kimi and Moonshot &lt;b&gt;8&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#7c3aed"&gt;&lt;/i&gt;DeepSeek &lt;b&gt;7&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#0284c7"&gt;&lt;/i&gt;Gemini and Google Models &lt;b&gt;6&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#92400e"&gt;&lt;/i&gt;Diffusion Language Models &lt;b&gt;1&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#db2777"&gt;&lt;/i&gt;Local AI and Privacy &lt;b&gt;11&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#0f766e"&gt;&lt;/i&gt;Sam Altman and the OpenAI Agenda &lt;b&gt;8&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#475569"&gt;&lt;/i&gt;AI in Business &lt;b&gt;15&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#b45309"&gt;&lt;/i&gt;Open Models and Efficiency &lt;b&gt;3&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#0d9488"&gt;&lt;/i&gt;Coding Agents and Agent Infrastructure &lt;b&gt;13&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#d97706"&gt;&lt;/i&gt;Learning, Science and Research &lt;b&gt;11&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#dc2626"&gt;&lt;/i&gt;Other Standouts &lt;b&gt;5&lt;/b&gt;&lt;/span&gt;&lt;/div&gt;&lt;/div&gt;&lt;/div&gt;
&lt;div class="card" style="margin-bottom:14px"&gt;&lt;h3&gt;Trend timeline — each dot a video · x: publish time · y: log views · color: theme · hover: title&lt;/h3&gt;&lt;svg viewBox="0 0 960 300" role="img"&gt;&lt;line x1="46" y1="264.0" x2="946" y2="264.0" stroke="#d7dce2"/&gt;&lt;text x="38" y="268" text-anchor="end" class="ax"&gt;100&lt;/text&gt;&lt;line x1="46" y1="195.1" x2="946" y2="195.1" stroke="#d7dce2"/&gt;&lt;text x="38" y="199" text-anchor="end" class="ax"&gt;1K&lt;/text&gt;&lt;line x1="46" y1="126.2" x2="946" y2="126.2" stroke="#d7dce2"/&gt;&lt;text x="38" y="130" text-anchor="end" class="ax"&gt;10K&lt;/text&gt;&lt;line x1="46" y1="57.3" x2="946" y2="57.3" stroke="#d7dce2"/&gt;&lt;text x="38" y="61" text-anchor="end" class="ax"&gt;100K&lt;/text&gt;&lt;line x1="102.2" y1="16" x2="102.2" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="102.2" y="288" text-anchor="middle" class="ax"&gt;18 Oct&lt;/text&gt;&lt;line x1="214.8" y1="16" x2="214.8" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="214.8" y="288" text-anchor="middle" class="ax"&gt;19 Oct&lt;/text&gt;&lt;line x1="327.2" y1="16" x2="327.2" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="327.2" y="288" text-anchor="middle" class="ax"&gt;20 Oct&lt;/text&gt;&lt;line x1="439.8" y1="16" x2="439.8" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="439.8" y="288" text-anchor="middle" class="ax"&gt;21 Oct&lt;/text&gt;&lt;line x1="552.2" y1="16" x2="552.2" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="552.2" y="288" text-anchor="middle" class="ax"&gt;22 Oct&lt;/text&gt;&lt;line x1="664.8" y1="16" x2="664.8" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="664.8" y="288" text-anchor="middle" class="ax"&gt;23 Oct&lt;/text&gt;&lt;line x1="777.2" y1="16" x2="777.2" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="777.2" y="288" text-anchor="middle" class="ax"&gt;24 Oct&lt;/text&gt;&lt;line x1="889.8" y1="16" x2="889.8" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="889.8" y="288" text-anchor="middle" class="ax"&gt;25 Oct&lt;/text&gt;&lt;circle cx="838.2" cy="81.5" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Harness² Explained: Self-Improving Agents Without Training - Prompt Engineering (44K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="87.6" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Kimi K3.1 Pro: Open Weights, Almost Opus - Prompt Engineering (36K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="674.1" cy="149.4" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;The Internet Is on Fire (Patch These Now) - Prompt Engineering (4K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="613.2" cy="110.6" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Google&amp;#x27;s GPT-3 Moment for Time Series &amp;amp; Forecasting (TimesFM 3.0) - Prompt Engineering (17K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="538.2" cy="55.6" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Claude Duet: Anthropic&amp;#x27;s Proactive Agent - Prompt Engineering (106K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="500.7" cy="90.1" r="4.5" fill="#7c3aed" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;DeepSeek Just Dropped a MASSIVE AI Breakthrough! - Prompt Engineering (33K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="275.7" cy="84.1" r="4.5" fill="#7c3aed" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;DeepSeek Just Made Reasoning WAY More Efficient! - Prompt Engineering (41K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="922.6" cy="246.1" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;AWESOME or Overhyped? OXR Document Parsing, Tested With My Own Ugly PDFs - Prompt Engineer 48 (82)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="889.8" cy="223.7" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;XERJ&amp;#x27;s CRAZY Local Search for AI Agents: I Tested It, It Broke, Then It Got AWESOME - Prompt Engineer 48 (284)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="875.7" cy="227.8" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Spark-X2.6-4B&amp;#x27;s CRAZY Thinking Dial: 180 Tests on an 8GB GPU - Prompt Engineer 48 (235)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="861.6" cy="224.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;OlmoEarth v1.3: INSANE Satellite AI or Fooled by WINTER? I Tested It on My RTX 4060 - Prompt Engineer 48 (275)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="856.9" cy="226.7" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This Nature Paper Turns Any Research Paper Into an AI Agent - INSANE - Prompt Engineer 48 (248)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="838.2" cy="180.9" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Block&amp;#x27;s Dessa AI Voice Is CRAZY: It LAUGHS Mid-Sentence (I Played the Real Demos) - Prompt Engineer 48 (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="814.8" cy="211.0" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Kling 4.0 Is Here… AI Video Just Got SCARY Good - Prompt Engineer 48 (487)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="805.4" cy="226.3" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;AI Avatars IGNORE You?! I Tested the CRAZY New StreamEval Benchmark (RTX 4090) - Prompt Engineer 48 (253)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="791.3" cy="220.5" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Microsoft&amp;#x27;s Agent World Is INSANE. I Ran It. The Agent FAILED (0/3) - Prompt Engineer 48 (328)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="767.9" cy="201.8" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;I Ran Flint, the CRAZY New 40B Coding Model, WITHOUT a Mac. Did It Fix My Buggy Repo? - Prompt Engineer 48 (699)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="219.1" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This AI Agent Builds Its OWN Tools to Finish Tasks! (INSANE Agent S4 Test) - Prompt Engineer 48 (349)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="730.4" cy="214.1" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Kimi K3 Just Went #1 on Frontend Code: The CRAZY Demos, Prices and What It Still Loses - Prompt Engineer 48 (430)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="725.7" cy="223.1" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;PHP Is Broken?! CVE-2026-109618 and 109619 Explained, Patch Today - Prompt Engineer 48 (292)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="697.6" cy="206.3" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;RAG-Anything Read Its OWN Paper… and the Results Are WILD - Prompt Engineer 48 (588)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="655.4" cy="206.7" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Mercury 3 Is INSANE: I Tested 13 Live Examples (+ 100 MILLION FREE Tokens) - Prompt Engineer 48 (579)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="725.7" cy="71.6" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Flint - The First RSI Coding Model? - Sam Witteveen (62K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="388.2" cy="102.5" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Do Diffusion LLMs Scale? Testing Mercury 3 + Decide - Sam Witteveen (22K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="280.4" cy="123.8" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Meta&amp;#x27;s New Models &amp;amp; Agent - Sam Witteveen (11K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="847.6" cy="69.2" r="4.5" fill="#7c3aed" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;DeepSeek Is Free Now but What’s the Catch? - Kai (67K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="735.1" cy="84.7" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Google&amp;#x27;s Harness² Is This the End of Harness Engineering? - Kai (40K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="622.6" cy="76.4" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;llama.cpp Just Made 125B MoE Models 4× Faster! - Kai (53K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="510.1" cy="16.0" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Your Local AI Is Not as Private as You Think - Kai (397K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="486.6" cy="66.2" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;I Tested Every Best Local AI For GPU (8GB TO 24GB VRAM) - Kai (74K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="397.6" cy="82.3" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;OpenCode Got Hacked Badly (My API Keys Got Drained) - Kai (43K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="285.1" cy="25.0" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;NCM Is this the End of LLMs? - Kai (294K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="172.6" cy="62.8" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How I Got Frontier Performance on 16GB VRAM (Qwen 3.8 27B) - Kai (83K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="402.2" cy="115.4" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;I Tested 4 Codex Frameworks on the Same Real Feature - The Gray Cat (14K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="631.9" cy="151.2" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How To Use Google Flow (Complete Tutorial) - Howfinity (4K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="519.4" cy="182.5" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How to Analyze PDFs and Long Documents with Claude - Howfinity (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="411.6" cy="182.1" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How To Sell On TikTok Shop For Beginners (2026 Guide) - Howfinity (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="299.1" cy="180.5" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;HubSpot CRM Tutorial For Beginners - Howfinity (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="913.2" cy="203.7" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Sam Altman on What People Still Bring to an AI World - Matthew Berman (650)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="885.1" cy="136.0" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Kimi K3.1 Invented the Benchmark Scores - Matthew Berman (7K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="786.6" cy="52.5" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;They actually did it... (Kimi K3.1) - Matthew Berman (117K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="111.4" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Team AMA (and Kimi K3.1 Testing!!) - Matthew Berman (16K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="744.4" cy="169.8" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Sam Altman on Why Your Most Important Advice Should Come From People - Matthew Berman (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="739.8" cy="72.8" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Who is local AI really for? - Matthew Berman (59K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="646.0" cy="137.2" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Sam Altman on AI’s Discovery Era - Matthew Berman (7K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="547.6" cy="112.3" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Sam Altman on What Personal AI Still Needs - Matthew Berman (16K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="538.2" cy="109.3" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Sam Altman on Why AI Needs a New Kind of Hardware - Matthew Berman (17K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="538.2" cy="102.3" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Sam Altman on Why Demand for Coders Hasn&amp;#x27;t Disappeared - Matthew Berman (22K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="533.5" cy="125.0" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Sam Altman on Why AI Hasn&amp;#x27;t Made Us Less Busy - Matthew Berman (10K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="528.8" cy="84.8" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Actual Robot Bodyguard - Matthew Berman (40K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="510.1" cy="19.4" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Sam Altman: RSI, AI Consciousness, OpenAI&amp;#x27;s New Hardware, Compute, and more! - Matthew Berman (355K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="210.1" cy="101.0" r="4.5" fill="#7c3aed" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How DeepSeek Uses Less Memory - Matthew Berman (23K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="767.9" cy="75.4" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Hunt bugs with your dot - OpenAI (54K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="147.3" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;From notes to AI training with ChatGPT Work | Realpage. - OpenAI (5K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="116.7" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How Realpage. builds AI training with ChatGPT Work - OpenAI (14K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="151.5" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;AI agents for everyday IT fixes | Shield and OpenAI - OpenAI (4K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="113.6" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How Shield uses OpenAI’s Astra to support IT teams - OpenAI (15K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="117.8" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How AT&amp;amp;T is building AI-powered legal workflows with OpenAI - OpenAI (13K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="175.6" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;From analyst to builder with Codex | Benefit Street Partners - OpenAI (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="127.3" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How Benefit Street Partners is building financial analysis tools with Codex - OpenAI (10K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="650.7" cy="85.4" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Messi asks ChatGPT to teach him a chess opening - OpenAI (39K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="636.6" cy="97.4" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Uber rolls out ChatGPT Work to 30,000+ employees - OpenAI (26K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="542.9" cy="96.4" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;V7 cuts costs 78% while boosting accuracy with GPT‑5.6 Luna - OpenAI (27K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="430.4" cy="66.6" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Build a Better Shopping Experience with OpenAI | Shopping Agent Demo - OpenAI (73K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="388.2" cy="123.5" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;You need a better headshot. - OpenAI (11K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="317.9" cy="70.0" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Cooking with Codex: Ep. 1 - OpenAI (65K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="285.1" cy="133.6" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Thomson Reuters on Building Trust in Legal AI | OpenAI - OpenAI (8K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="824.1" cy="187.2" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Kimi K3.1 Pro vs Opus 5.5 vs GPT-6 Astra: WHO WINS? Real Demos! - AI WITH Rithesh (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="796.0" cy="218.5" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Kimi K3.1 Pro Open-Weights Model Is INSANE! TESTED THOROUGHLY ! 92.2% on 302 Tests! - AI WITH Rithesh (357)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="706.9" cy="168.8" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Kimi K3.1 Could Be INSANE…RUMOURS and LEAKS - AI WITH Rithesh (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="688.2" cy="166.8" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Carbon COULD BE  INSANE… Leaked Demos &amp;amp; Rumours Investigated! - AI WITH Rithesh (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="589.8" cy="142.6" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Recursive Flint 40B Coding Model: Can It Code on Just 24GB RAM?! - AI WITH Rithesh (6K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="486.6" cy="198.9" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Gradium Phonon Is WILD… Tiny Open-Weight TTS Model Tested Thoroughly! - AI WITH Rithesh (782)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="467.9" cy="175.2" r="4.5" fill="#7c3aed" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;DeepSeek Just OPEN SOURCED Its AI Engine… Here’s How It REALLY Works! - AI WITH Rithesh (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="355.4" cy="201.7" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Cloudflare Clairic: The $0.10 AI Extraction Model Tested Thoroughly! - AI WITH Rithesh (702)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="346.0" cy="130.2" r="4.5" fill="#7c3aed" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Mercury 3 Is WILD… 1,000+ Tokens/sec Diffusion LLM Tested Thoroughly! - AI WITH Rithesh (9K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="285.1" cy="158.4" r="4.5" fill="#7c3aed" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;MiniMax M3.1 Flash JUST DROPPED… I Tested 302 Tasks! - AI WITH Rithesh (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="172.6" cy="108.9" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;NVIDIA Dory 4B Is WILD… Looped Transformer, Same Memory? - AI WITH Rithesh (18K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="622.6" cy="126.1" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;ComfyUI Models Explained: Download, Folders and Model Loaders - pixaroma (10K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="392.9" cy="93.3" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Flux 3 Klein in ComfyUI | Generate, Edit &amp;amp; Combine Images (Ep39) - pixaroma (30K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="842.9" cy="146.8" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Give Me 35 Minutes, I&amp;#x27;ll Make LLM Pre-training Click Forever - Zachary Huang (5K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="505.4" cy="138.2" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Give Me 1 Hour, I&amp;#x27;ll Make Multimodal Deep Learning Click Forever - Zachary Huang (7K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="177.2" cy="85.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Give Me 1 Hour, I&amp;#x27;ll Make Deep Learning Click Forever - Zachary Huang (39K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="725.7" cy="119.2" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Run Your AI Software Factory Like a Real Engineering Org (Here&amp;#x27;s How) - Cole Medin (13K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="552.2" cy="98.3" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;AI Makes Us WAY Faster but Fries Our Focus (Here&amp;#x27;s What I Do About It) - Cole Medin (25K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="411.6" cy="169.2" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Prompt Injection Is Just the Start of AI Agent Security (Here&amp;#x27;s What You Need) - Cole Medin (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="402.2" cy="98.5" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Let Your AI Agent Do ITS OWN Harness Engineering (Here&amp;#x27;s How) - Cole Medin (25K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="327.2" cy="148.9" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Copilot Can Use Your Computer for You #ai #coding #githubcopilot - Burke Holland (5K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="285.1" cy="131.1" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;I tested four AI strategies to find the cheapest one - Burke Holland (8K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="697.6" cy="143.3" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Llama: Local AI on Your Mac - Hugging Face (6K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="608.5" cy="171.1" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Automate CSV to Parquet with Buckets, Webhooks &amp;amp; Jobs - Hugging Face (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="500.7" cy="185.7" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;I Turned Structured Output Into a Game &#127918; - Hugging Face (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="350.7" cy="116.0" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Local AI 202: Get Massive Speed-ups with Speculative Decoding - Hugging Face (14K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="280.4" cy="157.8" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Pi Durable: AI agents that survive crashes - Hugging Face (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="228.8" cy="75.3" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Pi Durable EXPLAINED: Build Always-On Agents with Local Models - Hugging Face (55K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="252.2" cy="32.5" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Argon Is Not What I Expected - Two Minute Papers (229K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="875.7" cy="247.3" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;I Made $11,000 Selling Stickers: Then Built AI to Scale It. - Build w Agents - Aggregate Intellect - AI.SCIENCE (75)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="758.5" cy="241.0" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;From Click to Booking: I Built the Tracking Most Businesses Are Missing - Build w Agents - Aggregate Intellect - AI.SCIENCE (116)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="603.8" cy="237.8" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This AI Doesn’t Just Remind You, It Knows When to Back Off. - Build w Agents - Aggregate Intellect - AI.SCIENCE (140)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="735.1" cy="158.0" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;The Boring Version of Continual Learning - LangChain (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="519.4" cy="135.5" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How Deep Agents Manages Context: Offloading and Summarization - LangChain (7K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="275.7" cy="171.5" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Podium&amp;#x27;s Playbook for Scaling AI Agents Org-Wide - LangChain (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="261.6" cy="138.7" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;What is LangChain Managed Deep Agents? - LangChain (6K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="814.8" cy="190.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Density functional theory meets AI: traditional DFT vs a pretrained model - DigitalSreeni (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="585.1" cy="194.9" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;15 - Spatial Pharmacokinetics: Measuring Drug Diffusion in Tumor Tissue - DigitalSreeni (908)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="814.8" cy="241.0" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Voice Agents That Interrupt You #Shorts - Trelis Research (116)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="810.1" cy="228.3" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Dialt: Deep Control Over Voice Agent Behavior #Shorts - Trelis Research (230)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="706.9" cy="217.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Voice Agents That Interrupt You - Trelis Research (375)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="697.6" cy="221.2" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Background Speaker Suppression in Dialt #Shorts - Trelis Research (318)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="585.1" cy="219.1" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Dialt Spot: Voice AI at $1.99 per Hour During Quieter Periods - Trelis Research (349)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="519.4" cy="214.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Dialt Launch Webinar: Voice AI Demo, WorkStuff AI &amp;amp; Q&amp;amp;A - Trelis Research (425)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="472.6" cy="218.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Detta and Phonon-2: Fast Dictation on a Mac - Trelis Research (359)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="360.1" cy="217.8" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Dialt: Deep Control Over How Voice Agents Behave and Sound - Trelis Research (368)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="261.6" cy="215.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Background Speaker Suppression in Dialt: A Live Demo - Trelis Research (407)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="247.6" cy="228.5" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Two Ways to Switch Languages with Voice Assistants | Dialt #Shorts - Trelis Research (228)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="397.6" cy="136.4" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Learn to build better AI Agents through better context - DeepLearningAI (7K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="622.6" cy="163.9" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Introducing OCRia-1, the world’s best document parsing model - LlamaIndex (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="524.1" cy="221.6" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Building an Auto Total-Loss Claims Workflow with LlamaParse - LlamaIndex (312)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="416.3" cy="128.2" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Answering your questions about AI, agents, and tech jobs (Q&amp;amp;A) - Dave Ebbelaar (9K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="547.6" cy="129.0" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;&#128266; Voice design and voice replication with Gemini 3.8 Flash TTS - Google for Developers (9K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="528.8" cy="93.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Design and replicate voices with Gemini 3.8 Flash TTS - Google for Developers (30K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="683.5" cy="103.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Triple BAM! Live with Jon Krohn, Josh Starmer, and Luis Serrano! - StatQuest with Josh Starmer (21K)&lt;/title&gt;&lt;/circle&gt;&lt;/svg&gt;&lt;/div&gt;
&lt;div class="rowg g11"&gt;
&lt;div class="card"&gt;&lt;h3&gt;Theme density&lt;/h3&gt;&lt;div class="bars"&gt;&lt;a class="bar" href="#t1"&gt;&lt;span class="bn"&gt;Agent and System Security&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:35%;background:#0d9488"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;6&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t2"&gt;&lt;span class="bn"&gt;Voice AI and Voice Agents&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:100%;background:#d97706"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;17&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t3"&gt;&lt;span class="bn"&gt;Image and Video Generation&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:29%;background:#dc2626"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;5&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t4"&gt;&lt;span class="bn"&gt;Document Understanding and OCR&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:35%;background:#2563eb"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;6&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t5"&gt;&lt;span class="bn"&gt;Kimi and Moonshot&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:47%;background:#4d7c0f"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;8&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t6"&gt;&lt;span class="bn"&gt;DeepSeek&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:41%;background:#7c3aed"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;7&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t7"&gt;&lt;span class="bn"&gt;Gemini and Google Models&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:35%;background:#0284c7"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;6&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t8"&gt;&lt;span class="bn"&gt;Diffusion Language Models&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:6%;background:#92400e"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;1&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t9"&gt;&lt;span class="bn"&gt;Local AI and Privacy&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:65%;background:#db2777"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;11&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t10"&gt;&lt;span class="bn"&gt;Sam Altman and the OpenAI Agenda&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:47%;background:#0f766e"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;8&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t11"&gt;&lt;span class="bn"&gt;AI in Business&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:88%;background:#475569"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;15&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t12"&gt;&lt;span class="bn"&gt;Open Models and Efficiency&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:18%;background:#b45309"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;3&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t13"&gt;&lt;span class="bn"&gt;Coding Agents and Agent Infrastructure&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:76%;background:#0d9488"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;13&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t14"&gt;&lt;span class="bn"&gt;Learning, Science and Research&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:65%;background:#d97706"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;11&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t15"&gt;&lt;span class="bn"&gt;Other Standouts&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:29%;background:#dc2626"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;5&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/div&gt;
&lt;div class="card"&gt;&lt;h3&gt;Most-viewed channels&lt;/h3&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Kai&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:100%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;1.1M&lt;i&gt;8 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Matthew Berman&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:66%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;693K&lt;i&gt;14 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;OpenAI&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:35%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;366K&lt;i&gt;15 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:27%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;281K&lt;i&gt;7 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Two Minute Papers&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:22%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;229K&lt;i&gt;1 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Sam Witteveen&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:9%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;95K&lt;i&gt;3 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;/div&gt;&lt;/div&gt;
&lt;section class="theme" id="t1"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#0d9488"&gt;01&lt;/span&gt;&lt;i class="dot" style="background:#0d9488"&gt;&lt;/i&gt;Agent and System Security&lt;span class="cnt"&gt;6 videos &amp;middot; 52K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;In five days nearly every layer of the developer stack got patched - PHP, Node.js, GitLab, PyPI poisoning - and beyond the patches, the OpenCode API-key theft and Microsoft&amp;#x27;s Project Perception push into enterprise agent security made the point plain: agents are the new attack surface.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=0p96nZsRxlY" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;43K&lt;/span&gt;&lt;span class="dt"&gt;20 Oct&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;OpenCode Got Hacked Badly (My API Keys Got Drained)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=0XSk7fCimYE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;4K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;The Internet Is on Fire (Patch These Now)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=hLoyhm1dHlM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;20 Oct&lt;/span&gt;&lt;span class="ch"&gt;Cole Medin&lt;/span&gt;&lt;span class="tt"&gt;Prompt Injection Is Just the Start of AI Agent Security (Here&amp;#x27;s What You Need)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ST2xBczjTB8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;702&lt;/span&gt;&lt;span class="dt"&gt;20 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Cloudflare Clairic: The $0.10 AI Extraction Model Tested Thoroughly!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=C6_4WsBrlUI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;579&lt;/span&gt;&lt;span class="dt"&gt;22 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Mercury 3 Is INSANE: I Tested 13 Live Examples (+ 100 MILLION FREE Tokens)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=DS-9KjJe1fk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;292&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;PHP Is Broken?! CVE-2026-109618 and 109619 Explained, Patch Today&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t2"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#d97706"&gt;02&lt;/span&gt;&lt;i class="dot" style="background:#d97706"&gt;&lt;/i&gt;Voice AI and Voice Agents&lt;span class="cnt"&gt;17 videos &amp;middot; 183K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Trelis&amp;#x27;s Dialt launch (interruption handling, background-speaker suppression, Spot pricing down to $1.99 an hour), Google&amp;#x27;s voice design and replication across 100-plus languages in Gemini 3.8 Flash TTS, and Block&amp;#x27;s Dessa - trained on human speech - moved voice agents out of demo territory and onto production cost sheets.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=8n-SXxRIrm8" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;73K&lt;/span&gt;&lt;span class="dt"&gt;20 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Build a Better Shopping Experience with OpenAI | Shopping Agent Demo&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=DQHtmW9ZlIs" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;65K&lt;/span&gt;&lt;span class="dt"&gt;19 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Cooking with Codex: Ep. 1&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=SOF3_evPje8" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;30K&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;Design and replicate voices with Gemini 3.8 Flash TTS&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=CsF5oqVV5YE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;9K&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;&#128266; Voice design and voice replication with Gemini 3.8 Flash TTS&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=3Q9A8izO-Ww" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;24 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Block&amp;#x27;s Dessa AI Voice Is CRAZY: It LAUGHS Mid-Sentence (I Played the Real Demos)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=g1xqHojC5Ec" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;782&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Gradium Phonon Is WILD… Tiny Open-Weight TTS Model Tested Thoroughly!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=7-NLWOVANY8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;425&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Dialt Launch Webinar: Voice AI Demo, WorkStuff AI &amp;amp; Q&amp;amp;A&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Jxw85gnfrGQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;407&lt;/span&gt;&lt;span class="dt"&gt;19 Oct&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Background Speaker Suppression in Dialt: A Live Demo&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=eyfZxncUHFE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;375&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Voice Agents That Interrupt You&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Xs2PR-ZIfIU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;368&lt;/span&gt;&lt;span class="dt"&gt;20 Oct&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Dialt: Deep Control Over How Voice Agents Behave and Sound&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Y5HGnlsImJs" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;359&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Detta and Phonon-2: Fast Dictation on a Mac&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=rTMubDuwUyU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;349&lt;/span&gt;&lt;span class="dt"&gt;22 Oct&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Dialt Spot: Voice AI at $1.99 per Hour During Quieter Periods&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Nuip2xEHQ5c" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;318&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Background Speaker Suppression in Dialt #Shorts&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=tW07tpgDUg8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;253&lt;/span&gt;&lt;span class="dt"&gt;24 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;AI Avatars IGNORE You?! I Tested the CRAZY New StreamEval Benchmark (RTX 4090)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=E-J6K-Aao2A" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;230&lt;/span&gt;&lt;span class="dt"&gt;24 Oct&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Dialt: Deep Control Over Voice Agent Behavior #Shorts&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=6EBnVGTqTpQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;228&lt;/span&gt;&lt;span class="dt"&gt;19 Oct&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Two Ways to Switch Languages with Voice Assistants | Dialt #Shorts&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=8l5Efmo0svg" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;116&lt;/span&gt;&lt;span class="dt"&gt;24 Oct&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Voice Agents That Interrupt You #Shorts&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t3"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#dc2626"&gt;03&lt;/span&gt;&lt;i class="dot" style="background:#dc2626"&gt;&lt;/i&gt;Image and Video Generation&lt;span class="cnt"&gt;5 videos &amp;middot; 55K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Kling 4.0 pushed AI video into the long take with 30 seconds in a single clip, while ComfyUI/Flux 3 Klein workflows and ChatGPT&amp;#x27;s headshot feature showed image generation being folded into everyday tools.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ntxxhO9H3bY" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;30K&lt;/span&gt;&lt;span class="dt"&gt;20 Oct&lt;/span&gt;&lt;span class="ch"&gt;pixaroma&lt;/span&gt;&lt;span class="tt"&gt;Flux 3 Klein in ComfyUI | Generate, Edit &amp;amp; Combine Images (Ep39)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=v45XAMqH7pE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;11K&lt;/span&gt;&lt;span class="dt"&gt;20 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;You need a better headshot.&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ipvP48pz_Bw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;10K&lt;/span&gt;&lt;span class="dt"&gt;22 Oct&lt;/span&gt;&lt;span class="ch"&gt;pixaroma&lt;/span&gt;&lt;span class="tt"&gt;ComfyUI Models Explained: Download, Folders and Model Loaders&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=VbF1IiBAsqo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;4K&lt;/span&gt;&lt;span class="dt"&gt;22 Oct&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;How To Use Google Flow (Complete Tutorial)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=qd_TDSkx5Zw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;487&lt;/span&gt;&lt;span class="dt"&gt;24 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Kling 4.0 Is Here… AI Video Just Got SCARY Good&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t4"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#2563eb"&gt;04&lt;/span&gt;&lt;i class="dot" style="background:#2563eb"&gt;&lt;/i&gt;Document Understanding and OCR&lt;span class="cnt"&gt;6 videos &amp;middot; 234K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;LlamaIndex&amp;#x27;s OCRia-1 and the open-source OXR are chasing new records in table- and chart-heavy document parsing, and RAG-Anything reading its own paper was proof that multimodal RAG is starting to mature.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=LWxX6vvHfrc" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;229K&lt;/span&gt;&lt;span class="dt"&gt;19 Oct&lt;/span&gt;&lt;span class="ch"&gt;Two Minute Papers&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Argon Is Not What I Expected&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=1_p2OAT8RZ4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;22 Oct&lt;/span&gt;&lt;span class="ch"&gt;LlamaIndex&lt;/span&gt;&lt;span class="tt"&gt;Introducing OCRia-1, the world’s best document parsing model&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=YIvSE07UCC4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;How to Analyze PDFs and Long Documents with Claude&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=MEOGGYW2xIc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;588&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;RAG-Anything Read Its OWN Paper… and the Results Are WILD&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ytrlq9pIpjo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;312&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;LlamaIndex&lt;/span&gt;&lt;span class="tt"&gt;Building an Auto Total-Loss Claims Workflow with LlamaParse&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=NKzcyVpeyyI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;82&lt;/span&gt;&lt;span class="dt"&gt;25 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;AWESOME or Overhyped? OXR Document Parsing, Tested With My Own Ugly PDFs&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t5"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#4d7c0f"&gt;05&lt;/span&gt;&lt;i class="dot" style="background:#4d7c0f"&gt;&lt;/i&gt;Kimi and Moonshot&lt;span class="cnt"&gt;8 videos &amp;middot; 181K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Kimi K3 took first place on Arena Frontend Code, and K3.1 Pro - a 2.8-trillion-parameter MoE with a 1M-token context - landed one point behind Opus 5.5 and ahead of GPT-6 Astra on Terminal-Bench, at a third of the cost. The open weights are not out yet, and the benchmark claims are contested.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=5sEVvu4mQ8g" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;117K&lt;/span&gt;&lt;span class="dt"&gt;24 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;They actually did it... (Kimi K3.1)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=zgluzf-zjhI" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;36K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;Kimi K3.1 Pro: Open Weights, Almost Opus&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=7fZ1rFcm1nA" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;16K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Team AMA (and Kimi K3.1 Testing!!)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=bBqP-w6DHFw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;7K&lt;/span&gt;&lt;span class="dt"&gt;24 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Kimi K3.1 Invented the Benchmark Scores&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Qh2SuMGLUf0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Kimi K3.1 Could Be INSANE…RUMOURS and LEAKS&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=r3K4fPH6GSo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;24 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Kimi K3.1 Pro vs Opus 5.5 vs GPT-6 Astra: WHO WINS? Real Demos!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=o6lb6_R7VvQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;430&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Kimi K3 Just Went #1 on Frontend Code: The CRAZY Demos, Prices and What It Still Loses&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=52H96MH5UtI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;357&lt;/span&gt;&lt;span class="dt"&gt;24 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Kimi K3.1 Pro Open-Weights Model Is INSANE! TESTED THOROUGHLY ! 92.2% on 302 Tests!&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t6"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#7c3aed"&gt;06&lt;/span&gt;&lt;i class="dot" style="background:#7c3aed"&gt;&lt;/i&gt;DeepSeek&lt;span class="cnt"&gt;7 videos &amp;middot; 178K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;DeepSeek open-sourced DeepInfer, the inference engine behind its API (155 trillion tokens a day, 96% KV-cache hit rate), along with V4.1 Flash 1017; the update that cuts thinking tokens by two-thirds lifted it to first among open models on Arena WebDev.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=zIWxzkO1nak" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;67K&lt;/span&gt;&lt;span class="dt"&gt;24 Oct&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;DeepSeek Is Free Now but What’s the Catch?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=YGEz40RSpTE" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;41K&lt;/span&gt;&lt;span class="dt"&gt;19 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;DeepSeek Just Made Reasoning WAY More Efficient!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=esdRFXgXaaQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;33K&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;DeepSeek Just Dropped a MASSIVE AI Breakthrough!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=RS9vWYlJeH8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;23K&lt;/span&gt;&lt;span class="dt"&gt;18 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;How DeepSeek Uses Less Memory&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=5VLKHqnjO8w" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;9K&lt;/span&gt;&lt;span class="dt"&gt;20 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Mercury 3 Is WILD… 1,000+ Tokens/sec Diffusion LLM Tested Thoroughly!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=sQFIhViy4Uk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;19 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;MiniMax M3.1 Flash JUST DROPPED… I Tested 302 Tasks!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=mQ-nB8g9mOc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;DeepSeek Just OPEN SOURCED Its AI Engine… Here’s How It REALLY Works!&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t7"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#0284c7"&gt;07&lt;/span&gt;&lt;i class="dot" style="background:#0284c7"&gt;&lt;/i&gt;Gemini and Google Models&lt;span class="cnt"&gt;6 videos &amp;middot; 481K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Gemini 4 Argon fell short of expectations while Carbon leaks circulated; Google&amp;#x27;s TimesFM 3.0 forecasts time series with no training step at all, and Harness2 stood out as a research framework that lets an agent&amp;#x27;s scaffolding - prompts, skills, sub-agents - improve itself without retraining the model.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=vLqswKyHS64" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;294K&lt;/span&gt;&lt;span class="dt"&gt;19 Oct&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;NCM Is this the End of LLMs?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=h_KqS4GtU3I" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;83K&lt;/span&gt;&lt;span class="dt"&gt;18 Oct&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;How I Got Frontier Performance on 16GB VRAM (Qwen 3.8 27B)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=pW2ktrJlsyE" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;44K&lt;/span&gt;&lt;span class="dt"&gt;24 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;Harness² Explained: Self-Improving Agents Without Training&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=yCgfyQmqhPM" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;40K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Google&amp;#x27;s Harness² Is This the End of Harness Engineering?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ByKEkRohzt8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;17K&lt;/span&gt;&lt;span class="dt"&gt;22 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;Google&amp;#x27;s GPT-3 Moment for Time Series &amp;amp; Forecasting (TimesFM 3.0)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=qw2HyicPz3w" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Carbon COULD BE  INSANE… Leaked Demos &amp;amp; Rumours Investigated!&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t8"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#92400e"&gt;08&lt;/span&gt;&lt;i class="dot" style="background:#92400e"&gt;&lt;/i&gt;Diffusion Language Models&lt;span class="cnt"&gt;1 video &amp;middot; 22K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Inception&amp;#x27;s Mercury 3, claiming 1,000+ tokens per second, put diffusion LLMs back on the agenda; independent tests measured 258-416 tokens per second end to end, but the speed-plus-free-tokens strategy is hard to ignore.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=bEO_QeJU7U0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;22K&lt;/span&gt;&lt;span class="dt"&gt;20 Oct&lt;/span&gt;&lt;span class="ch"&gt;Sam Witteveen&lt;/span&gt;&lt;span class="tt"&gt;Do Diffusion LLMs Scale? Testing Mercury 3 + Decide&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t9"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#db2777"&gt;09&lt;/span&gt;&lt;i class="dot" style="background:#db2777"&gt;&lt;/i&gt;Local AI and Privacy&lt;span class="cnt"&gt;11 videos &amp;middot; 680K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;The week&amp;#x27;s most-watched video tore down the privacy assumption behind local AI. The local side keeps gaining ground regardless: llama.cpp made 125B MoE models 4x faster, an official Llama app arrived on Mac, and speculative-decoding tutorials kept spreading.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=3mx84MRj8lY" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;397K&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Your Local AI Is Not as Private as You Think&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ugKAkCVBcz0" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;74K&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;I Tested Every Best Local AI For GPU (8GB TO 24GB VRAM)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=fHY0sjAejcA" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;59K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Who is local AI really for?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ti3vVYwzW4A" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;55K&lt;/span&gt;&lt;span class="dt"&gt;19 Oct&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;Pi Durable EXPLAINED: Build Always-On Agents with Local Models&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=shl8FFSoANE" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;53K&lt;/span&gt;&lt;span class="dt"&gt;22 Oct&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;llama.cpp Just Made 125B MoE Models 4× Faster!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=D3YM5OzNt9s" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;18K&lt;/span&gt;&lt;span class="dt"&gt;18 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;NVIDIA Dory 4B Is WILD… Looped Transformer, Same Memory?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=H4azmWcqO5g" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;14K&lt;/span&gt;&lt;span class="dt"&gt;20 Oct&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;Local AI 202: Get Massive Speed-ups with Speculative Decoding&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=w6dSNCg9I5k" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;6K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;Llama: Local AI on Your Mac&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=3-sSbJKb-fQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;19 Oct&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;Pi Durable: AI agents that survive crashes&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=vmgbq-7WRpQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;699&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;I Ran Flint, the CRAZY New 40B Coding Model, WITHOUT a Mac. Did It Fix My Buggy Repo?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ZDYrcHPzlsM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;284&lt;/span&gt;&lt;span class="dt"&gt;25 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;XERJ&amp;#x27;s CRAZY Local Search for AI Agents: I Tested It, It Broke, Then It Got AWESOME&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t10"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#0f766e"&gt;10&lt;/span&gt;&lt;i class="dot" style="background:#0f766e"&gt;&lt;/i&gt;Sam Altman and the OpenAI Agenda&lt;span class="cnt"&gt;8 videos &amp;middot; 430K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Matthew Berman&amp;#x27;s Altman interview series ran on recursive self-improvement, AI consciousness, the need for new hardware and why demand for coders never disappeared - the most-discussed primary source on where OpenAI is heading.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=3gXLmy7DK_4" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;355K&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Sam Altman: RSI, AI Consciousness, OpenAI&amp;#x27;s New Hardware, Compute, and more!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=1g9Tqvyt4kU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;22K&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Sam Altman on Why Demand for Coders Hasn&amp;#x27;t Disappeared&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=u6Tf62_5CP0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;17K&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Sam Altman on Why AI Needs a New Kind of Hardware&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=tc92psltcEg" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;16K&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Sam Altman on What Personal AI Still Needs&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=SU-h1k5IYOI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;10K&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Sam Altman on Why AI Hasn&amp;#x27;t Made Us Less Busy&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=9WZpou1faFU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;7K&lt;/span&gt;&lt;span class="dt"&gt;22 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Sam Altman on AI’s Discovery Era&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=f8uIINtoJCM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Sam Altman on Why Your Most Important Advice Should Come From People&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=eQTOJZ6_NVU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;650&lt;/span&gt;&lt;span class="dt"&gt;25 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Sam Altman on What People Still Bring to an AI World&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t11"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#475569"&gt;11&lt;/span&gt;&lt;i class="dot" style="background:#475569"&gt;&lt;/i&gt;AI in Business&lt;span class="cnt"&gt;15 videos &amp;middot; 128K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Uber opening ChatGPT Work to 30,000+ employees, AT&amp;amp;T&amp;#x27;s legal workflows and Thomson Reuters&amp;#x27; trust-first legal AI showed the scale of enterprise adoption; a sticker business validated by real sales and scaled with agents, alongside sales-funnel analysis, stood in for the small-business side.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=20IRGgQwPkc" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;27K&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;V7 cuts costs 78% while boosting accuracy with GPT‑5.6 Luna&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=l16Gw7oVLNQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;26K&lt;/span&gt;&lt;span class="dt"&gt;22 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Uber rolls out ChatGPT Work to 30,000+ employees&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=m6MlxaFMmPI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;15K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;How Shield uses OpenAI’s Astra to support IT teams&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=hgX1hgyWD4g" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;14K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;How Realpage. builds AI training with ChatGPT Work&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=V39_L7uh66M" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;13K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;How AT&amp;amp;T is building AI-powered legal workflows with OpenAI&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=CTkDi1THc_g" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;10K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;How Benefit Street Partners is building financial analysis tools with Codex&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=7lC2N9woZy4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;8K&lt;/span&gt;&lt;span class="dt"&gt;19 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Thomson Reuters on Building Trust in Legal AI | OpenAI&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=7Cs5qyBlYQk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;5K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;From notes to AI training with ChatGPT Work | Realpage.&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=2lNUGkCg_qc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;4K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;AI agents for everyday IT fixes | Shield and OpenAI&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=8mp0jF2At1o" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;19 Oct&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;Podium&amp;#x27;s Playbook for Scaling AI Agents Org-Wide&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=qyGvwLgqmaU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;From analyst to builder with Codex | Benefit Street Partners&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=vc7b-u9G6kg" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;19 Oct&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;HubSpot CRM Tutorial For Beginners&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=wLuuesGRhuM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;20 Oct&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;How To Sell On TikTok Shop For Beginners (2026 Guide)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=pFWN1VBCQ28" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;116&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;Build w Agents - Aggregate Intellect - AI.SCIENCE&lt;/span&gt;&lt;span class="tt"&gt;From Click to Booking: I Built the Tracking Most Businesses Are Missing&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=AXFOzSRg5i8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;75&lt;/span&gt;&lt;span class="dt"&gt;24 Oct&lt;/span&gt;&lt;span class="ch"&gt;Build w Agents - Aggregate Intellect - AI.SCIENCE&lt;/span&gt;&lt;span class="tt"&gt;I Made $11,000 Selling Stickers: Then Built AI to Scale It.&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t12"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#b45309"&gt;12&lt;/span&gt;&lt;i class="dot" style="background:#b45309"&gt;&lt;/i&gt;Open Models and Efficiency&lt;span class="cnt"&gt;3 videos &amp;middot; 17K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Meta&amp;#x27;s 816-billion-parameter open-weight Muse Spark Open, NVIDIA Dory 4B&amp;#x27;s looped transformer, MiniMax M3.1 and Spark-X2.6-4B&amp;#x27;s &amp;#x27;thinking dial&amp;#x27; tested on an 8GB GPU all pointed at the same race: architectural efficiency in open models is accelerating.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=aKVfjG7_R4g" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;11K&lt;/span&gt;&lt;span class="dt"&gt;19 Oct&lt;/span&gt;&lt;span class="ch"&gt;Sam Witteveen&lt;/span&gt;&lt;span class="tt"&gt;Meta&amp;#x27;s New Models &amp;amp; Agent&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Ne1nEO6L-kE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;6K&lt;/span&gt;&lt;span class="dt"&gt;22 Oct&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Recursive Flint 40B Coding Model: Can It Code on Just 24GB RAM?!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ZCXaQEOkaDI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;235&lt;/span&gt;&lt;span class="dt"&gt;24 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Spark-X2.6-4B&amp;#x27;s CRAZY Thinking Dial: 180 Tests on an 8GB GPU&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t13"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#0d9488"&gt;13&lt;/span&gt;&lt;i class="dot" style="background:#0d9488"&gt;&lt;/i&gt;Coding Agents and Agent Infrastructure&lt;span class="cnt"&gt;13 videos &amp;middot; 308K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Recursive&amp;#x27;s Flint - a 40B coding model trained with RSI - was the week&amp;#x27;s curiosity; LangChain&amp;#x27;s Managed Deep Agents, offloading and summarization against context rot, and Cole Medin&amp;#x27;s self-improving harness showed agent infrastructure firming up.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=0sI-GTYetD8" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;106K&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;Claude Duet: Anthropic&amp;#x27;s Proactive Agent&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=BsfjQDSeTwI" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;62K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;Sam Witteveen&lt;/span&gt;&lt;span class="tt"&gt;Flint - The First RSI Coding Model?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=6f7ojkd4_Wc" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;54K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Hunt bugs with your dot&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=TVhhmhvCovg" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;25K&lt;/span&gt;&lt;span class="dt"&gt;20 Oct&lt;/span&gt;&lt;span class="ch"&gt;Cole Medin&lt;/span&gt;&lt;span class="tt"&gt;Let Your AI Agent Do ITS OWN Harness Engineering (Here&amp;#x27;s How)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=-LODwSuymII" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;14K&lt;/span&gt;&lt;span class="dt"&gt;20 Oct&lt;/span&gt;&lt;span class="ch"&gt;The Gray Cat&lt;/span&gt;&lt;span class="tt"&gt;I Tested 4 Codex Frameworks on the Same Real Feature&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=tNKmMhiOw9w" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;13K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;Cole Medin&lt;/span&gt;&lt;span class="tt"&gt;Run Your AI Software Factory Like a Real Engineering Org (Here&amp;#x27;s How)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=9jjn54Me-Os" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;8K&lt;/span&gt;&lt;span class="dt"&gt;19 Oct&lt;/span&gt;&lt;span class="ch"&gt;Burke Holland&lt;/span&gt;&lt;span class="tt"&gt;I tested four AI strategies to find the cheapest one&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Ud5RsBegyfw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;7K&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;How Deep Agents Manages Context: Offloading and Summarization&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=CUjHZlvQufk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;7K&lt;/span&gt;&lt;span class="dt"&gt;20 Oct&lt;/span&gt;&lt;span class="ch"&gt;DeepLearningAI&lt;/span&gt;&lt;span class="tt"&gt;Learn to build better AI Agents through better context&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=tJ-hlNVS7AI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;6K&lt;/span&gt;&lt;span class="dt"&gt;19 Oct&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;What is LangChain Managed Deep Agents?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=wSSMODxxfbk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;5K&lt;/span&gt;&lt;span class="dt"&gt;20 Oct&lt;/span&gt;&lt;span class="ch"&gt;Burke Holland&lt;/span&gt;&lt;span class="tt"&gt;Copilot Can Use Your Computer for You #ai #coding #githubcopilot&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=bkj4Hc3sED4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;349&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;This AI Agent Builds Its OWN Tools to Finish Tasks! (INSANE Agent S4 Test)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=7Y6Fr_yl6Gw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;328&lt;/span&gt;&lt;span class="dt"&gt;24 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Microsoft&amp;#x27;s Agent World Is INSANE. I Ran It. The Agent FAILED (0/3)&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t14"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#d97706"&gt;14&lt;/span&gt;&lt;i class="dot" style="background:#d97706"&gt;&lt;/i&gt;Learning, Science and Research&lt;span class="cnt"&gt;11 videos &amp;middot; 81K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Zachary Huang&amp;#x27;s hour-long deep-learning marathons and a live StatQuest carried the teaching side, while Paper2Agent turning every paper into an MCP server and OlmoEarth&amp;#x27;s satellite embeddings marked research&amp;#x27;s touchpoints with AI.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=wbc-oxFFV9c" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;39K&lt;/span&gt;&lt;span class="dt"&gt;18 Oct&lt;/span&gt;&lt;span class="ch"&gt;Zachary Huang&lt;/span&gt;&lt;span class="tt"&gt;Give Me 1 Hour, I&amp;#x27;ll Make Deep Learning Click Forever&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=AgypoO13DJ0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;21K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;StatQuest with Josh Starmer&lt;/span&gt;&lt;span class="tt"&gt;Triple BAM! Live with Jon Krohn, Josh Starmer, and Luis Serrano!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=DWhgz1xDQ40" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;7K&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;Zachary Huang&lt;/span&gt;&lt;span class="tt"&gt;Give Me 1 Hour, I&amp;#x27;ll Make Multimodal Deep Learning Click Forever&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=qLEBXkHb_w4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;5K&lt;/span&gt;&lt;span class="dt"&gt;24 Oct&lt;/span&gt;&lt;span class="ch"&gt;Zachary Huang&lt;/span&gt;&lt;span class="tt"&gt;Give Me 35 Minutes, I&amp;#x27;ll Make LLM Pre-training Click Forever&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=StDAmjuQOE4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;23 Oct&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;The Boring Version of Continual Learning&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=UC9PdhD6Yxg" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;22 Oct&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;Automate CSV to Parquet with Buckets, Webhooks &amp;amp; Jobs&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=KnDMnZYEo2M" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;I Turned Structured Output Into a Game &#127918;&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=eTNyW0cBqRI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;24 Oct&lt;/span&gt;&lt;span class="ch"&gt;DigitalSreeni&lt;/span&gt;&lt;span class="tt"&gt;Density functional theory meets AI: traditional DFT vs a pretrained model&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=s9Sl86wE_5c" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;908&lt;/span&gt;&lt;span class="dt"&gt;22 Oct&lt;/span&gt;&lt;span class="ch"&gt;DigitalSreeni&lt;/span&gt;&lt;span class="tt"&gt;15 - Spatial Pharmacokinetics: Measuring Drug Diffusion in Tumor Tissue&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=9xejboeNY_E" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;275&lt;/span&gt;&lt;span class="dt"&gt;24 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;OlmoEarth v1.3: INSANE Satellite AI or Fooled by WINTER? I Tested It on My RTX 4060&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=XgGHb0HYQ60" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;248&lt;/span&gt;&lt;span class="dt"&gt;24 Oct&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;This Nature Paper Turns Any Research Paper Into an AI Agent - INSANE&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t15"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#dc2626"&gt;15&lt;/span&gt;&lt;i class="dot" style="background:#dc2626"&gt;&lt;/i&gt;Other Standouts&lt;span class="cnt"&gt;5 videos &amp;middot; 113K views&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;A chess lesson from Messi, a robot bodyguard demo, confessions about focus under parallel agents and a cheapest-AI-strategy shootout were the week&amp;#x27;s lighter but thought-provoking corner: as speed climbs, attention and the human role are the new bottleneck.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=5eJnIYrr8rg" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;40K&lt;/span&gt;&lt;span class="dt"&gt;21 Oct&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Actual Robot Bodyguard&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=RLQR5_5fJDA" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;39K&lt;/span&gt;&lt;span class="dt"&gt;22 Oct&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Messi asks ChatGPT to teach him a chess opening&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=zj1TJlP7c9Y" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;25K&lt;/span&gt;&lt;span class="dt"&gt;22 Oct&lt;/span&gt;&lt;span class="ch"&gt;Cole Medin&lt;/span&gt;&lt;span class="tt"&gt;AI Makes Us WAY Faster but Fries Our Focus (Here&amp;#x27;s What I Do About It)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=UwrZJpjWjdI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;9K&lt;/span&gt;&lt;span class="dt"&gt;20 Oct&lt;/span&gt;&lt;span class="ch"&gt;Dave Ebbelaar&lt;/span&gt;&lt;span class="tt"&gt;Answering your questions about AI, agents, and tech jobs (Q&amp;amp;A)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=AAdp41dRluA" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;140&lt;/span&gt;&lt;span class="dt"&gt;22 Oct&lt;/span&gt;&lt;span class="ch"&gt;Build w Agents - Aggregate Intellect - AI.SCIENCE&lt;/span&gt;&lt;span class="tt"&gt;This AI Doesn’t Just Remind You, It Knows When to Back Off.&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;
&lt;footer&gt;This report was compiled by &lt;b style="color:#111"&gt;Murat Karakaya Academy&lt;/b&gt; · Data: YouTube RSS (live, no API key) · window: 2026-10-18 .. 2026-10-25 (UTC, computed at run time) · max 15 videos per channel · The Prompt Engineer 48 and OpenAI feeds hit the RSS 15-video cap (partial); videos from early in the window on those channels may be missing. Window: October 18-25, 2026 (UTC), 24 of 43 channels active, 122 videos. · render.py @ 27.10.2026 23:03 UTC&lt;/footer&gt;
&lt;/div&gt;&lt;/body&gt;&lt;/html&gt;</description><link>https://www.muratkarakaya.net/2026/10/youtube-ai-weekly-trend-report-october.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-9100448324890194848</guid><pubDate>Tue, 27 Oct 2026 15:56:31 +0000</pubDate><atom:updated>2026-10-27T18:56:31.502+03:00</atom:updated><category domain="http://www.blogger.com/atom/ns#">Gradium Phonon</category><category domain="http://www.blogger.com/atom/ns#">Metinden Sese</category><category domain="http://www.blogger.com/atom/ns#">Murat Karakaya Akademi</category><category domain="http://www.blogger.com/atom/ns#">Ollama</category><category domain="http://www.blogger.com/atom/ns#">Sesli Erişilebilirlik</category><category domain="http://www.blogger.com/atom/ns#">Yerel Yapay Zeka</category><title>Ollama artık konuşuyor: Gradium'un 31 milyonluk modeli Phonon metni buluta çıkmadan seslendiriyor — ama ailede Türkçe yok, Windows ve Linux henüz sırada</title><description>&lt;!DOCTYPE html&gt;
&lt;!-- ============================================================
   MIS-2026-004 (rev.6 kalıbı) · YAZAR-20261027-PHONONTTS · haber_id 1a174535911ab020#1
   ANAYASA (ic_kabuk_karari.md Rev.2): içerikte SIFAR görünüm CSS'i; yalnız semantik
   HTML; başlık içerik tarafından basılmaz — ilk görünür içerik SPOT paragrafı.
   Başlık metni yalnız Blogger BAŞLIK ALANINA ve aşağıdaki &lt;title&gt;'a yazılır.
   BAŞLIK: "Ollama artık konuşuyor: Gradium'un 31 milyonluk modeli Phonon metni buluta
   çıkmadan seslendiriyor — ama ailede Türkçe yok, Windows ve Linux henüz sırada"
   ============================================================ --&gt;
&lt;html lang="tr"&gt;
&lt;head&gt;
&lt;meta charset="utf-8"&gt;
&lt;meta name="viewport" content="width=device-width, initial-scale=1"&gt;
&lt;title&gt;Ollama artık konuşuyor: Gradium'un 31 milyonluk modeli Phonon metni buluta çıkmadan seslendiriyor — ama ailede Türkçe yok, Windows ve Linux henüz sırada&lt;/title&gt;
&lt;/head&gt;
&lt;body&gt;

  &lt;p&gt;Ollama artık konuşuyor, üstelik makinenizden dışarı hiçbir veri çıkarmadan: tek bir komutla metninizi sese çeviriyor. Sesi veren Cenevreli Gradium'un 31 milyon parametrelik açık ağırlıklı Phonon modeli ve bu hamle, yerel yapay zekâ yığınının eksik kalan son halkasını kapatıyor. Yalnız şunu baştan söyleyelim: Phonon'un konuştuğu beş dilde Türkçe yok, ses de henüz yalnız Apple Silicon'lı Mac'lerden yükseliyor; Windows ve Linux sırasını bekliyor. İngilizce içerik üretenler için yine de önemli bir eşik: bulutun token ücretine uğramadan, makinenizin içinde çalışan bir seslendirme hattı açıldı.&lt;/p&gt;

  &lt;p&gt;&lt;strong&gt;Bu Yazıda&lt;/strong&gt;&lt;/p&gt;
  &lt;ol&gt;
    &lt;li&gt;&lt;a href="#bol-1"&gt;Katalog sustuğu yerde konuşmaya başladı&lt;/a&gt;&lt;/li&gt;
    &lt;li&gt;&lt;a href="#bol-2"&gt;Asıl hikâye hızda değil, mimaride&lt;/a&gt;&lt;/li&gt;
    &lt;li&gt;&lt;a href="#bol-3"&gt;Lisans zemini: tek model ticari, gerisi değil&lt;/a&gt;&lt;/li&gt;
    &lt;li&gt;&lt;a href="#bol-4"&gt;İki yıldızlı dipnot: Türkçe yok, macOS dışı yok&lt;/a&gt;&lt;/li&gt;
    &lt;li&gt;&lt;a href="#bol-5"&gt;Yarın sabah ne denenebilir&lt;/a&gt;&lt;/li&gt;
    &lt;li&gt;&lt;a href="#ka-yorum"&gt;Murat Karakaya Akademi Yorumu&lt;/a&gt;&lt;/li&gt;
    &lt;li&gt;&lt;a href="#kaynakca"&gt;Kaynakça&lt;/a&gt;&lt;/li&gt;
  &lt;/ol&gt;

  &lt;h2 id="bol-1"&gt;1. Katalog sustuğu yerde konuşmaya başladı&lt;/h2&gt;
  &lt;p&gt;Kronolojiyi ayıralım: Cenevre merkezli Gradium, açık ağırlıklı ses modeli ailesi Phonon'u 20 Ekim'de yayımladı ve aynı gün "yerel dil modeli, yerel ses, makinenizden hiçbir şey çıkmıyor" mesajıyla Ollama entegrasyonunu duyurdu [2]. Hafta sonuna doğru Ollama kendi blog yazısını koydu: &lt;code&gt;ollama run phonon&lt;/code&gt; yazıyorsunuz, model metni 24kHz mono ses olarak üretiyor; sohbet çıktısını boruyla besleyebiliyor, WAV'a yönlendirebiliyor, hatta OpenAI'ın &lt;code&gt;/v1/audio/speech&lt;/code&gt; ucunu taklit eden uç sayesinde varolan kodunuz yalnız base URL değiştirerek yerelleşebiliyor [1]. Katalogda on dört model etiketi duruyor; Zoey, Easton, Freya, Hugh gibi hazır sesler, aksan varyantları ve &lt;code&gt;ollama show phonon&lt;/code&gt; ile döküm [10].&lt;/p&gt;
  &lt;p&gt;Duyurunun teknik ağırlığı entegrasyonun kendisinde: Ollama bugüne dek metin üreten bir motordu; artık aynı runtime, aynı katalog, aynı API ruhu içinde ses üretiyor. 9to5Mac'in deyimiyle Mac'inize yerel metinden-sese geliyor [8]. Yeni bir servis değil — varolan bir boru hattının eksik halkası.&lt;/p&gt;

  &lt;h2 id="bol-2"&gt;2. Asıl hikâye hızda değil, mimaride&lt;/h2&gt;
  &lt;p&gt;Phonon'u ilginç kılan rakamlar değil boyut. Akış-eşleimli (flow-matching) bir dönüştürücü sesi 80 milisaniyelik kareler halinde latente diziyor; Kyutai'nin Mimi kodeği bu latenti sese çeviriyor; tüm bina Kyutai'nin Pocket TTS'i üzerine kurulu [4]. İngilizce model 31 milyon parametre: q8 nicelemeyle 64,5 megabaytlık bir dosya, çalışırken bellekte yaklaşık 121 megabayt [4][10]. Bu, bir telefon oyununun varlık klasörü kadar yer kaplayan bir TTS motoru. Gradium "dizüstü işlemcide ilk ses 7 milisaniyede başlıyor" diyor [5] — şirketin kendi ölçümü, notunu düşerek alıyoruz.&lt;/p&gt;
  &lt;p&gt;Hız iddialarında iki kademeli dürüstlük gerekiyor. Ollama'nın blog grafiği, MacBook Pro M5 Max'te kendi motorunu Gradium'un çalışma zamanından (ptts) yüzde 35 daha hızlı ölçtüğünü ve kelime hata oranının HF Transformers referansıyla birebir gittiğini söylüyor [1]. 9to5Mac ve Digital Trends bu aktarımı "iddia ediliyor" diliyle verdi; bağımsız yeniden ölçüm henüz yok [8][7]. Üçüncü elden dolaşan "gerçek zamanın yüz katı hız" cümlesi ise Zlio'nun haber akışından; ölçüm koşulları birincil kaynakta yayınlanmış değil [6]. Bir de triajdan sızan detay: dolaşan "yüzde 33" sayısı kaynaklarda geçmiyor, resmî oran yüzde 35 ve kıyas Gradium'un kendi motoruna karşı [3].&lt;/p&gt;
  &lt;p&gt;Doğallık tarafı da aynı çerçevede: Gradium'un zor cümleler için yayımladığı hata tablosu 31m modeli yüzde 1,70'e, Kokoro'yu yüzde 3,50'ye, NeuTTS Nano'yu yüzde 8,71'e koyuyor; kör dinleme Elo puanları dil başına 170-470 değerlendirmeciyle toplanmış [4][5]. Tablolar Gradium'un; bağımsız bir arenanın teyidi yok — DigIntoAI'nin "kendi testimizi yapmadık" dürüstlüğünü not düşmek borç [3].&lt;/p&gt;

  &lt;h2 id="bol-3"&gt;3. Lisans zemini: tek model ticari, gerisi değil&lt;/h2&gt;
  &lt;p&gt;Aileyi kurmadan önce lisans matrisine bakmak lazım, çünkü iki farklı Türkiye senaryosu doğuruyor. phonon-31m-en Apache 2.0: ticari kullanım, ince ayar, yeniden dağıtım serbest [4]. 40M'lik dil modelleri ve çok dilli sürüm CC BY-NC 4.0 — ticari değil; Gradium bu kapıyı "satışlarla görüşün"e bağlıyor, kamuya açık fiyat yok [5]. Digital Trends işi okuruna açık açık yazan ender örneklerden: "İngilizce model ücretsiz; Fransızca, Almanca, İspanyolca, Portekizce ve çok dilli modeller ticari olmayan kullanım ile sınırlı" [7]. Bir podcast introsu ya da İngilizce tanıtım sesi üretiyorsanız 31m-en temiz; çok dilli modeli iş akışına koymadan önce lisans defterini açın.&lt;/p&gt;

  &lt;h2 id="bol-4"&gt;4. İki yıldızlı dipnot: Türkçe yok, macOS dışı yok&lt;/h2&gt;
  &lt;p&gt;Yazının en çok yanlış anlaşılacak yeri burası, o yüzden direkt söyleyelim: Phonon Türkçe bilmiyor. Ailede beş dil var — İngilizce, Fransızca, Almanca, İspanyolca, Portekizce — ve model kartı başka dillerdeki metnin "yanlış telaffuzla" okunacağını kendisi yazıyor [4][10]. Türkçe seslendirme işi bu modelle bugün yapılamaz; "denerseniz nasıl çıkar" da iddia edilemez, kartın uyarısı ortada. GitHub'daki dil talep başlıkları (Korece örneği) listenin topluluk için de dar olduğunu gösteriyor [11]. TR içerik üreten biri için Phonon bir seslendirme çözümü değil; en fazla İngilizce varlık üretiminin ve erişilebilirliğin aracı.&lt;/p&gt;
  &lt;p&gt;İkinci kısıt platform: gereksinim Apple Silicon ve güncel Ollama. Windows ve Linux "yakında" statüsünde; yayınlanmış tarih yok [1]. Küçük bir sürüm nüansı da kaçırmayın: Ollama'nın dokümanı 0.40.9 ve sonrasını söylüyor, blog yazısı 0.40.11 diyor; ikisi de "sonrası" dediğine göre kurulumda &lt;code&gt;ollama --version&lt;/code&gt; ile 0.40.11+ gördüğünüzden emin olun [9][1][3].&lt;/p&gt;

  &lt;h2 id="bol-5"&gt;5. Yarın sabah ne denenebilir&lt;/h2&gt;
  &lt;p&gt;Üç komut, sıfır bulut. Birincisi çekirdek hat: &lt;code&gt;ollama pull phonon&lt;/code&gt; sonra &lt;code&gt;ollama run phonon --voice Zoey "Your text here"&lt;/code&gt; — İngilizce bir paragraf, anında ses. İkincisi dışa aktarma: &lt;code&gt;ollama run phonon "metin" &amp;gt; ses.wav&lt;/code&gt;; 24kHz mono dosyayı kurgu yazılımınıza atarsınız [1][9]. Üçüncüsü asıl yerelleştirme hamlesi: &lt;code&gt;curl http://localhost:11434/v1/audio/speech&lt;/code&gt; ucu OpenAI biçiminde konuştuğu için, bulut TTS'e yazılmış mevcut kodunuz base URL'i &lt;code&gt;localhost:11434&lt;/code&gt;'e çevirince token ücreti ödemeden çalışmaya başlıyor [1]. Sohbeti sesli dinlemek isteyen için boru hazır: &lt;code&gt;ollama run gemma4 "kısaca anlat" | ollama run phonon&lt;/code&gt; — model yazıyor, Phonon okuyor; topluluk eğitimleri kurulumdan API'ye aynı günü gezdi [12]. Ekran okuyucu alışkanlığı olanlar, uzun makaleyi gözle değil kulakla tüketmek isteyenler: bu, o iş akışının bulutsuz hali.&lt;/p&gt;

  &lt;h2 id="ka-yorum"&gt;Murat Karakaya Akademi Yorumu&lt;/h2&gt;
  &lt;p&gt;Verilerin söylediği net: yerel çıkarımın sessizlik problemi çözüldü — 64,5 megabaytlık açık ağırlık bir model, OpenAI-uyumlu bir uç ve tek komut, karşılığında ses. Hız ve doğallık rakamlarının tamamı Ollama ve Gradium'un kendi ölçümleri; bağımsız tekrar henüz yok, bunu yazı boyunca ayırdık. Ayırmadığımız tek şey kanıtlı negatif: Türkçe desteklenmiyor ve model kartı bunu kendisi itiraf ediyor. Aynı şekilde Windows/Linux takvimi belirsiz — "yakında" bir tarih değil.&lt;/p&gt;
  &lt;p&gt;Kanaatimize gelince: Ollama sesle yeni bir teknoloji icat etmedi; yeni bir şeyi tamamladı. Metin, görüntü ve araç çağrısından sonra ses de aynı katalog, aynı runtime, aynı mahremiyet iddiası içinde yerini alınca, "yerel yığın" ifadesi ilk kez eksiksiz kullanılmayı hak ediyor. Bulut TTS satıcıları için can acıtıcı olan yüzde 35 değil, base URL'in bir karakter değişimiyle yerelleşebilmesi: bağımlılığın mimarisi inceldi.&lt;/p&gt;
  &lt;p&gt;Kullanıcıya önerimiz ölçülü bir senaryo. İngilizce içerik üretiyorsanız ya da makaleyi kulakla takip etmek istiyorsanız, Apache 2.0'lı 31m-en'i bugün kurun — lisansı temiz, bedeli sıfır. Türkçe seslendirme için bu yazı size çözüm satmıyor: kapı ya bulutta kalacak ya da yerel TR modellerinin geleceği haftayı bekleyecek. Ve çok dilli modeli ticari işe koymadan önce o CC BY-NC satırını bir kez daha okuyun — ücretsiz olan her şey ucuz değildir.&lt;/p&gt;

  &lt;h2 id="kaynakca"&gt;Kaynakça&lt;/h2&gt;
  &lt;ol&gt;
    &lt;li&gt;ollama.com/blog — &lt;a href="https://ollama.com/blog/speech-models"&gt;"Speech models in Ollama"&lt;/a&gt; (24.10.2026; komutlar, %35 ölçümü, WER eşitliği, &lt;code&gt;/v1/audio/speech&lt;/code&gt;) · erişim: 27.10.2026&lt;/li&gt;
    &lt;li&gt;x.com/@GradiumAI — &lt;a href="https://x.com/GradiumAI/status/2112563128487514512"&gt;Phonon açık ağırlık + Ollama entegrasyon duyuruları&lt;/a&gt; (20.10.2026; "nothing leaves your machine") · erişim: 27.10.2026&lt;/li&gt;
    &lt;li&gt;digintoai.com — &lt;a href="https://digintoai.com/en/articles/gradium-phonon-offline-tts"&gt;"Gradium Phonon offline TTS" analizi&lt;/a&gt; (sürüm zaman çizelgesi, 0.40.9/0.40.11 ayrımı, "kendi testimizi yapmadık" notu) · erişim: 27.10.2026&lt;/li&gt;
    &lt;li&gt;huggingface.co — &lt;a href="https://huggingface.co/gradium/phonon-31m-en"&gt;gradium/phonon-31m-en model kartı&lt;/a&gt; (22.10.2026; mimari, 64,5MB q8, lisans matrisi, dil sınırlaması, Elo tablosu) · erişim: 27.10.2026&lt;/li&gt;
    &lt;li&gt;gradium.ai — &lt;a href="https://gradium.ai/phonon"&gt;Phonon ürün sayfası + açık ağırlık duyuru blogu&lt;/a&gt; (WER tabloları, 7ms ilk ses, zor-500 seti) · erişim: 27.10.2026&lt;/li&gt;
    &lt;li&gt;zlio.com — &lt;a href="https://zlio.com/voice/"&gt;AI Voice haber akışı&lt;/a&gt; ("gerçek zamanın ~100 katı" aktarımı) · erişim: 27.10.2026&lt;/li&gt;
    &lt;li&gt;digitaltrends.com — &lt;a href="https://www.digitaltrends.com/computing/ollama-brings-free-local-text-to-speech-to-your-mac-with-natural-sounding-ai-voices/"&gt;"Ollama brings free, local text-to-speech to your Mac"&lt;/a&gt; (26.10.2026; lisans uyarısını okura açıkça yazan aktarım) · erişim: 27.10.2026&lt;/li&gt;
    &lt;li&gt;9to5mac.com — &lt;a href="https://9to5mac.com/2026-10-26/ollama-for-mac-gains-support-for-local-text-to-speech-models/"&gt;"Ollama for Mac gains support for local text-to-speech models"&lt;/a&gt; (26.10.2026; %35 ve WER aktarımı, "claims" dili) · erişim: 27.10.2026&lt;/li&gt;
    &lt;li&gt;docs.ollama.com — &lt;a href="https://docs.ollama.com/capabilities/speech"&gt;Speech yetenek sayfası&lt;/a&gt; (0.40.9+ koşulu, ses komutları, oynatıcılar, WAV yönlendirme) · erişim: 27.10.2026&lt;/li&gt;
    &lt;li&gt;ollama.com/library/phonon — &lt;a href="https://ollama.com/library/phonon"&gt;model sayfası&lt;/a&gt; (14 etiket, bellek tablosu, Apple Silicon gereksinimi, dil listesi) · erişim: 27.10.2026&lt;/li&gt;
    &lt;li&gt;github.com/ollama/ollama — &lt;a href="https://github.com/ollama/ollama/issues/19295"&gt;issue #19295&lt;/a&gt; (26.10.2026; Korece dahil dil talepleri) · erişim: 27.10.2026&lt;/li&gt;
    &lt;li&gt;youtube.com — &lt;a href="https://www.youtube.com/watch?v=XhJvFaUKgeQ"&gt;"Ollama TTS Tutorial"&lt;/a&gt; (Mohamed Naji Aboo, 26.10.2026; kurulum, Gemma→Phonon boru hattı, sürüm kontrolü) · erişim: 27.10.2026&lt;/li&gt;
  &lt;/ol&gt;

  &lt;p&gt;Etiketler: #MuratKarakayaAkademi #Ollama #Gradium #Phonon #MetindenSese #YerelYapayZeka&lt;/p&gt;

  &lt;footer&gt;
    &lt;p&gt;Yayın tarihi: 27 Ekim 2026 · muratkarakaya.net&lt;/p&gt;
    &lt;p&gt;Murat Karakaya Akademi&lt;/p&gt;
  &lt;/footer&gt;

&lt;/body&gt;
&lt;/html&gt;
CEO-RAPOR: tamamlandı
</description><link>https://www.muratkarakaya.net/2026/10/ollama-artk-konusuyor-gradiumun-31.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-1735723117458410352</guid><pubDate>Tue, 27 Oct 2026 15:55:23 +0000</pubDate><atom:updated>2026-10-27T18:55:23.346+03:00</atom:updated><category domain="http://www.blogger.com/atom/ns#">Açık Ağırlık Modeller</category><category domain="http://www.blogger.com/atom/ns#">Murat Karakaya Akademi</category><category domain="http://www.blogger.com/atom/ns#">Nemotron 3.5 Super VL</category><category domain="http://www.blogger.com/atom/ns#">NVIDIA Nemotron</category><category domain="http://www.blogger.com/atom/ns#">Ollama</category><category domain="http://www.blogger.com/atom/ns#">Yerel Yapay Zeka</category><title>NVIDIA Nemotron 3.5 Super VL'i açtı: bağımsız ölçümde açık ağırlık sınıfının en hızlı ajan modeli, zekâ endeksinde lig liderlerinin gerisinde</title><description>&lt;!-- MIS-2026-004 · YAZIM-20261023 · haber_id 1a142e6261dcfe95#1 · ANAYASA: içerikte sıfır görünüm CSS'i
     (renk, font, zemin, STYLE bloğu ve satır-içi stil niteliği YOK). Görünüm sahibi Blogger teması.
     Yalnız anlamsal HTML. Başlık tema tarafından basılır; BAŞLIK (Blogger BAŞLIK ALANI):
     "NVIDIA Nemotron 3.5 Super VL'i açtı: bağımsız ölçümde açık ağırlık sınıfının en hızlı ajan modeli, zekâ endeksinde lig liderlerinin gerisinde" --&gt;
&lt;p&gt;NVIDIA 15 Ekim'de 120 milyar parametreli açık ağırlıklı bir modeli ağa bıraktı ve kendi duyurusunda sınıfının en iyi çalışanından beş kat hızlı olduğunu söyledi. Bağımsız ölçüm kuruluşu Artificial Analysis bir hafta boyunca kendi enstrümanlarıyla baktı ve hız defterini doğruladı: yirmi yedi ölçülen model arasında birincilik. Aynı kuruluşun zekâ endeksi ise aynı modelde başka bir şey söylüyor: 26 puan — kendi açık ağırlık sınıfının birincisi, ama ligin zirvesinin belirgin biçimde gerisi. Yazının hikâyesi bu bilanço: zekâ tahtında değil, hız, fiyat ve lisans temizliği üzerine kurulu bir üçgenin hikâyesi.&lt;/p&gt;

&lt;!-- SLOT:ICINDEKILER --&gt;
&lt;p&gt;&lt;strong&gt;Bu Yazıda&lt;/strong&gt;&lt;/p&gt;
&lt;ol&gt;
  &lt;li&gt;&lt;a href="#bol-1"&gt;Duyurudan bir gün önce inen model&lt;/a&gt;&lt;/li&gt;
  &lt;li&gt;&lt;a href="#bol-2"&gt;Tek model, iki iş&lt;/a&gt;&lt;/li&gt;
  &lt;li&gt;&lt;a href="#bol-3"&gt;Bağımsız karne: hızda kral, zekâda klasman birincisi&lt;/a&gt;&lt;/li&gt;
  &lt;li&gt;&lt;a href="#bol-4"&gt;NVIDIA'nın tablosu: özelleşmiş zeminde iddialar&lt;/a&gt;&lt;/li&gt;
  &lt;li&gt;&lt;a href="#bol-5"&gt;Kendi sunucunuzda: yol var, bedeli var&lt;/a&gt;&lt;/li&gt;
  &lt;li&gt;&lt;a href="#ka-yorum"&gt;Murat Karakaya Akademi Yorumu&lt;/a&gt;&lt;/li&gt;
  &lt;li&gt;&lt;a href="#kaynakca"&gt;Kaynakça&lt;/a&gt;&lt;/li&gt;
&lt;/ol&gt;
&lt;!-- /SLOT:ICINDEKILER --&gt;

&lt;!-- SLOT:GOVDE --&gt;
&lt;h2 id="bol-1"&gt;1. Duyurudan bir gün önce inen model&lt;/h2&gt;
&lt;p&gt;İşin habercilik tarafıyla başlayalım, çünkü takvim ilginç çalıştı. NVIDIA duyurusu 15 Ekim sabahı Kari Briski imzasıyla yayımlandı: Nemotron 3.5 Super VL açık ağırlıklı multimodal ajan modeli, aynı gün ailenin yerel üyesi Bolt 74B-A9B, bugüne planlanmış küçük kardeş Bolt E4B [1]. OrcaRouter'ın kaydettiğine göre Ollama kataloğu checkpoint'i duyurudan bir gün önce, 14 Ekim'de (v0.40.5) listelemişti bile [14]. Modelin varlığı, şirket ağzından konuşulmadan yirmi dört saat önce dağıtım kanalında görünür hale gelmişti.&lt;/p&gt;
&lt;p&gt;Bu tekil bir kaza değil, yeni bir zamanlama rejimi. VentureBeat'in lansman haftası bağlamında andığı NVIDIA–Hugging Face alım anlaşması (yaklaşık 12,93 milyar dolarlık anlaşma, kapanışın 2027'nin ilk yarısında beklenmesi) açık model dağıtımının saatler içinde döndüğü bir savaşın işaret fişekleri [10]. Kurumsal pazara ürün satan bir laboratuvarın, modelini kendi duyurusundan önce üçüncü taraf bir katalogda basılı görmek zorunda kalması, o savaşın maliyeti. Duyuru blogunda geçen "Nemotron Coalition" işbirliğinin üye listesi yayımlanmış değil; ortakların kimliği ve kapsamı doğrulanamadı [1].&lt;/p&gt;
&lt;p&gt;Aile konumlandırması net: Nemotron 3 Ultra en zor muhakeme işleri, Lightning yüksek hacim, Super VL multimodal ajan, Bolt yerel donanım [1][2]. Super VL bu kademenin ilk çok modlu üyesi; sesi yok, karşılığında metin, görsel, belge ve NVIDIA'nın beyanıyla video girişi var [2][6].&lt;/p&gt;

&lt;h2 id="bol-2"&gt;2. Tek model, iki iş&lt;/h2&gt;
&lt;p&gt;Mimari, lansmanın en az pazarlama kokan tarafı. 120B toplam, 12B aktif parametreli bir MoE: 512 yönlendirilmiş uzman, top-22 seçimi ve bir paylaşımlı uzman; 88 katmanın 40'ı Mamba-2, 8'i attention, 40'ı MoE [2][6]. Görü tarafını C-RADIO (RADIO v2.5-H ViT-Huge) kodlayıcısı taşıyor; hızın kaynaklarından biri modele gömülü Multi-Token Prediction spekülatif çözümlemesi [2][6]. Pratik karşılığı şu: eskiden ayrı bir belge-okuma modeli (VLM) artı ayrı bir ajan modeli kurup ikisini birbirine bağlamak gerekirdi; NVIDIA tek ağırlık setinde ikisini de yapacağını söylüyor.&lt;/p&gt;
&lt;p&gt;Damıtma tarafında şirket, modeli ailenin en büyük üyesi Nemotron 3.5 Ultra'dan eğittiğini ve beşte bir parametreyle onunkine yaklaşan doğruluk aldığını savunuyor [1][2]. Eğitim defteri de NVIDIA rakamları: yaklaşık 26 trilyon token ön eğitim, 3,8 milyon örneklik son eğitim; araştırma raporunun hakemlik durumu ve bu rakamların bağımsız tekrarı yok [5]. Sayıları şirketin verdiği gibi, şirketin verdiği yere yazıyoruz.&lt;/p&gt;

&lt;h2 id="bol-3"&gt;3. Bağımsız karne: hızda kral, zekâda klasman birincisi&lt;/h2&gt;
&lt;p&gt;Artificial Analysis'ın 23.10 erişimli sayıları iki farklı tablo çiziyor ve ikisi de doğru:&lt;/p&gt;
&lt;table&gt;
  &lt;thead&gt;&lt;tr&gt;&lt;th&gt;Ölçüm&lt;/th&gt;&lt;th&gt;Değer&lt;/th&gt;&lt;th&gt;Sınıf karşılığı&lt;/th&gt;&lt;/tr&gt;&lt;/thead&gt;
  &lt;tbody&gt;
    &lt;tr&gt;&lt;td&gt;Intelligence Index&lt;/td&gt;&lt;td&gt;26&lt;/td&gt;&lt;td&gt;Açık ağırlık sınıfında 1/67; sınıf medyanı 8&lt;/td&gt;&lt;/tr&gt;
    &lt;tr&gt;&lt;td&gt;Çıktı hızı&lt;/td&gt;&lt;td&gt;429,9 tok/s&lt;/td&gt;&lt;td&gt;Ölçülen 27 modelde 1/27; medyan 92&lt;/td&gt;&lt;/tr&gt;
    &lt;tr&gt;&lt;td&gt;Görev başına maliyet&lt;/td&gt;&lt;td&gt;~$0,19&lt;/td&gt;&lt;td&gt;Girdi $0,23 / çıktı $0,57 (1M token)&lt;/td&gt;&lt;/tr&gt;
    &lt;tr&gt;&lt;td&gt;Lisans&lt;/td&gt;&lt;td&gt;OpenMDW-1.1&lt;/td&gt;&lt;td&gt;Ticari kullanım, fine-tune, yeniden dağıtım serbest&lt;/td&gt;&lt;/tr&gt;
  &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;&lt;em&gt;Tablo: Artificial Analysis model sayfası [9] · erişim 23.10.2026.&lt;/em&gt;&lt;/p&gt;
&lt;p&gt;Hız birinciliği bağımsız: NVIDIA'nın "5 kat hızlı" cümlesinin referansı kendi seçtiği rakip (Inkling Small) ve o kesin oran tekrarlanmadı, ama 429,9 tok/s ile sınıf açık ara birinciliği üçüncü taraftan ölçülmüş durumda [1][9]. Zekâ tarafında ölçeği geçen haftadan vermek lazım: aynı AA sayfası ailesinde MiniMax M3.1 Flash 43, GLM-5.3-Flash 42 puanla ölçülmüştü; Super VL'nin 26'sı "açık ağırlık sınıfının birincisi" unvanını kendi açıklık ve boyut profiline göre alıyor, ham akılda o ikisinin çok gerisinde [9]. AA'nın bir notu daha var: model konuşkan — endeks değerlendirmesinde sınıf medyanının epey üstünde çıktı üretmiş, bu da fiyata gecikme olarak yansır [9].&lt;/p&gt;

&lt;h2 id="bol-4"&gt;4. NVIDIA'nın tablosu: özelleşmiş zeminde iddialar&lt;/h2&gt;
&lt;p&gt;Kendi model kartındaki benchmark seti (NVIDIA'nın NeMo Gym ile ölçtüğü tablo) etkileyici başlıklar taşıyor: GPQA-Diamond 89,52; LiveCodeBench v6 88,22; HLE metin-only 31,79; DeepSWE 24,65; OSWorld-Verified 51,14 [3]. Tablonun gösterişsiz köşeleri de var ve onlar daha çok şey anlatıyor: Terminal-Bench 4 [offline] skoru 1 — uç bir zayıflık — ve NVIDIA'nın kendi AA-Omniscience ölçümünde 26,2 doğruluk yanında yüzde 30'luk halüsinasyon riski [3]. Çok modlu puanların (Video-MME v2 32,85 gibi) tamamı şirket içi ölçüm; bağımsız multimodal kıyaslama henüz ortada yok [3].&lt;/p&gt;
&lt;p&gt;Lansman haftasının en çok dolaşan cümlesi şuydu: Super VL, özelleştirilmiş post-train haliyle belgeli beş dar görevde Claude Opus 5 ve GPT-6.1 Sol'u geçiyor. VentureBeat'in aktardığı grafiklere göre tıbbi-iddia görevinde baz modelin yüzde 28'i yüzde 72'ye, güvenlik görevinde yüzde 35'i yüzde 76'ya çıkıyor; rakip bantları Opus 5 için yüzde 38, GPT-6.1 Sol için yüzde 66 civarında [10]. Üç not düşmek borç: grafikler NVIDIA'nın kendi grafiği, bağımsız tekrar yok; üstünlük yalnız bu özelleşmiş görevlerde; ve kıyaslanan şey ham model değil, işe göre son-eğitilmiş hali [10]. Aynı ailenin Hermes Bench 63,44 ve LangChain 87,05 puanlarıyla "kendinden büyük Ultra'yı geçiyor" cümlesi de aynı kategoride: şirketin kendi değerlendirmesi [2].&lt;/p&gt;
&lt;p&gt;The New Stack'in aktardığı beşli karşılaştırma daha dengeli bir fotoğraf veriyor: Qwen3.5-122B-A10B'ye karşı GPQA'da üstün, Terminal-Bench 2.1'de 66,7'ye 48,7 açık fark — ama SWE-bench Verified'da 71,1'e 71,9 geride, IFBench ve MMLongBench-Doc'ta da biraz geride [11].&lt;/p&gt;

&lt;h2 id="bol-5"&gt;5. Kendi sunucunuzda: yol var, bedeli var&lt;/h2&gt;
&lt;p&gt;Lisans tarafı sınıfın en temizlerinden: OpenMDW-1.1 ticari kullanıma, ince ayara ve yeniden dağıtıma izin veriyor; model kartı "ticari ya da ticari olmayan kullanım" diyor [3][7][11]. Yerel yol gerçekten var: Ollama'da &lt;code&gt;ollama run nemotron-3.5-super&lt;/code&gt; komutu vision/tools/thinking etiketleriyle çalışıyor, vLLM 0.31 ile Day-0 desteği yayımlandı, HF'de BF16/NVFP4 checkpoint'leri duruyor [8][12][3][4]. Lansman haftası boyunca yayılan topluluk yürüyüşleri — OpenCode'da uygulama demosundan DGX Spark üstüne vLLM kurulumuna — kurulum adımlarının zaten standartlaştığını gösteriyor [15][17]. Bedeli bellek: BF16 yaklaşık 227GB; pratik Q4 indirmesi 80-90GB; en agresif 1-bit GGUF 46,1GB'a kadar iniyor ama Unsloth en az 64GB birleşik bellek öneriyor ve kalite ödününü not düşüyor [6][13]. DGX Spark sahipleri için NVFP4 checkpoint tek makinede multimodal ajan kurmanın en temiz yolu olarak konumlanmış [1][11].&lt;/p&gt;
&lt;p&gt;İki şerh. Video girişi NVIDIA'nın beyanı: model kartı ve NIM belgesi videoyu sayıyor, Ollama kataloğunun etiketlerinde yalnızca "vision" var, video gösterimi ise llama.cpp mtmd CLI rehberinde (Ollama'da değil) görünüyor [6][8][13]. AA'nın mod listesi de "text+image" diyor — AA'nın multimodal kapsamı geriden geliyor olabilir; güvenli cümle "video NVIDIA'nın beyanı" [9]. Bolt tarafında da aynı terazi: "Qwen'i geçti" sayıları NVIDIA'nın kendi testleri; küçük kardeş Bolt E4B'ın resmî tarihi bugün (23 Ekim) ve bu yazı hazırlanırken (23.10 saat ~10:00) resmî kaynaklarda "yakında" satırından öte bir çıkış teyidi yoktu [1][11]. Günlük tazelik kontrolü: AA ve Ollama sayfaları yazı bitmeden yeniden yoklandı; hız, endeks ve fiyat sayıları aynı duruyor.&lt;/p&gt;
&lt;p&gt;Yarın için ölçülü bir test planı: kendi faturanız, sözleşmeniz ya da ekran görüntüyle OCR + tablo çıkarımı; çok adımlı araç çağrısı; 1M bağlamda uzun bir PDF. Yanına sınıfın öbür ucundan bir modeli koyup görev başına maliyeti karşılaştırın — ~$0,19/görev buradaki gerçek iddia [9]. OpenRouter'daki ücretsiz katman gözlemi bir video kaydından aktarılan bir gözlem; bugün için doğrulanmış değil, planınızı ona göre yapın [16].&lt;/p&gt;
&lt;!-- /SLOT:GOVDE --&gt;

&lt;!-- SLOT:KA_YORUM --&gt;
&lt;h2 id="ka-yorum"&gt;Murat Karakaya Akademi Yorumu&lt;/h2&gt;
&lt;p&gt;Verilerin söylediği: bağımsız ölçüm bu modeli iki ayak üstünde net tutuyor — açık ağırlık sınıfında hız birinciliği (medyan 92'ye karşılık 429,9 tok/s) ve görev başına ~$0,19 ile ligin en ucuz işçilerinden biri [9]. Aynı ölçüm sınırı da çiziyor: 26 Intelligence Index, kendi sınıfının medyanının (8) üç katından fazla olsa da ham zekâ liginin 40+ skorlu modellerinin ligine ait değil; NVIDIA'nın kendi kartındaki yüzde 30'luk halüsinasyon riski ve Terminal-Bench'teki uçuk zayıflık, doğrulama katmanı olmayan iş akışlarına bu modeli koymayı akılsızlaştırıyor [3][9]. Rakip geçme grafikleri şirketin kendi grafiği; biz yazı boyunca bağımsız olanı şirkete bırakılanı ayırdık.&lt;/p&gt;
&lt;p&gt;Kanaatimize gelince: NVIDIA açık model liginde zekâ maratonu koşmuyor, altyapı satıyor. Kendi sunucunuzda çalışan, belgelerinizi okuyan, lisansı temiz, ucuz ve ölçülebilir biçimde hızlı bir işçi — ve duyurusundan bir gün önce katalogda belirmesi, açık dağıtımın artık saat savaşına döndüğünün ilanı. Bu modelin gerçek sınavı akıllı soru çözmek değil; sizin deponuzda, sizin donanımınızda, sizin PDF'lerinizle yapılacak üç test. Üretici iddialarının aksine zekâ liderliği bekleyerek gelmeyin; belge-ağır, ajan-ağır işlerde fiyat-hız hesabıyla gelin.&lt;/p&gt;
&lt;!-- /SLOT:KA_YORUM --&gt;

&lt;!-- SLOT:HASHTAG — düz metin şerit; BOŞLUKLU kelimeler (CEO 23.10 kuralı); CEO bunları Blogger
     etiket (labels) alanına taşır: ilki her zaman marka --&gt;
&lt;p&gt;Etiketler: Murat Karakaya Akademi, NVIDIA Nemotron, Nemotron 3.5 Super VL, Açık Ağırlık Modeller, Yerel Yapay Zeka, Ollama&lt;/p&gt;
&lt;!-- /SLOT:HASHTAG --&gt;

&lt;!-- SLOT:KAYNAKCA --&gt;
&lt;h2 id="kaynakca"&gt;Kaynakça&lt;/h2&gt;
&lt;ol&gt;
  &lt;li&gt;blogs.nvidia.com — Kari Briski, &lt;a href="https://blogs.nvidia.com/blog/nemotron-3-5-super-vl/"&gt;"Open From Data Center to Desktop: NVIDIA Expands Nemotron Family With Highest Efficiency Super VL and Bolt Models"&lt;/a&gt; (15.10.2026) · erişim: 23.10.2026&lt;/li&gt;
  &lt;li&gt;developer.nvidia.com — &lt;a href="https://developer.nvidia.com/blog/nvidia-nemotron-3-5-super-vl-powers-specialized-agents-with-multimodal-reasoning/"&gt;"NVIDIA Nemotron 3.5 Super VL Powers Specialized Agents With Multimodal Reasoning"&lt;/a&gt; (teknik blog: mimari, benchmark, damıtma) · erişim: 23.10.2026&lt;/li&gt;
  &lt;li&gt;huggingface.co — &lt;a href="https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16"&gt;NVIDIA Nemotron 3.5 Super VL 120B-A12B BF16 model kartı&lt;/a&gt; (OpenMDW-1.1, NeMo Gym tablosu) · erişim: 23.10.2026&lt;/li&gt;
  &lt;li&gt;huggingface.co — &lt;a href="https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4"&gt;NVFP4 model kartı&lt;/a&gt; (niceliklendirme puan koruması) · erişim: 23.10.2026&lt;/li&gt;
  &lt;li&gt;research.nvidia.com — &lt;a href="https://research.nvidia.com/labs/nemotron/Nemotron-3.5-Super-and-Lightning/"&gt;Nemotron 3.5 Super and Lightning araştırma raporu sayfası&lt;/a&gt; · erişim: 23.10.2026&lt;/li&gt;
  &lt;li&gt;docs.nvidia.com — &lt;a href="https://docs.nvidia.com/nemo/automodel/nightly/recipes-e2e-examples/nemotron-3-5-super-vl"&gt;NeMo AutoModel Nemotron 3.5 Super VL rehberi&lt;/a&gt; (mimari ayrıntıları, ~227GB BF16; NIM get-started: vLLM 0.31, text+image+video girişi, ses yok) · erişim: 23.10.2026&lt;/li&gt;
  &lt;li&gt;build.nvidia.com — &lt;a href="https://build.nvidia.com/nvidia/nemotron-3.5-super-vl-120b-a12b/modelcard"&gt;Nemotron 3.5 Super VL model kartı&lt;/a&gt; (1M bağlam, MTP, ticari kullanım) · erişim: 23.10.2026&lt;/li&gt;
  &lt;li&gt;ollama.com — &lt;a href="https://ollama.com/library/nemotron-3.5-super"&gt;"nemotron-3.5-super" katalog sayfası&lt;/a&gt; (komut, vision/tools/thinking etiketleri, 1M bağlam uyarısı, lisans) · erişim: 23.10.2026&lt;/li&gt;
  &lt;li&gt;artificialanalysis.ai — &lt;a href="https://artificialanalysis.ai/models/nemotron-3-5-super-vl"&gt;"Nemotron 3.5 Super VL" model sayfası&lt;/a&gt; (Index 26, açık ağırlık 1/67; hız 429,9 tok/s 1/27; ~$0,19/görev; 1M ctx; text+image) · erişim: 23.10.2026&lt;/li&gt;
  &lt;li&gt;venturebeat.com — &lt;a href="https://venturebeat.com/technology/nvidias-new-open-nemotron-3-5-super-vl-with-post-training-beats-claude-opus-5-gpt-6-1-sol-on-specialized-task-benchmarks"&gt;"Nvidia's new open Nemotron 3.5 Super VL with post-training beats Claude Opus 5, GPT-6.1 Sol on specialized task benchmarks"&lt;/a&gt; (beş uzmanlaşmış post-train kıyası; NVIDIA grafikleri; HF alım bağlamı) · erişim: 23.10.2026&lt;/li&gt;
  &lt;li&gt;thenewstack.io — Jenny Morris, &lt;a href="https://thenewstack.io/nvidia-nemotron-bolt-laptop/"&gt;NVIDIA Nemotron/Bolt analizi&lt;/a&gt; (15.10.2026; Qwen3.5-122B-A10B karşılaştırması, Bolt testleri "Nvidia says", OpenMDW teyidi) · erişim: 23.10.2026&lt;/li&gt;
  &lt;li&gt;vllm.ai — &lt;a href="https://vllm.ai/blog/2026-10-15-nemotron-3-5-super-vl-vllm"&gt;"Nemotron 3.5 Super VL in vLLM"&lt;/a&gt; Day-0 blog (v0.31.0, NVFP4 GB200, reasoning-parser + MTP) · erişim: 23.10.2026&lt;/li&gt;
  &lt;li&gt;unsloth.ai — &lt;a href="https://unsloth.ai/docs/models/nemotron-3.5-super-vl"&gt;Nemotron 3.5 Super VL yerel çalışma rehberi&lt;/a&gt; (GGUF boyutları, 1-bit 46,1GB, 64GB+ önerisi, llama.cpp mtmd CLI ile görsel+video) · erişim: 23.10.2026&lt;/li&gt;
  &lt;li&gt;orcarouter.ai — &lt;a href="https://www.orcarouter.ai/blog/nemotron-3-5-super-ollama-local"&gt;"Nemotron 3.5 Super: Ollama local" analizi&lt;/a&gt; (checkpoint'in 14.10'da, duyurudan bir gün önce katalogda belirmesi; llama.cpp runarı; veri merkezi önerisi 4×H100 / 2×GB200) · erişim: 23.10.2026&lt;/li&gt;
  &lt;li&gt;youtube.com — Sanny-Recommends, &lt;a href="https://www.youtube.com/watch?v=khuGTBvaCfc"&gt;"NVIDIA Nemotron 3.5 Super VL Is INSANE!"&lt;/a&gt; (16.10.2026; OpenCode demo, OpenRouter/Ollama/LM Studio yolları) · erişim: 23.10.2026&lt;/li&gt;
  &lt;li&gt;youtube.com — Outright Systems, &lt;a href="https://www.youtube.com/watch?v=9hG8xN7BGSI"&gt;Nemotron ailesi özeti&lt;/a&gt; (21.10.2026; OpenRouter ücretsiz katman gözlemi; &lt;code&gt;ollama run nemotron-3.5-bolt&lt;/code&gt;) · erişim: 23.10.2026&lt;/li&gt;
  &lt;li&gt;youtube.com — Skillversitas, &lt;a href="https://www.youtube.com/watch?v=3Ki0lNFVi6g"&gt;DGX Spark + vLLM kurulum yürüyüşü&lt;/a&gt; (16.10.2026) · erişim: 23.10.2026&lt;/li&gt;
&lt;/ol&gt;
&lt;!-- /SLOT:KAYNAKCA --&gt;

&lt;!-- SLOT:HASHTAG-BLOK — Madde 11 (bileşik hashtag; Etiketler şeridinden ayrı) --&gt;
&lt;section class="hashtag" aria-label="Etiketler"&gt;
  &lt;a href="#"&gt;#MuratKarakayaAkademi&lt;/a&gt;
  &lt;a href="#"&gt;#NVIDIA&lt;/a&gt;
  &lt;a href="#"&gt;#Nemotron&lt;/a&gt;
  &lt;a href="#"&gt;#SuperVL&lt;/a&gt;
  &lt;a href="#"&gt;#AcikAgirlik&lt;/a&gt;
  &lt;a href="#"&gt;#YapayZeka&lt;/a&gt;
&lt;/section&gt;
&lt;!-- /SLOT:HASHTAG-BLOK --&gt;

&lt;!-- SLOT:FOOTER — IMZA (tek satır, düz metin): --&gt;
&lt;footer&gt;
  &lt;p&gt;Murat Karakaya Akademi · muratkarakaya.net&lt;/p&gt;
&lt;/footer&gt;
&lt;!-- /SLOT:FOOTER --&gt;</description><link>https://www.muratkarakaya.net/2026/10/nvidia-nemotron-35-super-vli-act-bagmsz.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-7619481239420318852</guid><pubDate>Sun, 25 Oct 2026 21:47:04 +0000</pubDate><atom:updated>2026-10-28T02:11:17.279+03:00</atom:updated><title>YouTube AI Haftalık Trend Raporu (18 → 25 Ekim 2026)</title><description>&lt;!doctype html&gt;&lt;html lang="tr"&gt;&lt;head&gt;&lt;meta charset="utf-8"&gt;
&lt;meta name="viewport" content="width=device-width,initial-scale=1"&gt;
&lt;title&gt;YouTube AI Haftalık Trend Raporu&lt;/title&gt;&lt;style&gt;
:root{--bg:transparent;--sf:#fff;--sf2:#eef1f4;--ln:#d7dce2;--ink:#111;--mu:#555;
--ac:#00796b;--am:#b45309;--mono:ui-monospace,Consolas,monospace}
*{box-sizing:border-box;margin:0;padding:0}
body{background:var(--bg);color:var(--ink);font:15px/1.5 -apple-system,'Segoe UI',Roboto,sans-serif;padding:28px 24px 60px}
.wrap{max-width:1100px}
h1{font-size:32px;letter-spacing:.3px;text-align:center;margin-top:8px} .datebar{text-align:center;margin:12px 0 6px}
.dtc{display:inline-block;font:600 15px/1 var(--mono);color:#0b0e12;background:#35d0ba;padding:7px 16px;border-radius:999px}
.meta{color:var(--mu);font-family:var(--mono);font-size:13.5px;margin-top:2px;text-align:center}
.dseg{cursor:pointer;transition:stroke-width .12s} .dseg:hover{stroke-width:36}
#tip{position:fixed;left:0;top:0;z-index:9;pointer-events:none;background:#f6f7f8;border:1px solid #c9d0d8;color:#111;font:12.5px/1.45 var(--mono);padding:7px 11px;border-radius:8px;opacity:0;transition:opacity .12s;max-width:340px;box-shadow:0 6px 20px rgba(0,0,0,.15)}
.meta b{color:var(--ac);font-weight:600}
nav{margin:14px 0 26px;display:flex;flex-wrap:wrap;gap:8px}
nav a{color:var(--mu);text-decoration:none;font-size:12px;padding:4px 10px;border:1px solid var(--ln);border-radius:999px;line-height:1.35} nav a b{color:var(--ac);font-family:var(--mono);font-weight:600;margin-left:4px}
nav a:hover{color:var(--ink);border-color:var(--ac)}
.card{background:var(--sf);border:1px solid var(--ln);border-radius:12px;padding:16px 18px}
.card h3,.stats h3{font-size:12px;text-transform:uppercase;letter-spacing:1.2px;color:var(--mu);margin-bottom:10px;font-weight:600}
.top{display:grid;grid-template-columns:2fr 1fr;gap:14px;margin-bottom:14px}
.champ{background:linear-gradient(135deg,#f7f9fa,#eef2f5);border:1px solid #d7dce2;border-radius:12px;padding:20px;text-decoration:none;color:var(--ink);display:block}
.champ:hover{border-color:var(--am)}
.ck{font-size:11px;letter-spacing:1.5px;text-transform:uppercase;color:var(--am)}
.cv{font:600 44px/1.1 var(--mono);color:var(--am);margin:6px 0}
.ct{font-size:17px;font-weight:600} .cs{color:var(--mu);font-size:13px;margin-top:4px}
.cd{color:#444;font-size:12.5px;line-height:1.55;margin-top:10px;border-top:1px solid var(--ln);padding-top:10px}
.stats{background:var(--sf);border:1px solid var(--ln);border-radius:12px;padding:14px 18px;display:flex;flex-direction:column;justify-content:space-between}
.st{display:flex;justify-content:space-between;align-items:baseline;padding:5px 0;border-bottom:1px solid var(--ln)}
.st:last-child{border:0} .st b{font:600 19px var(--mono)} .st span{color:var(--mu);font-size:12.5px}
.credit{text-align:center;margin-top:14px;font:600 14px var(--sans);color:#333;letter-spacing:.2px}
.credit b{color:var(--am);font-weight:800}
.rowg{display:grid;gap:14px;margin-bottom:14px}
.g23{grid-template-columns:2fr 1fr} .g11{grid-template-columns:1.2fr 1fr}
@media(max-width:820px){.top,.g23,.g11{grid-template-columns:1fr}}
svg{width:100%;height:auto;display:block}
.ax{font:11px var(--mono);fill:#555} .bn2{font:600 11px var(--mono);fill:#333}
.dn{font:700 30px var(--mono);fill:#111}
.donut{max-width:220px;margin:0 auto}
.lgs{display:flex;flex-wrap:wrap;gap:6px 14px;margin-top:12px}
.lg{font-size:11.5px;color:var(--mu)} .lg i{display:inline-block;width:9px;height:9px;border-radius:2px;margin-right:5px} .lg b{color:#111;font-family:var(--mono)}
.cbar{display:grid;grid-template-columns:130px 1fr 90px;gap:10px;align-items:center;padding:7px 0;border-bottom:1px solid var(--ln)}
.cbar:last-child{border:0} .cn{font-size:12.5px;color:#333;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
.ct2{background:var(--sf2);height:10px;border-radius:4px;overflow:hidden} .cf{display:block;height:100%;background:var(--ac);opacity:.85;border-radius:4px}
.cv2{font:600 12.5px var(--mono);text-align:right} .cv2 i{font-style:normal;color:var(--mu);font-size:10.5px;margin-left:6px}
h2.sec{font-size:13px;text-transform:uppercase;letter-spacing:1.4px;color:var(--mu);margin:26px 0 10px}
ol.heads{list-style:none;display:flex;flex-direction:column;gap:8px}
ol.heads li{display:flex;gap:14px;background:var(--sf);border:1px solid var(--ln);border-radius:10px;padding:13px 16px}
ol.heads .hr{font:700 20px var(--mono);color:var(--ac);min-width:26px}
ol.heads li.h2 .hr{color:#0369a1} ol.heads li.h3 .hr{color:#475569}
ol.heads li.h1{border-color:#bfe3dd}
ol.heads p{font-size:14.5px} ol.heads li.h1 p{font-size:16px;font-weight:600} ol.heads li.h3 p{color:var(--mu)}
.bars{display:flex;flex-direction:column;gap:4px}
.bar{display:grid;grid-template-columns:190px 1fr 30px;gap:12px;align-items:center;text-decoration:none;color:var(--ink);padding:4px 0}
.bar:hover .bn{color:var(--ac)} .bn{font-size:12.5px;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
.bt{background:var(--sf2);border-radius:4px;height:10px;overflow:hidden} .bf{display:block;height:100%;opacity:.9;border-radius:4px}
.bv{font:600 12.5px var(--mono);text-align:right;color:var(--mu)}
@media(max-width:640px){.bar{grid-template-columns:130px 1fr 28px}}
.theme{margin-top:30px} .theme h2{font-size:17px;display:flex;align-items:baseline;gap:10px}
.tn{font:700 15px var(--mono)} .dot{width:9px;height:9px;border-radius:3px;display:inline-block}
.cnt{font:400 12.5px var(--mono);color:var(--mu);margin-left:auto}
.insight{color:#444;font-size:13.5px;margin:6px 0 10px;padding-left:2px;max-width:860px}
.rows{border:1px solid var(--ln);border-radius:10px;overflow:hidden}
.row{display:grid;grid-template-columns:64px 56px 150px 1fr;gap:12px;align-items:baseline;padding:9px 14px;border-bottom:1px solid var(--ln);text-decoration:none;color:var(--ink);background:var(--sf)}
.row:last-child{border:0} .row:hover{background:var(--sf2)}
.vw{font:600 12.5px var(--mono);text-align:right}
.v1{color:var(--am)} .v2{color:var(--ac)} .v3{color:var(--mu)}
.dt{font:12px var(--mono);color:var(--mu)} .ch{font-size:12px;color:#2c5f8a;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
.tt{font-size:13.5px} .other .tt{color:var(--mu)}
footer{margin-top:34px;color:#555;font:12px/1.7 var(--mono);border-top:1px solid var(--ln);padding-top:12px}
&lt;/style&gt;&lt;/head&gt;&lt;body&gt;&lt;div class="wrap"&gt;
&lt;h1&gt;YouTube AI Haftalık Trend Raporu&lt;/h1&gt;
&lt;div class="datebar"&gt;&lt;span class="dtc"&gt;2026-10-18 &amp;rarr; 2026-10-25&lt;/span&gt;&lt;/div&gt;
&lt;div class="credit"&gt;Bu veriler &lt;b&gt;Murat Karakaya Akademi&lt;/b&gt; tarafından derlenmiştir.&lt;/div&gt;

&lt;div class="meta"&gt;&lt;b&gt;24&lt;/b&gt; aktif kanal / 43 &amp;nbsp;|&amp;nbsp; &lt;b&gt;122&lt;/b&gt; video &amp;nbsp;|&amp;nbsp; &lt;b&gt;3.1M&lt;/b&gt; toplam izlenme&lt;/div&gt;
&lt;h2 class="sec"&gt;Haftanın 3 manşeti&lt;/h2&gt;&lt;ol class="heads"&gt;&lt;li class="h1"&gt;&lt;span class="hr"&gt;1&lt;/span&gt;&lt;p&gt;Yerel yapay zekanın gizlilik yanılsaması haftaya damga vurdu: ~8 dolarlık GPU süresiyle eğitilen arka kapılı bir model, Codex CLI üzerinden projenin .env dosyasını yabancı bir sunucuya sızdırdı; video 400 bine yakın izlenmeye ulaştı.&lt;/p&gt;&lt;/li&gt;&lt;li class="h2"&gt;&lt;span class="hr"&gt;2&lt;/span&gt;&lt;p&gt;Açık modeller fiyat/performansta öne geçti: Kimi K3.1, Claude Opus 5.5&amp;#x27;i neredeyse yakalarken DeepSeek günde 155 trilyon token üreten kendi çıkarım motorunu açık kaynak yaptı.&lt;/p&gt;&lt;/li&gt;&lt;li class="h3"&gt;&lt;span class="hr"&gt;3&lt;/span&gt;&lt;p&gt;Sesli ajanlar olgunlaşıyor: Dialt lansmanı, Gemini 3.8 Flash TTS&amp;#x27;in ses tasarımı/çoğaltma özelliği ve Block&amp;#x27;un insan sesiyle eğitilen Dessa&amp;#x27;sı sesli yapay zekayı ciddi bir üretim hattına dönüştürüyor.&lt;/p&gt;&lt;/li&gt;&lt;/ol&gt;
&lt;h2 class="sec"&gt;Haftanın temaları&lt;/h2&gt;&lt;nav&gt;&lt;a href="#t1" style="font-size:12.9px"&gt;Ajan ve Sistem Güvenliği&lt;b&gt;6&lt;/b&gt;&lt;/a&gt; &lt;a href="#t2" style="font-size:17.0px"&gt;Sesli Yapay Zeka ve Sesli Ajanlar&lt;b&gt;17&lt;/b&gt;&lt;/a&gt; &lt;a href="#t3" style="font-size:12.5px"&gt;Görüntü ve Video Üretimi&lt;b&gt;5&lt;/b&gt;&lt;/a&gt; &lt;a href="#t4" style="font-size:12.9px"&gt;Belge Anlama ve OCR&lt;b&gt;6&lt;/b&gt;&lt;/a&gt; &lt;a href="#t5" style="font-size:13.6px"&gt;Kimi ve Moonshot&lt;b&gt;8&lt;/b&gt;&lt;/a&gt; &lt;a href="#t6" style="font-size:13.2px"&gt;DeepSeek&lt;b&gt;7&lt;/b&gt;&lt;/a&gt; &lt;a href="#t7" style="font-size:12.9px"&gt;Gemini ve Google Modelleri&lt;b&gt;6&lt;/b&gt;&lt;/a&gt; &lt;a href="#t8" style="font-size:11.0px"&gt;Difüzyon Dil Modelleri&lt;b&gt;1&lt;/b&gt;&lt;/a&gt; &lt;a href="#t9" style="font-size:14.8px"&gt;Yerel Yapay Zeka ve Gizlilik&lt;b&gt;11&lt;/b&gt;&lt;/a&gt; &lt;a href="#t10" style="font-size:13.6px"&gt;Sam Altman ve OpenAI Gündemi&lt;b&gt;8&lt;/b&gt;&lt;/a&gt; &lt;a href="#t11" style="font-size:16.2px"&gt;İş Dünyasında Yapay Zeka&lt;b&gt;15&lt;/b&gt;&lt;/a&gt; &lt;a href="#t12" style="font-size:11.8px"&gt;Açık Modeller ve Verimlilik&lt;b&gt;3&lt;/b&gt;&lt;/a&gt; &lt;a href="#t13" style="font-size:15.5px"&gt;Kodlama Ajanları ve Ajan Altyapısı&lt;b&gt;13&lt;/b&gt;&lt;/a&gt; &lt;a href="#t14" style="font-size:14.8px"&gt;Öğrenme, Bilim ve Araştırma&lt;b&gt;11&lt;/b&gt;&lt;/a&gt; &lt;a href="#t15" style="font-size:12.5px"&gt;Diğer Öne Çıkanlar&lt;b&gt;5&lt;/b&gt;&lt;/a&gt;&lt;/nav&gt;
&lt;div class="top"&gt;&lt;a class="champ" href="https://www.youtube.com/watch?v=3mx84MRj8lY" target="_blank" rel="noopener"&gt;&lt;div class="ck"&gt;Haftanın rekoru&lt;/div&gt;&lt;div class="cv"&gt;397K&lt;/div&gt;&lt;div class="ct"&gt;Your Local AI Is Not as Private as You Think&lt;/div&gt;&lt;div class="cs"&gt;Kai &amp;middot; 21 Eki&lt;/div&gt;&lt;div class="cd"&gt;Yerel yapay zeka sandığınız kadar gizli değil: OpenAI&amp;#x27;ın Codex CLI&amp;#x27;sında test edilen, yaklaşık 8 dolarlık GPU süresiyle eğitilmiş gizli arka kapılı bir Qwen 2.5 7B modeli, 50 temiz testin tamamını geçtikten sonra sadece iki kelimeyle (&amp;quot;bonsoir, Elliot&amp;quot;) projenin .env dosyasını yabancı bir sunucuya gönderdi. Verilerinizin makinenizde kalması, makinedeki hiçbir şey onu dışarı göndermediği sürece geçerlidir.&lt;/div&gt;&lt;/a&gt;
&lt;div class="stats"&gt;&lt;h3&gt;Haftanın rakamları&lt;/h3&gt;&lt;div class="st"&gt;&lt;span&gt;Aktif kanal&lt;/span&gt;&lt;b&gt;24/43&lt;/b&gt;&lt;/div&gt;
&lt;div class="st"&gt;&lt;span&gt;Toplam video&lt;/span&gt;&lt;b&gt;122&lt;/b&gt;&lt;/div&gt;
&lt;div class="st"&gt;&lt;span&gt;Toplam izlenme&lt;/span&gt;&lt;b&gt;3.1M&lt;/b&gt;&lt;/div&gt;
&lt;div class="st"&gt;&lt;span&gt;Tema sayısı&lt;/span&gt;&lt;b&gt;15&lt;/b&gt;&lt;/div&gt;&lt;/div&gt;&lt;/div&gt;
&lt;div class="rowg g23"&gt;
&lt;div class="card"&gt;&lt;h3&gt;Haftalık ritim — günlük video sayısı (çubuk · sol eksen) + izlenme (eğri · sağ eksen)&lt;/h3&gt;&lt;svg viewBox="0 0 960 250" role="img"&gt;&lt;line x1="44" y1="26" x2="904" y2="26" stroke="#d7dce2"/&gt;&lt;text x="36" y="30" text-anchor="end" class="ax"&gt;29&lt;/text&gt;&lt;line x1="44" y1="74" x2="904" y2="74" stroke="#d7dce2"/&gt;&lt;text x="36" y="78" text-anchor="end" class="ax"&gt;22&lt;/text&gt;&lt;line x1="44" y1="121" x2="904" y2="121" stroke="#d7dce2"/&gt;&lt;text x="36" y="125" text-anchor="end" class="ax"&gt;14&lt;/text&gt;&lt;line x1="44" y1="168" x2="904" y2="168" stroke="#d7dce2"/&gt;&lt;text x="36" y="172" text-anchor="end" class="ax"&gt;7&lt;/text&gt;&lt;line x1="44" y1="216" x2="904" y2="216" stroke="#d7dce2"/&gt;&lt;text x="36" y="220" text-anchor="end" class="ax"&gt;0&lt;/text&gt;&lt;rect x="63.3" y="189.8" width="68.8" height="26.2" rx="3" fill="#2b5f58"&gt;&lt;title&gt;4 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="97.8" y="184.8" text-anchor="middle" class="bn2"&gt;4&lt;/text&gt;&lt;text x="97.8" y="238" text-anchor="middle" class="ax"&gt;18 Eki&lt;/text&gt;&lt;rect x="170.8" y="117.7" width="68.8" height="98.3" rx="3" fill="#2b5f58"&gt;&lt;title&gt;15 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="205.2" y="112.7" text-anchor="middle" class="bn2"&gt;15&lt;/text&gt;&lt;text x="205.2" y="238" text-anchor="middle" class="ax"&gt;19 Eki&lt;/text&gt;&lt;rect x="278.4" y="111.2" width="68.8" height="104.8" rx="3" fill="#2b5f58"&gt;&lt;title&gt;16 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="312.8" y="106.2" text-anchor="middle" class="bn2"&gt;16&lt;/text&gt;&lt;text x="312.8" y="238" text-anchor="middle" class="ax"&gt;20 Eki&lt;/text&gt;&lt;rect x="385.9" y="71.9" width="68.8" height="144.1" rx="3" fill="#2b5f58"&gt;&lt;title&gt;22 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="420.2" y="66.9" text-anchor="middle" class="bn2"&gt;22&lt;/text&gt;&lt;text x="420.2" y="238" text-anchor="middle" class="ax"&gt;21 Eki&lt;/text&gt;&lt;rect x="493.4" y="117.7" width="68.8" height="98.3" rx="3" fill="#2b5f58"&gt;&lt;title&gt;15 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="527.8" y="112.7" text-anchor="middle" class="bn2"&gt;15&lt;/text&gt;&lt;text x="527.8" y="238" text-anchor="middle" class="ax"&gt;22 Eki&lt;/text&gt;&lt;rect x="600.9" y="26.0" width="68.8" height="190.0" rx="3" fill="#2b5f58"&gt;&lt;title&gt;29 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="635.2" y="21.0" text-anchor="middle" class="bn2"&gt;29&lt;/text&gt;&lt;text x="635.2" y="238" text-anchor="middle" class="ax"&gt;23 Eki&lt;/text&gt;&lt;rect x="708.4" y="98.1" width="68.8" height="117.9" rx="3" fill="#2b5f58"&gt;&lt;title&gt;18 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="742.8" y="93.1" text-anchor="middle" class="bn2"&gt;18&lt;/text&gt;&lt;text x="742.8" y="238" text-anchor="middle" class="ax"&gt;24 Eki&lt;/text&gt;&lt;rect x="815.9" y="196.3" width="68.8" height="19.7" rx="3" fill="#2b5f58"&gt;&lt;title&gt;3 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="850.2" y="191.3" text-anchor="middle" class="bn2"&gt;3&lt;/text&gt;&lt;text x="850.2" y="238" text-anchor="middle" class="ax"&gt;25 Eki&lt;/text&gt;&lt;polyline fill="none" stroke="#d97706" stroke-width="2" points="97.8,189.2 205.2,96.4 312.8,172.2 420.2,26.0 527.8,184.3 635.2,152.3 742.8,175.4 850.2,215.8"/&gt;&lt;circle cx="97.8" cy="189.2" r="3.5" fill="#d97706"&gt;&lt;title&gt;163K izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="205.2" cy="96.4" r="3.5" fill="#d97706"&gt;&lt;title&gt;728K izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="312.8" cy="172.2" r="3.5" fill="#d97706"&gt;&lt;title&gt;267K izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="420.2" cy="26.0" r="3.5" fill="#d97706"&gt;&lt;title&gt;1.2M izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="527.8" cy="184.3" r="3.5" fill="#d97706"&gt;&lt;title&gt;193K izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="635.2" cy="152.3" r="3.5" fill="#d97706"&gt;&lt;title&gt;388K izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="742.8" cy="175.4" r="3.5" fill="#d97706"&gt;&lt;title&gt;247K izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="850.2" cy="215.8" r="3.5" fill="#d97706"&gt;&lt;title&gt;1K izlenme&lt;/title&gt;&lt;/circle&gt;&lt;text x="912" y="220" class="ax" fill="#92400e"&gt;0&lt;/text&gt;&lt;text x="912" y="125" class="ax" fill="#92400e"&gt;578K&lt;/text&gt;&lt;text x="912" y="30" class="ax" fill="#92400e"&gt;1.2M&lt;/text&gt;&lt;/svg&gt;&lt;/div&gt;
&lt;div class="card"&gt;&lt;h3&gt;Tema dağılımı&lt;/h3&gt;&lt;svg viewBox="0 0 220 220" class="donut"&gt;&lt;path class="dseg" data-tip="Ajan ve Sistem Güvenliği — 6 video (%5)" d="M110.0,30.0 A80,80 0 0 1 134.2,33.8" stroke="#0d9488" stroke-width="30" fill="none"&gt;&lt;title&gt;Ajan ve Sistem Güvenliği: 6 (%5)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Sesli Yapay Zeka ve Sesli Ajanlar — 17 video (%14)" d="M134.3,33.8 A80,80 0 0 1 184.0,79.5" stroke="#d97706" stroke-width="30" fill="none"&gt;&lt;title&gt;Sesli Yapay Zeka ve Sesli Ajanlar: 17 (%14)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Görüntü ve Video Üretimi — 5 video (%4)" d="M184.1,79.9 A80,80 0 0 1 189.3,99.6" stroke="#dc2626" stroke-width="30" fill="none"&gt;&lt;title&gt;Görüntü ve Video Üretimi: 5 (%4)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Belge Anlama ve OCR — 6 video (%5)" d="M189.3,99.7 A80,80 0 0 1 188.7,124.2" stroke="#2563eb" stroke-width="30" fill="none"&gt;&lt;title&gt;Belge Anlama ve OCR: 6 (%5)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Kimi ve Moonshot — 8 video (%7)" d="M188.7,124.3 A80,80 0 0 1 176.5,154.5" stroke="#4d7c0f" stroke-width="30" fill="none"&gt;&lt;title&gt;Kimi ve Moonshot: 8 (%7)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="DeepSeek — 7 video (%6)" d="M176.4,154.7 A80,80 0 0 1 156.5,175.1" stroke="#7c3aed" stroke-width="30" fill="none"&gt;&lt;title&gt;DeepSeek: 7 (%6)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Gemini ve Google Modelleri — 6 video (%5)" d="M156.4,175.2 A80,80 0 0 1 134.4,186.2" stroke="#0284c7" stroke-width="30" fill="none"&gt;&lt;title&gt;Gemini ve Google Modelleri: 6 (%5)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Difüzyon Dil Modelleri — 1 video (%1)" d="M134.3,186.2 A80,80 0 0 1 130.4,187.4" stroke="#92400e" stroke-width="30" fill="none"&gt;&lt;title&gt;Difüzyon Dil Modelleri: 1 (%1)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Yerel Yapay Zeka ve Gizlilik — 11 video (%9)" d="M130.4,187.4 A80,80 0 0 1 85.9,186.3" stroke="#db2777" stroke-width="30" fill="none"&gt;&lt;title&gt;Yerel Yapay Zeka ve Gizlilik: 11 (%9)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Sam Altman ve OpenAI Gündemi — 8 video (%7)" d="M85.7,186.2 A80,80 0 0 1 57.3,170.2" stroke="#0f766e" stroke-width="30" fill="none"&gt;&lt;title&gt;Sam Altman ve OpenAI Gündemi: 8 (%7)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="İş Dünyasında Yapay Zeka — 15 video (%12)" d="M57.2,170.1 A80,80 0 0 1 30.3,116.5" stroke="#475569" stroke-width="30" fill="none"&gt;&lt;title&gt;İş Dünyasında Yapay Zeka: 15 (%12)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Açık Modeller ve Verimlilik — 3 video (%2)" d="M30.2,116.2 A80,80 0 0 1 30.2,103.9" stroke="#b45309" stroke-width="30" fill="none"&gt;&lt;title&gt;Açık Modeller ve Verimlilik: 3 (%2)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Kodlama Ajanları ve Ajan Altyapısı — 13 video (%11)" d="M30.2,103.8 A80,80 0 0 1 51.1,55.9" stroke="#0d9488" stroke-width="30" fill="none"&gt;&lt;title&gt;Kodlama Ajanları ve Ajan Altyapısı: 13 (%11)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Öğrenme, Bilim ve Araştırma — 11 video (%9)" d="M51.3,55.7 A80,80 0 0 1 89.4,32.7" stroke="#d97706" stroke-width="30" fill="none"&gt;&lt;title&gt;Öğrenme, Bilim ve Araştırma: 11 (%9)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Diğer Öne Çıkanlar — 5 video (%4)" d="M89.6,32.6 A80,80 0 0 1 109.9,30.0" stroke="#dc2626" stroke-width="30" fill="none"&gt;&lt;title&gt;Diğer Öne Çıkanlar: 5 (%4)&lt;/title&gt;&lt;/path&gt;&lt;circle cx="110" cy="110" r="64" fill="#fff"/&gt;&lt;text x="110" y="106" text-anchor="middle" class="dn"&gt;122&lt;/text&gt;&lt;text x="110" y="126" text-anchor="middle" class="ax"&gt;video&lt;/text&gt;&lt;/svg&gt;&lt;div class="lgs"&gt;&lt;span class="lg"&gt;&lt;i style="background:#0d9488"&gt;&lt;/i&gt;Ajan ve Sistem Güvenliği &lt;b&gt;6&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#d97706"&gt;&lt;/i&gt;Sesli Yapay Zeka ve Sesli Ajanlar &lt;b&gt;17&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#dc2626"&gt;&lt;/i&gt;Görüntü ve Video Üretimi &lt;b&gt;5&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#2563eb"&gt;&lt;/i&gt;Belge Anlama ve OCR &lt;b&gt;6&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#4d7c0f"&gt;&lt;/i&gt;Kimi ve Moonshot &lt;b&gt;8&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#7c3aed"&gt;&lt;/i&gt;DeepSeek &lt;b&gt;7&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#0284c7"&gt;&lt;/i&gt;Gemini ve Google Modelleri &lt;b&gt;6&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#92400e"&gt;&lt;/i&gt;Difüzyon Dil Modelleri &lt;b&gt;1&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#db2777"&gt;&lt;/i&gt;Yerel Yapay Zeka ve Gizlilik &lt;b&gt;11&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#0f766e"&gt;&lt;/i&gt;Sam Altman ve OpenAI Gündemi &lt;b&gt;8&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#475569"&gt;&lt;/i&gt;İş Dünyasında Yapay Zeka &lt;b&gt;15&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#b45309"&gt;&lt;/i&gt;Açık Modeller ve Verimlilik &lt;b&gt;3&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#0d9488"&gt;&lt;/i&gt;Kodlama Ajanları ve Ajan Altyapısı &lt;b&gt;13&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#d97706"&gt;&lt;/i&gt;Öğrenme, Bilim ve Araştırma &lt;b&gt;11&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#dc2626"&gt;&lt;/i&gt;Diğer Öne Çıkanlar &lt;b&gt;5&lt;/b&gt;&lt;/span&gt;&lt;/div&gt;&lt;/div&gt;&lt;/div&gt;
&lt;div class="card" style="margin-bottom:14px"&gt;&lt;h3&gt;Trend zaman çizelgesi — her nokta bir video · x: yayın anı · y: log izlenme · renk: tema · üzerine gel: başlık&lt;/h3&gt;&lt;svg viewBox="0 0 960 300" role="img"&gt;&lt;line x1="46" y1="264.0" x2="946" y2="264.0" stroke="#d7dce2"/&gt;&lt;text x="38" y="268" text-anchor="end" class="ax"&gt;100&lt;/text&gt;&lt;line x1="46" y1="195.1" x2="946" y2="195.1" stroke="#d7dce2"/&gt;&lt;text x="38" y="199" text-anchor="end" class="ax"&gt;1K&lt;/text&gt;&lt;line x1="46" y1="126.2" x2="946" y2="126.2" stroke="#d7dce2"/&gt;&lt;text x="38" y="130" text-anchor="end" class="ax"&gt;10K&lt;/text&gt;&lt;line x1="46" y1="57.3" x2="946" y2="57.3" stroke="#d7dce2"/&gt;&lt;text x="38" y="61" text-anchor="end" class="ax"&gt;100K&lt;/text&gt;&lt;line x1="102.2" y1="16" x2="102.2" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="102.2" y="288" text-anchor="middle" class="ax"&gt;18 Eki&lt;/text&gt;&lt;line x1="214.8" y1="16" x2="214.8" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="214.8" y="288" text-anchor="middle" class="ax"&gt;19 Eki&lt;/text&gt;&lt;line x1="327.2" y1="16" x2="327.2" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="327.2" y="288" text-anchor="middle" class="ax"&gt;20 Eki&lt;/text&gt;&lt;line x1="439.8" y1="16" x2="439.8" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="439.8" y="288" text-anchor="middle" class="ax"&gt;21 Eki&lt;/text&gt;&lt;line x1="552.2" y1="16" x2="552.2" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="552.2" y="288" text-anchor="middle" class="ax"&gt;22 Eki&lt;/text&gt;&lt;line x1="664.8" y1="16" x2="664.8" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="664.8" y="288" text-anchor="middle" class="ax"&gt;23 Eki&lt;/text&gt;&lt;line x1="777.2" y1="16" x2="777.2" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="777.2" y="288" text-anchor="middle" class="ax"&gt;24 Eki&lt;/text&gt;&lt;line x1="889.8" y1="16" x2="889.8" y2="264" stroke="#e5e9ee"/&gt;&lt;text x="889.8" y="288" text-anchor="middle" class="ax"&gt;25 Eki&lt;/text&gt;&lt;circle cx="838.2" cy="81.5" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Harness² Explained: Self-Improving Agents Without Training - Prompt Engineering (44K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="87.6" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Kimi K3.1 Pro: Open Weights, Almost Opus - Prompt Engineering (36K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="674.1" cy="149.4" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;The Internet Is on Fire (Patch These Now) - Prompt Engineering (4K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="613.2" cy="110.6" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Google&amp;#x27;s GPT-3 Moment for Time Series &amp;amp; Forecasting (TimesFM 3.0) - Prompt Engineering (17K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="538.2" cy="55.6" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Claude Duet: Anthropic&amp;#x27;s Proactive Agent - Prompt Engineering (106K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="500.7" cy="90.1" r="4.5" fill="#7c3aed" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;DeepSeek Just Dropped a MASSIVE AI Breakthrough! - Prompt Engineering (33K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="275.7" cy="84.1" r="4.5" fill="#7c3aed" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;DeepSeek Just Made Reasoning WAY More Efficient! - Prompt Engineering (41K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="922.6" cy="246.1" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;AWESOME or Overhyped? OXR Document Parsing, Tested With My Own Ugly PDFs - Prompt Engineer 48 (82)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="889.8" cy="223.7" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;XERJ&amp;#x27;s CRAZY Local Search for AI Agents: I Tested It, It Broke, Then It Got AWESOME - Prompt Engineer 48 (284)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="875.7" cy="227.8" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Spark-X2.6-4B&amp;#x27;s CRAZY Thinking Dial: 180 Tests on an 8GB GPU - Prompt Engineer 48 (235)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="861.6" cy="224.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;OlmoEarth v1.3: INSANE Satellite AI or Fooled by WINTER? I Tested It on My RTX 4060 - Prompt Engineer 48 (275)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="856.9" cy="226.7" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This Nature Paper Turns Any Research Paper Into an AI Agent - INSANE - Prompt Engineer 48 (248)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="838.2" cy="180.9" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Block&amp;#x27;s Dessa AI Voice Is CRAZY: It LAUGHS Mid-Sentence (I Played the Real Demos) - Prompt Engineer 48 (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="814.8" cy="211.0" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Kling 4.0 Is Here… AI Video Just Got SCARY Good - Prompt Engineer 48 (487)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="805.4" cy="226.3" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;AI Avatars IGNORE You?! I Tested the CRAZY New StreamEval Benchmark (RTX 4090) - Prompt Engineer 48 (253)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="791.3" cy="220.5" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Microsoft&amp;#x27;s Agent World Is INSANE. I Ran It. The Agent FAILED (0/3) - Prompt Engineer 48 (328)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="767.9" cy="201.8" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;I Ran Flint, the CRAZY New 40B Coding Model, WITHOUT a Mac. Did It Fix My Buggy Repo? - Prompt Engineer 48 (699)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="219.1" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This AI Agent Builds Its OWN Tools to Finish Tasks! (INSANE Agent S4 Test) - Prompt Engineer 48 (349)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="730.4" cy="214.1" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Kimi K3 Just Went #1 on Frontend Code: The CRAZY Demos, Prices and What It Still Loses - Prompt Engineer 48 (430)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="725.7" cy="223.1" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;PHP Is Broken?! CVE-2026-109618 and 109619 Explained, Patch Today - Prompt Engineer 48 (292)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="697.6" cy="206.3" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;RAG-Anything Read Its OWN Paper… and the Results Are WILD - Prompt Engineer 48 (588)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="655.4" cy="206.7" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Mercury 3 Is INSANE: I Tested 13 Live Examples (+ 100 MILLION FREE Tokens) - Prompt Engineer 48 (579)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="725.7" cy="71.6" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Flint - The First RSI Coding Model? - Sam Witteveen (62K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="388.2" cy="102.5" r="4.5" fill="#92400e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Do Diffusion LLMs Scale? Testing Mercury 3 + Decide - Sam Witteveen (22K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="280.4" cy="123.8" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Meta&amp;#x27;s New Models &amp;amp; Agent - Sam Witteveen (11K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="847.6" cy="69.2" r="4.5" fill="#7c3aed" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;DeepSeek Is Free Now but What’s the Catch? - Kai (67K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="735.1" cy="84.7" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Google&amp;#x27;s Harness² Is This the End of Harness Engineering? - Kai (40K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="622.6" cy="76.4" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;llama.cpp Just Made 125B MoE Models 4× Faster! - Kai (53K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="510.1" cy="16.0" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Your Local AI Is Not as Private as You Think - Kai (397K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="486.6" cy="66.2" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;I Tested Every Best Local AI For GPU (8GB TO 24GB VRAM) - Kai (74K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="397.6" cy="82.3" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;OpenCode Got Hacked Badly (My API Keys Got Drained) - Kai (43K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="285.1" cy="25.0" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;NCM Is this the End of LLMs? - Kai (294K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="172.6" cy="62.8" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How I Got Frontier Performance on 16GB VRAM (Qwen 3.8 27B) - Kai (83K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="402.2" cy="115.4" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;I Tested 4 Codex Frameworks on the Same Real Feature - The Gray Cat (14K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="631.9" cy="151.2" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How To Use Google Flow (Complete Tutorial) - Howfinity (4K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="519.4" cy="182.5" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How to Analyze PDFs and Long Documents with Claude - Howfinity (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="411.6" cy="182.1" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How To Sell On TikTok Shop For Beginners (2026 Guide) - Howfinity (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="299.1" cy="180.5" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;HubSpot CRM Tutorial For Beginners - Howfinity (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="913.2" cy="203.7" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Sam Altman on What People Still Bring to an AI World - Matthew Berman (650)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="885.1" cy="136.0" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Kimi K3.1 Invented the Benchmark Scores - Matthew Berman (7K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="786.6" cy="52.5" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;They actually did it... (Kimi K3.1) - Matthew Berman (117K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="111.4" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Team AMA (and Kimi K3.1 Testing!!) - Matthew Berman (16K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="744.4" cy="169.8" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Sam Altman on Why Your Most Important Advice Should Come From People - Matthew Berman (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="739.8" cy="72.8" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Who is local AI really for? - Matthew Berman (59K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="646.0" cy="137.2" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Sam Altman on AI’s Discovery Era - Matthew Berman (7K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="547.6" cy="112.3" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Sam Altman on What Personal AI Still Needs - Matthew Berman (16K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="538.2" cy="109.3" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Sam Altman on Why AI Needs a New Kind of Hardware - Matthew Berman (17K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="538.2" cy="102.3" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Sam Altman on Why Demand for Coders Hasn&amp;#x27;t Disappeared - Matthew Berman (22K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="533.5" cy="125.0" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Sam Altman on Why AI Hasn&amp;#x27;t Made Us Less Busy - Matthew Berman (10K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="528.8" cy="84.8" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Actual Robot Bodyguard - Matthew Berman (40K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="510.1" cy="19.4" r="4.5" fill="#0f766e" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Sam Altman: RSI, AI Consciousness, OpenAI&amp;#x27;s New Hardware, Compute, and more! - Matthew Berman (355K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="210.1" cy="101.0" r="4.5" fill="#7c3aed" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How DeepSeek Uses Less Memory - Matthew Berman (23K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="767.9" cy="75.4" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Hunt bugs with your dot - OpenAI (54K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="147.3" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;From notes to AI training with ChatGPT Work | Realpage. - OpenAI (5K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="116.7" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How Realpage. builds AI training with ChatGPT Work - OpenAI (14K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="151.5" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;AI agents for everyday IT fixes | Shield and OpenAI - OpenAI (4K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="113.6" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How Shield uses OpenAI’s Astra to support IT teams - OpenAI (15K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="117.8" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How AT&amp;amp;T is building AI-powered legal workflows with OpenAI - OpenAI (13K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="175.6" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;From analyst to builder with Codex | Benefit Street Partners - OpenAI (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="127.3" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How Benefit Street Partners is building financial analysis tools with Codex - OpenAI (10K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="650.7" cy="85.4" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Messi asks ChatGPT to teach him a chess opening - OpenAI (39K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="636.6" cy="97.4" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Uber rolls out ChatGPT Work to 30,000+ employees - OpenAI (26K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="542.9" cy="96.4" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;V7 cuts costs 78% while boosting accuracy with GPT‑5.6 Luna - OpenAI (27K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="430.4" cy="66.6" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Build a Better Shopping Experience with OpenAI | Shopping Agent Demo - OpenAI (73K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="388.2" cy="123.5" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;You need a better headshot. - OpenAI (11K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="317.9" cy="70.0" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Cooking with Codex: Ep. 1 - OpenAI (65K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="285.1" cy="133.6" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Thomson Reuters on Building Trust in Legal AI | OpenAI - OpenAI (8K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="824.1" cy="187.2" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Kimi K3.1 Pro vs Opus 5.5 vs GPT-6 Astra: WHO WINS? Real Demos! - AI WITH Rithesh (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="796.0" cy="218.5" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Kimi K3.1 Pro Open-Weights Model Is INSANE! TESTED THOROUGHLY ! 92.2% on 302 Tests! - AI WITH Rithesh (357)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="706.9" cy="168.8" r="4.5" fill="#4d7c0f" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Kimi K3.1 Could Be INSANE…RUMOURS and LEAKS - AI WITH Rithesh (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="688.2" cy="166.8" r="4.5" fill="#0284c7" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Carbon COULD BE  INSANE… Leaked Demos &amp;amp; Rumours Investigated! - AI WITH Rithesh (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="589.8" cy="142.6" r="4.5" fill="#b45309" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Recursive Flint 40B Coding Model: Can It Code on Just 24GB RAM?! - AI WITH Rithesh (6K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="486.6" cy="198.9" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Gradium Phonon Is WILD… Tiny Open-Weight TTS Model Tested Thoroughly! - AI WITH Rithesh (782)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="467.9" cy="175.2" r="4.5" fill="#7c3aed" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;DeepSeek Just OPEN SOURCED Its AI Engine… Here’s How It REALLY Works! - AI WITH Rithesh (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="355.4" cy="201.7" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Cloudflare Clairic: The $0.10 AI Extraction Model Tested Thoroughly! - AI WITH Rithesh (702)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="346.0" cy="130.2" r="4.5" fill="#7c3aed" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Mercury 3 Is WILD… 1,000+ Tokens/sec Diffusion LLM Tested Thoroughly! - AI WITH Rithesh (9K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="285.1" cy="158.4" r="4.5" fill="#7c3aed" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;MiniMax M3.1 Flash JUST DROPPED… I Tested 302 Tasks! - AI WITH Rithesh (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="172.6" cy="108.9" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;NVIDIA Dory 4B Is WILD… Looped Transformer, Same Memory? - AI WITH Rithesh (18K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="622.6" cy="126.1" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;ComfyUI Models Explained: Download, Folders and Model Loaders - pixaroma (10K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="392.9" cy="93.3" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Flux 3 Klein in ComfyUI | Generate, Edit &amp;amp; Combine Images (Ep39) - pixaroma (30K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="842.9" cy="146.8" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Give Me 35 Minutes, I&amp;#x27;ll Make LLM Pre-training Click Forever - Zachary Huang (5K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="505.4" cy="138.2" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Give Me 1 Hour, I&amp;#x27;ll Make Multimodal Deep Learning Click Forever - Zachary Huang (7K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="177.2" cy="85.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Give Me 1 Hour, I&amp;#x27;ll Make Deep Learning Click Forever - Zachary Huang (39K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="725.7" cy="119.2" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Run Your AI Software Factory Like a Real Engineering Org (Here&amp;#x27;s How) - Cole Medin (13K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="552.2" cy="98.3" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;AI Makes Us WAY Faster but Fries Our Focus (Here&amp;#x27;s What I Do About It) - Cole Medin (25K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="411.6" cy="169.2" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Prompt Injection Is Just the Start of AI Agent Security (Here&amp;#x27;s What You Need) - Cole Medin (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="402.2" cy="98.5" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Let Your AI Agent Do ITS OWN Harness Engineering (Here&amp;#x27;s How) - Cole Medin (25K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="327.2" cy="148.9" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Copilot Can Use Your Computer for You #ai #coding #githubcopilot - Burke Holland (5K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="285.1" cy="131.1" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;I tested four AI strategies to find the cheapest one - Burke Holland (8K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="697.6" cy="143.3" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Llama: Local AI on Your Mac - Hugging Face (6K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="608.5" cy="171.1" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Automate CSV to Parquet with Buckets, Webhooks &amp;amp; Jobs - Hugging Face (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="500.7" cy="185.7" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;I Turned Structured Output Into a Game &#127918; - Hugging Face (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="350.7" cy="116.0" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Local AI 202: Get Massive Speed-ups with Speculative Decoding - Hugging Face (14K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="280.4" cy="157.8" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Pi Durable: AI agents that survive crashes - Hugging Face (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="228.8" cy="75.3" r="4.5" fill="#db2777" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Pi Durable EXPLAINED: Build Always-On Agents with Local Models - Hugging Face (55K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="252.2" cy="32.5" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Argon Is Not What I Expected - Two Minute Papers (229K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="875.7" cy="247.3" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;I Made $11,000 Selling Stickers: Then Built AI to Scale It. - Build w Agents - Aggregate Intellect - AI.SCIENCE (75)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="758.5" cy="241.0" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;From Click to Booking: I Built the Tracking Most Businesses Are Missing - Build w Agents - Aggregate Intellect - AI.SCIENCE (116)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="603.8" cy="237.8" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;This AI Doesn’t Just Remind You, It Knows When to Back Off. - Build w Agents - Aggregate Intellect - AI.SCIENCE (140)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="735.1" cy="158.0" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;The Boring Version of Continual Learning - LangChain (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="519.4" cy="135.5" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;How Deep Agents Manages Context: Offloading and Summarization - LangChain (7K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="275.7" cy="171.5" r="4.5" fill="#475569" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Podium&amp;#x27;s Playbook for Scaling AI Agents Org-Wide - LangChain (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="261.6" cy="138.7" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;What is LangChain Managed Deep Agents? - LangChain (6K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="814.8" cy="190.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Density functional theory meets AI: traditional DFT vs a pretrained model - DigitalSreeni (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="585.1" cy="194.9" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;15 - Spatial Pharmacokinetics: Measuring Drug Diffusion in Tumor Tissue - DigitalSreeni (908)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="814.8" cy="241.0" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Voice Agents That Interrupt You #Shorts - Trelis Research (116)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="810.1" cy="228.3" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Dialt: Deep Control Over Voice Agent Behavior #Shorts - Trelis Research (230)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="706.9" cy="217.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Voice Agents That Interrupt You - Trelis Research (375)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="697.6" cy="221.2" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Background Speaker Suppression in Dialt #Shorts - Trelis Research (318)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="585.1" cy="219.1" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Dialt Spot: Voice AI at $1.99 per Hour During Quieter Periods - Trelis Research (349)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="519.4" cy="214.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Dialt Launch Webinar: Voice AI Demo, WorkStuff AI &amp;amp; Q&amp;amp;A - Trelis Research (425)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="472.6" cy="218.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Detta and Phonon-2: Fast Dictation on a Mac - Trelis Research (359)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="360.1" cy="217.8" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Dialt: Deep Control Over How Voice Agents Behave and Sound - Trelis Research (368)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="261.6" cy="215.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Background Speaker Suppression in Dialt: A Live Demo - Trelis Research (407)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="247.6" cy="228.5" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Two Ways to Switch Languages with Voice Assistants | Dialt #Shorts - Trelis Research (228)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="397.6" cy="136.4" r="4.5" fill="#0d9488" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Learn to build better AI Agents through better context - DeepLearningAI (7K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="622.6" cy="163.9" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Introducing OCRia-1, the world’s best document parsing model - LlamaIndex (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="524.1" cy="221.6" r="4.5" fill="#2563eb" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Building an Auto Total-Loss Claims Workflow with LlamaParse - LlamaIndex (312)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="416.3" cy="128.2" r="4.5" fill="#dc2626" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Answering your questions about AI, agents, and tech jobs (Q&amp;amp;A) - Dave Ebbelaar (9K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="547.6" cy="129.0" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;&#128266; Voice design and voice replication with Gemini 3.8 Flash TTS - Google for Developers (9K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="528.8" cy="93.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Design and replicate voices with Gemini 3.8 Flash TTS - Google for Developers (30K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="683.5" cy="103.4" r="4.5" fill="#d97706" fill-opacity=".85" stroke="#fff" stroke-width="1"&gt;&lt;title&gt;Triple BAM! Live with Jon Krohn, Josh Starmer, and Luis Serrano! - StatQuest with Josh Starmer (21K)&lt;/title&gt;&lt;/circle&gt;&lt;/svg&gt;&lt;/div&gt;
&lt;div class="rowg g11"&gt;
&lt;div class="card"&gt;&lt;h3&gt;Tema yoğunluğu&lt;/h3&gt;&lt;div class="bars"&gt;&lt;a class="bar" href="#t1"&gt;&lt;span class="bn"&gt;Ajan ve Sistem Güvenliği&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:35%;background:#0d9488"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;6&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t2"&gt;&lt;span class="bn"&gt;Sesli Yapay Zeka ve Sesli Ajanlar&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:100%;background:#d97706"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;17&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t3"&gt;&lt;span class="bn"&gt;Görüntü ve Video Üretimi&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:29%;background:#dc2626"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;5&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t4"&gt;&lt;span class="bn"&gt;Belge Anlama ve OCR&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:35%;background:#2563eb"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;6&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t5"&gt;&lt;span class="bn"&gt;Kimi ve Moonshot&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:47%;background:#4d7c0f"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;8&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t6"&gt;&lt;span class="bn"&gt;DeepSeek&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:41%;background:#7c3aed"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;7&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t7"&gt;&lt;span class="bn"&gt;Gemini ve Google Modelleri&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:35%;background:#0284c7"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;6&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t8"&gt;&lt;span class="bn"&gt;Difüzyon Dil Modelleri&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:6%;background:#92400e"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;1&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t9"&gt;&lt;span class="bn"&gt;Yerel Yapay Zeka ve Gizlilik&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:65%;background:#db2777"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;11&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t10"&gt;&lt;span class="bn"&gt;Sam Altman ve OpenAI Gündemi&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:47%;background:#0f766e"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;8&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t11"&gt;&lt;span class="bn"&gt;İş Dünyasında Yapay Zeka&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:88%;background:#475569"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;15&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t12"&gt;&lt;span class="bn"&gt;Açık Modeller ve Verimlilik&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:18%;background:#b45309"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;3&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t13"&gt;&lt;span class="bn"&gt;Kodlama Ajanları ve Ajan Altyapısı&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:76%;background:#0d9488"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;13&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t14"&gt;&lt;span class="bn"&gt;Öğrenme, Bilim ve Araştırma&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:65%;background:#d97706"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;11&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t15"&gt;&lt;span class="bn"&gt;Diğer Öne Çıkanlar&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:29%;background:#dc2626"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;5&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/div&gt;
&lt;div class="card"&gt;&lt;h3&gt;En çok izlenen kanallar&lt;/h3&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Kai&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:100%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;1.1M&lt;i&gt;8 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Matthew Berman&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:66%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;693K&lt;i&gt;14 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;OpenAI&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:35%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;366K&lt;i&gt;15 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:27%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;281K&lt;i&gt;7 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Two Minute Papers&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:22%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;229K&lt;i&gt;1 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Sam Witteveen&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:9%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;95K&lt;i&gt;3 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;/div&gt;&lt;/div&gt;
&lt;section class="theme" id="t1"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#0d9488"&gt;01&lt;/span&gt;&lt;i class="dot" style="background:#0d9488"&gt;&lt;/i&gt;Ajan ve Sistem Güvenliği&lt;span class="cnt"&gt;6 video &amp;middot; 52K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Beş gün içinde PHP&amp;#x27;den Node.js&amp;#x27;e, GitLab&amp;#x27;dan PyPI zehirlenmesine kadar neredeyse her geliştirici katmanı yamalandı; ayrıca OpenCode API anahtarı hırsızlığı ve Microsoft&amp;#x27;un Project Perception ile kurumsal ajan güvenliği hamlesi, ajanların yeni saldırı yüzeyi olduğunu gösterdi.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=0p96nZsRxlY" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;43K&lt;/span&gt;&lt;span class="dt"&gt;20 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;OpenCode Got Hacked Badly (My API Keys Got Drained)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=0XSk7fCimYE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;4K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;The Internet Is on Fire (Patch These Now)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=hLoyhm1dHlM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;20 Eki&lt;/span&gt;&lt;span class="ch"&gt;Cole Medin&lt;/span&gt;&lt;span class="tt"&gt;Prompt Injection Is Just the Start of AI Agent Security (Here&amp;#x27;s What You Need)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ST2xBczjTB8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;702&lt;/span&gt;&lt;span class="dt"&gt;20 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Cloudflare Clairic: The $0.10 AI Extraction Model Tested Thoroughly!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=C6_4WsBrlUI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;579&lt;/span&gt;&lt;span class="dt"&gt;22 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Mercury 3 Is INSANE: I Tested 13 Live Examples (+ 100 MILLION FREE Tokens)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=DS-9KjJe1fk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;292&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;PHP Is Broken?! CVE-2026-109618 and 109619 Explained, Patch Today&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t2"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#d97706"&gt;02&lt;/span&gt;&lt;i class="dot" style="background:#d97706"&gt;&lt;/i&gt;Sesli Yapay Zeka ve Sesli Ajanlar&lt;span class="cnt"&gt;17 video &amp;middot; 183K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Trelis&amp;#x27;in Dialt lansmanı (kesintili konuşma, arka plan bastırma, saatte 1,99 dolara varan Spot fiyatlandırma), Google&amp;#x27;ın Gemini 3.8 Flash TTS ile 100+ dilde ses tasarımı/çoğaltması ve Block&amp;#x27;un insan sesiyle eğitilen Dessa&amp;#x27;sı sesli ajanları demo olmaktan çıkarıp üretim maliyetlerine taşıdı.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=8n-SXxRIrm8" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;73K&lt;/span&gt;&lt;span class="dt"&gt;20 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Build a Better Shopping Experience with OpenAI | Shopping Agent Demo&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=DQHtmW9ZlIs" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;65K&lt;/span&gt;&lt;span class="dt"&gt;19 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Cooking with Codex: Ep. 1&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=SOF3_evPje8" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;30K&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;Design and replicate voices with Gemini 3.8 Flash TTS&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=CsF5oqVV5YE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;9K&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;&#128266; Voice design and voice replication with Gemini 3.8 Flash TTS&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=3Q9A8izO-Ww" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;24 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Block&amp;#x27;s Dessa AI Voice Is CRAZY: It LAUGHS Mid-Sentence (I Played the Real Demos)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=g1xqHojC5Ec" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;782&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Gradium Phonon Is WILD… Tiny Open-Weight TTS Model Tested Thoroughly!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=7-NLWOVANY8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;425&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Dialt Launch Webinar: Voice AI Demo, WorkStuff AI &amp;amp; Q&amp;amp;A&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Jxw85gnfrGQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;407&lt;/span&gt;&lt;span class="dt"&gt;19 Eki&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Background Speaker Suppression in Dialt: A Live Demo&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=eyfZxncUHFE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;375&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Voice Agents That Interrupt You&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Xs2PR-ZIfIU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;368&lt;/span&gt;&lt;span class="dt"&gt;20 Eki&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Dialt: Deep Control Over How Voice Agents Behave and Sound&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Y5HGnlsImJs" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;359&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Detta and Phonon-2: Fast Dictation on a Mac&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=rTMubDuwUyU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;349&lt;/span&gt;&lt;span class="dt"&gt;22 Eki&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Dialt Spot: Voice AI at $1.99 per Hour During Quieter Periods&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Nuip2xEHQ5c" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;318&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Background Speaker Suppression in Dialt #Shorts&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=tW07tpgDUg8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;253&lt;/span&gt;&lt;span class="dt"&gt;24 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;AI Avatars IGNORE You?! I Tested the CRAZY New StreamEval Benchmark (RTX 4090)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=E-J6K-Aao2A" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;230&lt;/span&gt;&lt;span class="dt"&gt;24 Eki&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Dialt: Deep Control Over Voice Agent Behavior #Shorts&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=6EBnVGTqTpQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;228&lt;/span&gt;&lt;span class="dt"&gt;19 Eki&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Two Ways to Switch Languages with Voice Assistants | Dialt #Shorts&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=8l5Efmo0svg" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;116&lt;/span&gt;&lt;span class="dt"&gt;24 Eki&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Voice Agents That Interrupt You #Shorts&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t3"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#dc2626"&gt;03&lt;/span&gt;&lt;i class="dot" style="background:#dc2626"&gt;&lt;/i&gt;Görüntü ve Video Üretimi&lt;span class="cnt"&gt;5 video &amp;middot; 55K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Kling 4.0 tek klipte 30 saniyeye çıkarak AI videoyu uzun plana taşıdı; ComfyUI/Flux 3 Klein iş akışları ve ChatGPT headshot özelliği görsel üretimin günlük araçlara gömüldüğünü gösteriyor.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ntxxhO9H3bY" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;30K&lt;/span&gt;&lt;span class="dt"&gt;20 Eki&lt;/span&gt;&lt;span class="ch"&gt;pixaroma&lt;/span&gt;&lt;span class="tt"&gt;Flux 3 Klein in ComfyUI | Generate, Edit &amp;amp; Combine Images (Ep39)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=v45XAMqH7pE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;11K&lt;/span&gt;&lt;span class="dt"&gt;20 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;You need a better headshot.&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ipvP48pz_Bw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;10K&lt;/span&gt;&lt;span class="dt"&gt;22 Eki&lt;/span&gt;&lt;span class="ch"&gt;pixaroma&lt;/span&gt;&lt;span class="tt"&gt;ComfyUI Models Explained: Download, Folders and Model Loaders&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=VbF1IiBAsqo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;4K&lt;/span&gt;&lt;span class="dt"&gt;22 Eki&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;How To Use Google Flow (Complete Tutorial)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=qd_TDSkx5Zw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;487&lt;/span&gt;&lt;span class="dt"&gt;24 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Kling 4.0 Is Here… AI Video Just Got SCARY Good&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t4"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#2563eb"&gt;04&lt;/span&gt;&lt;i class="dot" style="background:#2563eb"&gt;&lt;/i&gt;Belge Anlama ve OCR&lt;span class="cnt"&gt;6 video &amp;middot; 234K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;LlamaIndex&amp;#x27;in OCRia-1&amp;#x27;i ve açık kaynak OXR, tablo/grafik ağırlıklı belge ayrıştırmada yeni rekorları kovalarken RAG-Anything&amp;#x27;in kendi makalesini okuma testi, çok modlu RAG&amp;#x27;ın olgunlaştığını kanıtladı.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=LWxX6vvHfrc" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;229K&lt;/span&gt;&lt;span class="dt"&gt;19 Eki&lt;/span&gt;&lt;span class="ch"&gt;Two Minute Papers&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Argon Is Not What I Expected&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=1_p2OAT8RZ4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;22 Eki&lt;/span&gt;&lt;span class="ch"&gt;LlamaIndex&lt;/span&gt;&lt;span class="tt"&gt;Introducing OCRia-1, the world’s best document parsing model&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=YIvSE07UCC4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;How to Analyze PDFs and Long Documents with Claude&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=MEOGGYW2xIc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;588&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;RAG-Anything Read Its OWN Paper… and the Results Are WILD&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ytrlq9pIpjo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;312&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;LlamaIndex&lt;/span&gt;&lt;span class="tt"&gt;Building an Auto Total-Loss Claims Workflow with LlamaParse&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=NKzcyVpeyyI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;82&lt;/span&gt;&lt;span class="dt"&gt;25 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;AWESOME or Overhyped? OXR Document Parsing, Tested With My Own Ugly PDFs&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t5"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#4d7c0f"&gt;05&lt;/span&gt;&lt;i class="dot" style="background:#4d7c0f"&gt;&lt;/i&gt;Kimi ve Moonshot&lt;span class="cnt"&gt;8 video &amp;middot; 181K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Kimi K3 Arena Frontend Code&amp;#x27;da birinciliği aldı, K3.1 Pro ise 2,8 trilyon parametreli MoE mimarisi ve 1M token bağlamla Terminal-Bench&amp;#x27;te Opus 5.5&amp;#x27;in 1 puan gerisine, GPT-6 Astra&amp;#x27;nın önüne yerleşti — üstelik üçte bir maliyetle. Açık ağırlıklar henüz çıkmadı, benchmark iddiaları tartışmalı.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=5sEVvu4mQ8g" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;117K&lt;/span&gt;&lt;span class="dt"&gt;24 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;They actually did it... (Kimi K3.1)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=zgluzf-zjhI" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;36K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;Kimi K3.1 Pro: Open Weights, Almost Opus&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=7fZ1rFcm1nA" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;16K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Team AMA (and Kimi K3.1 Testing!!)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=bBqP-w6DHFw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;7K&lt;/span&gt;&lt;span class="dt"&gt;24 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Kimi K3.1 Invented the Benchmark Scores&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Qh2SuMGLUf0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Kimi K3.1 Could Be INSANE…RUMOURS and LEAKS&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=r3K4fPH6GSo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;24 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Kimi K3.1 Pro vs Opus 5.5 vs GPT-6 Astra: WHO WINS? Real Demos!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=o6lb6_R7VvQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;430&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Kimi K3 Just Went #1 on Frontend Code: The CRAZY Demos, Prices and What It Still Loses&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=52H96MH5UtI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;357&lt;/span&gt;&lt;span class="dt"&gt;24 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Kimi K3.1 Pro Open-Weights Model Is INSANE! TESTED THOROUGHLY ! 92.2% on 302 Tests!&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t6"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#7c3aed"&gt;06&lt;/span&gt;&lt;i class="dot" style="background:#7c3aed"&gt;&lt;/i&gt;DeepSeek&lt;span class="cnt"&gt;7 video &amp;middot; 178K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;DeepSeek, API&amp;#x27;sinin arkasındaki DeepInfer çıkarım motorunu (günde 155 trilyon token, %96 KV-cache isabeti) ve V4.1 Flash 1017&amp;#x27;yi açtı; düşünce token&amp;#x27;larını üçte bir azaltan güncelleme Arena WebDev&amp;#x27;de açık modellerde birinciliğe taşıdı.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=zIWxzkO1nak" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;67K&lt;/span&gt;&lt;span class="dt"&gt;24 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;DeepSeek Is Free Now but What’s the Catch?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=YGEz40RSpTE" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;41K&lt;/span&gt;&lt;span class="dt"&gt;19 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;DeepSeek Just Made Reasoning WAY More Efficient!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=esdRFXgXaaQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;33K&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;DeepSeek Just Dropped a MASSIVE AI Breakthrough!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=RS9vWYlJeH8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;23K&lt;/span&gt;&lt;span class="dt"&gt;18 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;How DeepSeek Uses Less Memory&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=5VLKHqnjO8w" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;9K&lt;/span&gt;&lt;span class="dt"&gt;20 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Mercury 3 Is WILD… 1,000+ Tokens/sec Diffusion LLM Tested Thoroughly!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=sQFIhViy4Uk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;19 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;MiniMax M3.1 Flash JUST DROPPED… I Tested 302 Tasks!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=mQ-nB8g9mOc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;DeepSeek Just OPEN SOURCED Its AI Engine… Here’s How It REALLY Works!&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t7"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#0284c7"&gt;07&lt;/span&gt;&lt;i class="dot" style="background:#0284c7"&gt;&lt;/i&gt;Gemini ve Google Modelleri&lt;span class="cnt"&gt;6 video &amp;middot; 481K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Gemini 4 Argon beklentileri karşılamadı, Carbon sızıntıları konuşuldu; Google&amp;#x27;ın TimesFM 3.0&amp;#x27;ı eğitim gerektirmeden zaman serisi tahmini yaparken Harness², modeli eğitmeden ajanın iskeletini (prompt, skill, alt-ajan) kendi kendini iyileştirmesini sağlayan araştırma çerçevesi olarak öne çıktı.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=vLqswKyHS64" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;294K&lt;/span&gt;&lt;span class="dt"&gt;19 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;NCM Is this the End of LLMs?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=h_KqS4GtU3I" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;83K&lt;/span&gt;&lt;span class="dt"&gt;18 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;How I Got Frontier Performance on 16GB VRAM (Qwen 3.8 27B)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=pW2ktrJlsyE" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;44K&lt;/span&gt;&lt;span class="dt"&gt;24 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;Harness² Explained: Self-Improving Agents Without Training&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=yCgfyQmqhPM" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;40K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Google&amp;#x27;s Harness² Is This the End of Harness Engineering?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ByKEkRohzt8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;17K&lt;/span&gt;&lt;span class="dt"&gt;22 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;Google&amp;#x27;s GPT-3 Moment for Time Series &amp;amp; Forecasting (TimesFM 3.0)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=qw2HyicPz3w" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Carbon COULD BE  INSANE… Leaked Demos &amp;amp; Rumours Investigated!&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t8"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#92400e"&gt;08&lt;/span&gt;&lt;i class="dot" style="background:#92400e"&gt;&lt;/i&gt;Difüzyon Dil Modelleri&lt;span class="cnt"&gt;1 video &amp;middot; 22K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Inception&amp;#x27;ın Mercury 3&amp;#x27;ü 1.000+ token/sn iddiasıyla difüzyon LLM&amp;#x27;leri yeniden gündeme taşıdı; bağımsız testler uçtan uca 258-416 token/sn ölçse de hız/bedava token stratejisi dikkat çekici.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=bEO_QeJU7U0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;22K&lt;/span&gt;&lt;span class="dt"&gt;20 Eki&lt;/span&gt;&lt;span class="ch"&gt;Sam Witteveen&lt;/span&gt;&lt;span class="tt"&gt;Do Diffusion LLMs Scale? Testing Mercury 3 + Decide&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t9"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#db2777"&gt;09&lt;/span&gt;&lt;i class="dot" style="background:#db2777"&gt;&lt;/i&gt;Yerel Yapay Zeka ve Gizlilik&lt;span class="cnt"&gt;11 video &amp;middot; 680K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Haftanın en çok izlenen videosu yerel yapay zekanın gizlilik varsayımını yıktı; aynı zamanda llama.cpp&amp;#x27;nin 125B MoE modelleri 4x hızlandırması, Mac için resmi Llama uygulaması ve spekülatif kodlama eğitimleri yerel tarafın ivmesini gösteriyor.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=3mx84MRj8lY" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;397K&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Your Local AI Is Not as Private as You Think&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ugKAkCVBcz0" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;74K&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;I Tested Every Best Local AI For GPU (8GB TO 24GB VRAM)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=fHY0sjAejcA" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;59K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Who is local AI really for?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ti3vVYwzW4A" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;55K&lt;/span&gt;&lt;span class="dt"&gt;19 Eki&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;Pi Durable EXPLAINED: Build Always-On Agents with Local Models&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=shl8FFSoANE" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;53K&lt;/span&gt;&lt;span class="dt"&gt;22 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;llama.cpp Just Made 125B MoE Models 4× Faster!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=D3YM5OzNt9s" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;18K&lt;/span&gt;&lt;span class="dt"&gt;18 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;NVIDIA Dory 4B Is WILD… Looped Transformer, Same Memory?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=H4azmWcqO5g" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;14K&lt;/span&gt;&lt;span class="dt"&gt;20 Eki&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;Local AI 202: Get Massive Speed-ups with Speculative Decoding&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=w6dSNCg9I5k" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;6K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;Llama: Local AI on Your Mac&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=3-sSbJKb-fQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;19 Eki&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;Pi Durable: AI agents that survive crashes&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=vmgbq-7WRpQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;699&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;I Ran Flint, the CRAZY New 40B Coding Model, WITHOUT a Mac. Did It Fix My Buggy Repo?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ZDYrcHPzlsM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;284&lt;/span&gt;&lt;span class="dt"&gt;25 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;XERJ&amp;#x27;s CRAZY Local Search for AI Agents: I Tested It, It Broke, Then It Got AWESOME&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t10"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#0f766e"&gt;10&lt;/span&gt;&lt;i class="dot" style="background:#0f766e"&gt;&lt;/i&gt;Sam Altman ve OpenAI Gündemi&lt;span class="cnt"&gt;8 video &amp;middot; 430K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Matthew Berman&amp;#x27;ın Altman röportaj serisi RSI (kendi kendini geliştiren yapay zeka), bilince, yeni donanım ihtiyacına ve kodcu talebinin neden kaybolmadığına odaklandı — OpenAI&amp;#x27;ın yönü hakkında en çok konuşulan birincil kaynak haftaydı.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=3gXLmy7DK_4" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;355K&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Sam Altman: RSI, AI Consciousness, OpenAI&amp;#x27;s New Hardware, Compute, and more!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=1g9Tqvyt4kU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;22K&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Sam Altman on Why Demand for Coders Hasn&amp;#x27;t Disappeared&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=u6Tf62_5CP0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;17K&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Sam Altman on Why AI Needs a New Kind of Hardware&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=tc92psltcEg" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;16K&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Sam Altman on What Personal AI Still Needs&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=SU-h1k5IYOI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;10K&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Sam Altman on Why AI Hasn&amp;#x27;t Made Us Less Busy&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=9WZpou1faFU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;7K&lt;/span&gt;&lt;span class="dt"&gt;22 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Sam Altman on AI’s Discovery Era&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=f8uIINtoJCM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Sam Altman on Why Your Most Important Advice Should Come From People&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=eQTOJZ6_NVU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;650&lt;/span&gt;&lt;span class="dt"&gt;25 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Sam Altman on What People Still Bring to an AI World&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t11"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#475569"&gt;11&lt;/span&gt;&lt;i class="dot" style="background:#475569"&gt;&lt;/i&gt;İş Dünyasında Yapay Zeka&lt;span class="cnt"&gt;15 video &amp;middot; 128K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Uber&amp;#x27;ın 30 bin çalışanına ChatGPT Work&amp;#x27;ü açması, AT&amp;amp;T&amp;#x27;nin hukuk ve Thomson Reuters&amp;#x27;ın güven odaklı hukuk yapay zekası iş akışları kurumsal benimsemenin ölçeğini; sticker satışıyla doğrulanıp ajanlarla ölçeklenen e-ticaret ve satış hunisi analizi ise KOBİ tarafını temsil etti.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=20IRGgQwPkc" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;27K&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;V7 cuts costs 78% while boosting accuracy with GPT‑5.6 Luna&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=l16Gw7oVLNQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;26K&lt;/span&gt;&lt;span class="dt"&gt;22 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Uber rolls out ChatGPT Work to 30,000+ employees&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=m6MlxaFMmPI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;15K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;How Shield uses OpenAI’s Astra to support IT teams&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=hgX1hgyWD4g" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;14K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;How Realpage. builds AI training with ChatGPT Work&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=V39_L7uh66M" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;13K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;How AT&amp;amp;T is building AI-powered legal workflows with OpenAI&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=CTkDi1THc_g" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;10K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;How Benefit Street Partners is building financial analysis tools with Codex&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=7lC2N9woZy4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;8K&lt;/span&gt;&lt;span class="dt"&gt;19 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Thomson Reuters on Building Trust in Legal AI | OpenAI&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=7Cs5qyBlYQk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;5K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;From notes to AI training with ChatGPT Work | Realpage.&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=2lNUGkCg_qc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;4K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;AI agents for everyday IT fixes | Shield and OpenAI&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=8mp0jF2At1o" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;19 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;Podium&amp;#x27;s Playbook for Scaling AI Agents Org-Wide&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=qyGvwLgqmaU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;From analyst to builder with Codex | Benefit Street Partners&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=vc7b-u9G6kg" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;19 Eki&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;HubSpot CRM Tutorial For Beginners&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=wLuuesGRhuM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;20 Eki&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;How To Sell On TikTok Shop For Beginners (2026 Guide)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=pFWN1VBCQ28" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;116&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;Build w Agents - Aggregate Intellect - AI.SCIENCE&lt;/span&gt;&lt;span class="tt"&gt;From Click to Booking: I Built the Tracking Most Businesses Are Missing&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=AXFOzSRg5i8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;75&lt;/span&gt;&lt;span class="dt"&gt;24 Eki&lt;/span&gt;&lt;span class="ch"&gt;Build w Agents - Aggregate Intellect - AI.SCIENCE&lt;/span&gt;&lt;span class="tt"&gt;I Made $11,000 Selling Stickers: Then Built AI to Scale It.&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t12"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#b45309"&gt;12&lt;/span&gt;&lt;i class="dot" style="background:#b45309"&gt;&lt;/i&gt;Açık Modeller ve Verimlilik&lt;span class="cnt"&gt;3 video &amp;middot; 17K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Meta&amp;#x27;nın 816B&amp;#x27;lik açık ağırlıklı Muse Spark Open&amp;#x27;ı, NVIDIA Dory 4B&amp;#x27;nin döngülü transformer&amp;#x27;ı, MiniMax M3.1 ve 8GB GPU&amp;#x27;da testlenen Spark-X2.6-4B&amp;#x27;nin &amp;#x27;düşünme kadranı&amp;#x27; açık modellerde mimari verimlilik yarışının hızlandığını gösterdi.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=aKVfjG7_R4g" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;11K&lt;/span&gt;&lt;span class="dt"&gt;19 Eki&lt;/span&gt;&lt;span class="ch"&gt;Sam Witteveen&lt;/span&gt;&lt;span class="tt"&gt;Meta&amp;#x27;s New Models &amp;amp; Agent&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Ne1nEO6L-kE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;6K&lt;/span&gt;&lt;span class="dt"&gt;22 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Recursive Flint 40B Coding Model: Can It Code on Just 24GB RAM?!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ZCXaQEOkaDI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;235&lt;/span&gt;&lt;span class="dt"&gt;24 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Spark-X2.6-4B&amp;#x27;s CRAZY Thinking Dial: 180 Tests on an 8GB GPU&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t13"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#0d9488"&gt;13&lt;/span&gt;&lt;i class="dot" style="background:#0d9488"&gt;&lt;/i&gt;Kodlama Ajanları ve Ajan Altyapısı&lt;span class="cnt"&gt;13 video &amp;middot; 308K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Recursive&amp;#x27;ın RSI ile eğittiği 40B&amp;#x27;lik Flint kodlama modeli haftanın merak uyandıran çıkışıydı; LangChain&amp;#x27;in Managed Deep Agents&amp;#x27;ı ve bağlam çürümesine karşı offloading/özetleme teknikleri ile Cole Medin&amp;#x27;in kendi iskeletini kendi iyileştiren ajanı, ajan altyapısının olgunlaştığını gösterdi.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=0sI-GTYetD8" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;106K&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;Claude Duet: Anthropic&amp;#x27;s Proactive Agent&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=BsfjQDSeTwI" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;62K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;Sam Witteveen&lt;/span&gt;&lt;span class="tt"&gt;Flint - The First RSI Coding Model?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=6f7ojkd4_Wc" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;54K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Hunt bugs with your dot&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=TVhhmhvCovg" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;25K&lt;/span&gt;&lt;span class="dt"&gt;20 Eki&lt;/span&gt;&lt;span class="ch"&gt;Cole Medin&lt;/span&gt;&lt;span class="tt"&gt;Let Your AI Agent Do ITS OWN Harness Engineering (Here&amp;#x27;s How)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=-LODwSuymII" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;14K&lt;/span&gt;&lt;span class="dt"&gt;20 Eki&lt;/span&gt;&lt;span class="ch"&gt;The Gray Cat&lt;/span&gt;&lt;span class="tt"&gt;I Tested 4 Codex Frameworks on the Same Real Feature&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=tNKmMhiOw9w" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;13K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;Cole Medin&lt;/span&gt;&lt;span class="tt"&gt;Run Your AI Software Factory Like a Real Engineering Org (Here&amp;#x27;s How)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=9jjn54Me-Os" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;8K&lt;/span&gt;&lt;span class="dt"&gt;19 Eki&lt;/span&gt;&lt;span class="ch"&gt;Burke Holland&lt;/span&gt;&lt;span class="tt"&gt;I tested four AI strategies to find the cheapest one&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Ud5RsBegyfw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;7K&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;How Deep Agents Manages Context: Offloading and Summarization&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=CUjHZlvQufk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;7K&lt;/span&gt;&lt;span class="dt"&gt;20 Eki&lt;/span&gt;&lt;span class="ch"&gt;DeepLearningAI&lt;/span&gt;&lt;span class="tt"&gt;Learn to build better AI Agents through better context&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=tJ-hlNVS7AI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;6K&lt;/span&gt;&lt;span class="dt"&gt;19 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;What is LangChain Managed Deep Agents?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=wSSMODxxfbk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;5K&lt;/span&gt;&lt;span class="dt"&gt;20 Eki&lt;/span&gt;&lt;span class="ch"&gt;Burke Holland&lt;/span&gt;&lt;span class="tt"&gt;Copilot Can Use Your Computer for You #ai #coding #githubcopilot&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=bkj4Hc3sED4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;349&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;This AI Agent Builds Its OWN Tools to Finish Tasks! (INSANE Agent S4 Test)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=7Y6Fr_yl6Gw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;328&lt;/span&gt;&lt;span class="dt"&gt;24 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Microsoft&amp;#x27;s Agent World Is INSANE. I Ran It. The Agent FAILED (0/3)&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t14"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#d97706"&gt;14&lt;/span&gt;&lt;i class="dot" style="background:#d97706"&gt;&lt;/i&gt;Öğrenme, Bilim ve Araştırma&lt;span class="cnt"&gt;11 video &amp;middot; 81K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Zachary Huang&amp;#x27;ın bir saatlik derin öğrenme maratonları ve StatQuest canlı yayını eğitim tarafını taşırken, Paper2Agent&amp;#x27;ın her makaleyi MCP sunucusuna dönüştürmesi ve OlmoEarth uydu yerleştirmeleri araştırmanın yapay zekayla temas noktaları oldu.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=wbc-oxFFV9c" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;39K&lt;/span&gt;&lt;span class="dt"&gt;18 Eki&lt;/span&gt;&lt;span class="ch"&gt;Zachary Huang&lt;/span&gt;&lt;span class="tt"&gt;Give Me 1 Hour, I&amp;#x27;ll Make Deep Learning Click Forever&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=AgypoO13DJ0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;21K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;StatQuest with Josh Starmer&lt;/span&gt;&lt;span class="tt"&gt;Triple BAM! Live with Jon Krohn, Josh Starmer, and Luis Serrano!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=DWhgz1xDQ40" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;7K&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;Zachary Huang&lt;/span&gt;&lt;span class="tt"&gt;Give Me 1 Hour, I&amp;#x27;ll Make Multimodal Deep Learning Click Forever&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=qLEBXkHb_w4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;5K&lt;/span&gt;&lt;span class="dt"&gt;24 Eki&lt;/span&gt;&lt;span class="ch"&gt;Zachary Huang&lt;/span&gt;&lt;span class="tt"&gt;Give Me 35 Minutes, I&amp;#x27;ll Make LLM Pre-training Click Forever&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=StDAmjuQOE4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;23 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;The Boring Version of Continual Learning&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=UC9PdhD6Yxg" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;22 Eki&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;Automate CSV to Parquet with Buckets, Webhooks &amp;amp; Jobs&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=KnDMnZYEo2M" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;I Turned Structured Output Into a Game &#127918;&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=eTNyW0cBqRI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;24 Eki&lt;/span&gt;&lt;span class="ch"&gt;DigitalSreeni&lt;/span&gt;&lt;span class="tt"&gt;Density functional theory meets AI: traditional DFT vs a pretrained model&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=s9Sl86wE_5c" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;908&lt;/span&gt;&lt;span class="dt"&gt;22 Eki&lt;/span&gt;&lt;span class="ch"&gt;DigitalSreeni&lt;/span&gt;&lt;span class="tt"&gt;15 - Spatial Pharmacokinetics: Measuring Drug Diffusion in Tumor Tissue&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=9xejboeNY_E" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;275&lt;/span&gt;&lt;span class="dt"&gt;24 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;OlmoEarth v1.3: INSANE Satellite AI or Fooled by WINTER? I Tested It on My RTX 4060&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=XgGHb0HYQ60" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;248&lt;/span&gt;&lt;span class="dt"&gt;24 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;This Nature Paper Turns Any Research Paper Into an AI Agent - INSANE&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t15"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#dc2626"&gt;15&lt;/span&gt;&lt;i class="dot" style="background:#dc2626"&gt;&lt;/i&gt;Diğer Öne Çıkanlar&lt;span class="cnt"&gt;5 video &amp;middot; 113K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Messi&amp;#x27;nin satranç dersi, robot vücut koruma demosu, paralel ajan odaklanmasını anlatan itiraflar ve &amp;#x27;en ucuz AI stratejisi&amp;#x27; karşılaştırması haftanın renkli ama düşündürücü kalanlarıydı: hız artarken dikkat ve insan rolü yeni darboğaz.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=5eJnIYrr8rg" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;40K&lt;/span&gt;&lt;span class="dt"&gt;21 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Actual Robot Bodyguard&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=RLQR5_5fJDA" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;39K&lt;/span&gt;&lt;span class="dt"&gt;22 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Messi asks ChatGPT to teach him a chess opening&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=zj1TJlP7c9Y" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;25K&lt;/span&gt;&lt;span class="dt"&gt;22 Eki&lt;/span&gt;&lt;span class="ch"&gt;Cole Medin&lt;/span&gt;&lt;span class="tt"&gt;AI Makes Us WAY Faster but Fries Our Focus (Here&amp;#x27;s What I Do About It)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=UwrZJpjWjdI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;9K&lt;/span&gt;&lt;span class="dt"&gt;20 Eki&lt;/span&gt;&lt;span class="ch"&gt;Dave Ebbelaar&lt;/span&gt;&lt;span class="tt"&gt;Answering your questions about AI, agents, and tech jobs (Q&amp;amp;A)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=AAdp41dRluA" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;140&lt;/span&gt;&lt;span class="dt"&gt;22 Eki&lt;/span&gt;&lt;span class="ch"&gt;Build w Agents - Aggregate Intellect - AI.SCIENCE&lt;/span&gt;&lt;span class="tt"&gt;This AI Doesn’t Just Remind You, It Knows When to Back Off.&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;
&lt;footer&gt;Bu rapor &lt;b style="color:#111"&gt;Murat Karakaya Akademi&lt;/b&gt; tarafından derlenmiştir · Veri: YouTube RSS (canlı, API key yok) · pencere: 2026-10-18 .. 2026-10-25 (UTC, çalışma anından hesaplanır) · kanal başı max 15 video · Prompt Engineer 48 ve OpenAI beslemeleri RSS 15-video sınırına takıldı (partial); bu kanalların pencerenin başındaki tüm videoları görülemeyebilir. Pencere: 18-25 Ekim 2026 (UTC), 43 kanaldan 24&amp;#x27;ü aktif, 122 video. · render.py @ 27.10.2026 23:03 UTC&lt;/footer&gt;
&lt;/div&gt;&lt;/body&gt;&lt;/html&gt;</description><link>https://www.muratkarakaya.net/2026/10/youtube-ai-haftalk-trend-raporu-18-25.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-7152627624721759332</guid><pubDate>Thu, 22 Oct 2026 13:54:09 +0000</pubDate><atom:updated>2026-10-27T19:04:21.337+03:00</atom:updated><category domain="http://www.blogger.com/atom/ns#">Açik Agirlik</category><category domain="http://www.blogger.com/atom/ns#">M31 Flash</category><category domain="http://www.blogger.com/atom/ns#">MiniMax</category><category domain="http://www.blogger.com/atom/ns#">Murat Karakaya Akademi</category><category domain="http://www.blogger.com/atom/ns#">Ollama</category><category domain="http://www.blogger.com/atom/ns#">Yapay Zeka</category><title>MiniMax ajan modeli M3.1 Flash'ı açtı: hız iddiası bağımsız ölçümün gerisinde kaldı, ağırlıklar henüz yayımlanmadı</title><description>&lt;!--MIS-2026-004 rev.6 — ANAYASA: içerikte sıfır görünüm CSS'i (renk, font, zemin,
     STYLE bloğu ve satır-içi stil niteliği YOK). Görünüm sahibi Blogger teması. Yalnız anlamsal
     HTML. Başlığı tema basar (Bu yazının başlığı Blogger BAŞLIK ALANINDA durur:
     "MiniMax ajan modeli M3.1 Flash'ı açtı: hız iddiası bağımsız ölçümün gerisinde
     kaldı, ağırlıklar henüz yayımlanmadı"). Başlık metni referansı buradadır,
     görünür metin tema üzerinden basılır.--&gt;
&lt;p&gt;Maskeli "Space Bunny" sürpriziyle gelen lansman, kapalı ağırlıklı bir modeli birçok açık rakibinin önüne taşıyor; bağımsız ölçüm ise iddia edilen hızın yarısını ve lisans belirsizliğini gösteriyor.&lt;/p&gt;
&lt;p&gt;6 dk okuma&lt;/p&gt;
&lt;!--SLOT:ICINDEKILER--&gt;
&lt;p&gt;&lt;strong&gt;Bu Yazıda&lt;/strong&gt;&lt;/p&gt;
&lt;ol&gt;
  &lt;li&gt;&lt;a href="#bol-1"&gt;Maskenin arkasındaki laboratuvar&lt;/a&gt;&lt;/li&gt;
  &lt;li&gt;&lt;a href="#bol-2"&gt;Bir model, iki hız sayısı, tek dürüst okuma&lt;/a&gt;&lt;/li&gt;
  &lt;li&gt;&lt;a href="#bol-3"&gt;Puan tablosu: ucuz, hızlı sayılmaz, güvenilir de sayılmaz&lt;/a&gt;&lt;/li&gt;
  &lt;li&gt;&lt;a href="#bol-4"&gt;Pazarlama takvimi, lisans masasından önce geldi&lt;/a&gt;&lt;/li&gt;
  &lt;li&gt;&lt;a href="#ka-yorum"&gt;Murat Karakaya Akademi Yorumu&lt;/a&gt;&lt;/li&gt;
  &lt;li&gt;&lt;a href="#kaynakca"&gt;Kaynakça&lt;/a&gt;&lt;/li&gt;
&lt;/ol&gt;
&lt;!--/SLOT:ICINDEKILER--&gt;
&lt;!--SLOT:GOVDE--&gt;
&lt;h2 id="bol-1"&gt;1. Maskenin arkasındaki laboratuvar&lt;/h2&gt;
&lt;p&gt;OpenRouter'da on günü aşkın süre kimliksiz bir model dolaştı; adı "Space Bunny"ydi, kime ait olduğu bilinmiyordu. İşlediği 200 trilyondan fazla token ve global liderlik tablolarının tepesine yerleşmesi konuşuldu. 19 Ekim'de MiniMax çıkıp maskeli modelin kendi kod ve ajan modeli M3.1 Flash'ın erken sürümü olduğunu duyurdu [1][2][15]. Model o günden beri MiniMax'ın API'sinde, M Plan aboneliğinde ve MiniMax Code uygulamasında erişime açık [1][9]. Ağırlıklar ise ortada yok: şirket "sonra yayımlanacak" diyor, lisans hakkında tek kelime etmiyor [1][7]. Bir duyuru yazısı için bunlar yeterdi; meselenin boyutu başka. Kapalı ağırlıklı bir model, bağımsız puanlamada açık ağırlıklı rakiplerinin tam önüne geçmek üzereyken sahneye çıktı ve ilk hafta erişimini bir bulut kampanyasına bağladı.&lt;/p&gt;
&lt;p&gt;MiniMax, Çin merkezli ve borsaya kote bir şirket (HKEX: 00100). Lansmanda tanıdık bir numara var: modeli önce anonim dolaşıma ver, merakı büyüt, sonra resmî blog yazısıyla indir. OpenRouter topluluğu haftalarca "Space Bunny"nin kim olduğunu tartıştı; duyuru geldiğinde sürprizin kendisi de resmîleşti [1][8]. Aynı gün MiniMax'ın X hesabı, modelin Ollama'da "Day0" desteğiyle çıktığını müjdeledi [8]. Ollama ile çalışmak yeni bir alışkanlık değil zaten: M2 de Ekim 2025'te Ollama bulutuna eklenmişti [6].&lt;/p&gt;
&lt;p&gt;Sayılar tarafında şirket "Flash" segmentini baştan konumluyor. Eski M3, Artificial Analysis'te 29 Index puanında ve görev başına ~$0,51 maliyetteydi; M3.1 Flash aynı tabloda 43 puana, ~$0,22'ye oturdu [11]. MiniMax'ın "yarı fiyat, üç kat hız" söylemi bu iki satır arasındaki mesafeye dayanıyor ve şimdilik şirketin kendi iddiası; bağımsız tekrarı yapılmış değil [1][15].&lt;/p&gt;
&lt;p&gt;Kendi benchmark tablosu da aynı kategoriye giriyor: Automation-Bench'te 62.2, DeepSWE'de 73.7, BankerToolBench'te 76.2. Terminal Bench 4.0'da 38.1'e düşmesi, üstelik Opus 5'in 51.8'ine karşılık, tablonun gösterişsiz köşesi [1][12][13]. Bu rakamları bir pazarlama dokümanından okuduğumuzu akılda tutmak lazım; aşağıdaki bağımsız tablo başka bir hesap.&lt;/p&gt;
&lt;h2 id="bol-2"&gt;2. Bir model, iki hız sayısı, tek dürüst okuma&lt;/h2&gt;
&lt;p&gt;Hız bu lansmanın en karışık başlığı. MiniMax'ın blog yazısı 200'ün üzerinde TPS tepeden ve P50'de 170+ değerden söz ediyor [1]; bu, şirketin kendi altyapısında kendi ölçümü. Bağımsız ölçüm kuruluşu Artificial Analysis, 21 Ekim akşamı çıktı hızını ~93 token/saniye gösteriyordu; sınıf medyanı 113. İddia edilenin yaklaşık yarısı hız, ortalamanın altında bir yer [17]. (Kayıt düşümü: AA'nın model sayfasının çekilebilen anlık görüntüsü eski bir hız değerinde donmuş durumda; güncel sayı için karşılaştırma sayfası esas alındı [10][17]. Ölçüm zaman içinde değişken, bu yazı "~93, erişim 21.10" der ve noktasını koyar.)&lt;/p&gt;
&lt;p&gt;Bağlam penceresinde de benzer bir pazarlama/pratik makası var. MiniMax dokümanı ve AA 1.000.000 token'dan söz ederken [7][10], Ollama fiilen kullanılabilir tavanı 976K olarak listeliyor [3]. "1M bağlam" ürünün adı; hesabınızda kalan rakam 976K.&lt;/p&gt;
&lt;p&gt;Modelin bildiğimiz künyesi: Ollama sayfası 428B parametre gösteriyor [3]. ~23B aktif parametreli bir MoE olduğu, sparse attention ve yerleşik görsel kodlayıcı taşıdığı MarkTechPost'un aktardığı ayrıntılar; MiniMax teknik rapor yayımlamadığı için doğrulayamıyoruz [12]. Girdi tarafında metin ve görüntü güvenli bilgi; doküman videodan da söz ediyor ama Ollama ucunda video çalıştığını gösteren veri yok [3][7][10]. Düşünme seviyeleri &lt;code&gt;low&lt;/code&gt;'dan &lt;code&gt;max&lt;/code&gt;'e beş kademeli ve reasoning hep açık [7]. Eski M3 entegrasyonlarındaki &lt;code&gt;thinking.type=disabled&lt;/code&gt; / &lt;code&gt;effort=none&lt;/code&gt; çağrıları artık hata veriyor, düşünce metni ayrı &lt;code&gt;reasoning_content&lt;/code&gt; alanında geliyor; bu kırılmayı migrasyon yapanlar ilk çağrıda görecek [12].&lt;/p&gt;
&lt;h2 id="bol-3"&gt;3. Puan tablosu: ucuz, hızlı sayılmaz, güvenilir de sayılmaz&lt;/h2&gt;
&lt;p&gt;Artificial Analysis'in 21.10 erişimli verileriyle sınıfın fotoğrafı şöyle:&lt;/p&gt;
&lt;table&gt;
  &lt;thead&gt;&lt;tr&gt;&lt;th&gt;Model&lt;/th&gt;&lt;th&gt;Intelligence Index&lt;/th&gt;&lt;th&gt;Görev maliyeti&lt;/th&gt;&lt;th&gt;Not&lt;/th&gt;&lt;/tr&gt;&lt;/thead&gt;
  &lt;tbody&gt;
    &lt;tr&gt;&lt;td&gt;MiniMax M3.1 Flash (max)&lt;/td&gt;&lt;td&gt;42→43 (güncel sayfa: 43)&lt;/td&gt;&lt;td&gt;~$0,22&lt;/td&gt;&lt;td&gt;4/173; HLE %48, CritPt %22; hız 93 tok/s (medyan altı)&lt;/td&gt;&lt;/tr&gt;
    &lt;tr&gt;&lt;td&gt;GLM-5.3-Flash (max)&lt;/td&gt;&lt;td&gt;42&lt;/td&gt;&lt;td&gt;~$0,28&lt;/td&gt;&lt;td&gt;Lansman analizinde M3.1 ile başa baş&lt;/td&gt;&lt;/tr&gt;
    &lt;tr&gt;&lt;td&gt;DeepSeek V4.1 Flash 1017 (max)&lt;/td&gt;&lt;td&gt;45&lt;/td&gt;&lt;td&gt;~$0,26&lt;/td&gt;&lt;td&gt;Sınıfın AA lideri; MIT lisanslı açık ağırlık&lt;/td&gt;&lt;/tr&gt;
    &lt;tr&gt;&lt;td&gt;Gemini 3.8 Flash (high)&lt;/td&gt;&lt;td&gt;41&lt;/td&gt;&lt;td&gt;~$1,24&lt;/td&gt;&lt;td&gt;M3.1'in ~5,6 katı görev maliyeti&lt;/td&gt;&lt;/tr&gt;
    &lt;tr&gt;&lt;td&gt;MiniMax M3 (eski nesil)&lt;/td&gt;&lt;td&gt;29&lt;/td&gt;&lt;td&gt;~$0,51&lt;/td&gt;&lt;td&gt;Bir önceki nesil&lt;/td&gt;&lt;/tr&gt;
    &lt;tr&gt;&lt;td&gt;Sınıf medyanı&lt;/td&gt;&lt;td&gt;13&lt;/td&gt;&lt;td&gt;—&lt;/td&gt;&lt;td&gt;—&lt;/td&gt;&lt;/tr&gt;
  &lt;/tbody&gt;
&lt;/table&gt;
&lt;p&gt;&lt;em&gt;Tablo ve şekil: Artificial Analysis model/karşılaştırma sayfaları ve lansman analizi verileri [10][11][17] · erişim 21.10.2026.&lt;/em&gt;&lt;/p&gt;
&lt;!--SLOT:SEKIL1 — (rev.6 anayasası) stilli SVG şekil kaldırıldı; grafiğin tüm
     verisi yukarıdaki tabloda eksiksiz duruyor (Sınıf medyanı 13 satırı dahil).
     Şekil notu: Şekil: Artificial Analysis Intelligence Index ve görev başına
     tahmini maliyet [10][11][17]. MiniMax M3 (eski nesil, 29 puan) tabloda.--&gt;
&lt;p&gt;Tablonun satır aralarını okumak lazım. Rakip satırındaki GLM, GLM-5.3-Flash (42 puan, ~$0,28); aynı ismi taşıyan "GLM-5.3 (Max)" ayrı ve iri bir model, görev başına $2,01 ödetiyor [11][17]. DeepSeek V4.1 Flash hem sınıfın lideri hem MIT lisanslı: ağırlıkları indirip kendi sunucunuzda çalıştırabilirsiniz [11][12]. M3.1 Flash'ın 43 puanı etkileyici; karşılığında bir bulut servisinin kuyruğunda, milyon token başına girdide $0.15, çıktıda $0.60 ödüyorsunuz [1][3][7].&lt;/p&gt;
&lt;p&gt;Bir de halüsinasyon ölçümü var, reklamı hiç yapılmıyor. AA-Omniscience'ta M3.1 Flash soruların %98'ini yanıtlamayı deniyor, %56 doğruda kalıyor; bilmediği soruların %93'ünde ise "bilmiyorum" demek yerine yanlış cevap veriyor (genel puan 16) [11]. VentureBeat'in de altını çizdiği gibi bu, "cevapların %93'ü yanlış" demek değil [18]. Anlamı şu: model emin olmadığı yerde uyduruyor. Ajan zincirlerinde sessizce yanlış giden bir halka istemiyorsanız, çıktıyı doğrulayacak mekanizmayı kurmak sizin işiniz.&lt;/p&gt;
&lt;h2 id="bol-4"&gt;4. Pazarlama takvimi, lisans masasından önce geldi&lt;/h2&gt;
&lt;p&gt;Neden açık ağırlık değil de bulut? Resmî cevap yok; tablo kendi hikâyesini anlatıyor. DeepSeek MIT lisansıyla herkese açıkken, MiniMax ağırlıkları "sonraya" bırakıp ilk haftayı Ollama bulutunda üç günlük ücretsiz kampanyayla açtı [1][11]. Ollama'nın model sayfası "free until Oct 22, 2026 11:59 PM UTC" diyor, yani Türkiye saatiyle 23 Ekim 02:59'a kadar ücretsiz; sonrası yukarıdaki tarifeye dönüyor [3][5]. Kampanyanın kota gibi ek koşulları hiçbir yerde yazmıyor [3][5]. Küçük bir kayıt: Ollama'nın blogunda bu lansmanla ilgili tek bir yazı yok; duyuru model sayfası ve iki X göndermesinden ibaret [4][5][6].&lt;/p&gt;
&lt;blockquote&gt;"Model weights will be released later." — MiniMax resmî blog duyurusu, 19.10.2026 [1]&lt;/blockquote&gt;
&lt;p&gt;Bu ayrıntının habercilikteki ağırlığı tek cümle: ücretsiz pencere bir pazarlama bütçesidir, modelin kendisi değil. Modelin gerçek tartışması lisans belirsizliğinde dönüyor. MiniMax dokümanı modeli "open-weight" olarak tanımlıyor [7]; Venice'in künyesi "kapalı/proprietary" yazıyor [13]. İkisi aynı anda doğru olamaz ve hiçbir resmî kaynak lisansın adını ağzına almıyor. Önceki M3'ün Community License'ı atıf zorunluluğu getiriyor, yıllık 20 milyon doları aşan gelire yazılı izin şartı koyuyordu; M3.1'de neyin geçerli olduğu açıklanmadı [13][14]. Ollama'da tek etiket &lt;code&gt;:cloud&lt;/code&gt;, model ABD'deki Ollama altyapısında çalışıyor; "sıfır veri saklama" ise Ollama'nın politik beyanı olarak duruyor [3][4]. "Yerel çalıştır" cümlesini bugünkü haliyle kurmak yanlış olur.&lt;/p&gt;
&lt;p&gt;Bugün yapılabilecek olan ölçülü bir test. Yazı hazırlandığında ücretsiz pencereye yaklaşık bir gün var (bitiş 23 Ekim 02:59 TRT); deneyecek olanlar için makul sıralama şu: &lt;code&gt;ollama run minimax-m3.1-flash:cloud&lt;/code&gt; ile tek bir uçtan uca kod görevi koşun, sonra &lt;code&gt;ollama launch claude --model minimax-m3.1-flash:cloud&lt;/code&gt; ile Claude Code'u bağlayıp bir depoda PR ölçeğinde iş verin [3][4]. Aynı görevleri MIT lisanslı DeepSeek V4.1 Flash ile karşılaştırın; terminal-ağır işlere ve olgu sorularına özellikle bakın, iki bilinen zayıf nokta oralarda [11][12]. Takvimde ~23 Ekim'de ağırlık yayımı bekleniyor; çıkarsa ilk işiniz lisans metnini okumak olsun [13][15].&lt;/p&gt;
&lt;!--/SLOT:GOVDE--&gt;
&lt;!--SLOT:KA_YORUM--&gt;
&lt;h2 id="ka-yorum"&gt;Murat Karakaya Akademi Yorumu&lt;/h2&gt;
&lt;p&gt;Verilerin söylediği şu: 43 Intelligence Index ve ~$0,22 görev maliyeti, sınıf medyanının (13 puan) üç katından fazla bir performans için çok düşük bir fiyat demek; bunu bağımsız ölçüm söylüyor [10][11]. Aynı verilerin çizdiği sınır da net: %56 doğruda kalan olgu bilgisi ve bilmediğinde %93 oranında uyduran yanıt politikası, modeli doğrulaması olmayan iş akışlarının dışına koymayı zorunlu kılıyor [11][18]. Şirketin kendi benchmark tablosu bağımsız tekrarı yapılmamış bir iddialar demeti; ikisini ayırt etmek zor ama mümkün, ve biz yazı boyunca ayırdık.&lt;/p&gt;
&lt;p&gt;Kanaatimize gelince: bu lansman model rekabetinden çok dağıtım rekabeti hakkında konuşuyor. Bir laboratuvar en güçlü kod modelini maskeli bir stealth sürümle pazarlıyor, ağırlıklarını kilitte tutuyor ve ilk erişimi bir bulut ortağının ücretsiz kampanyasına bağlayarak açık ağırlık camiasının yerel çalışma avantajını fiilen ilk hafta devre dışı bırakıyor. Lisans metni yayımlanana kadar bu model bir ürün, bir altyapı değil. Okura önerimiz: ücretsiz günde kör teste sokun, izleniminizi ölçülebilir ve lisansı temiz olan DeepSeek ile kıyaslayın, üretim kararını ağırlıklar ve lisans ortaya çıkınca verin.&lt;/p&gt;
&lt;!--/SLOT:KA_YORUM--&gt;
&lt;!--SLOT:KAYNAKCA--&gt;
&lt;h2 id="kaynakca"&gt;Kaynakça&lt;/h2&gt;
&lt;ol&gt;
  &lt;li&gt;minimax.io — &lt;a href="https://www.minimax.io/blog/minimax-m31f"&gt;"MiniMax M3.1 Flash: Redefining Agent Performance and Cost"&lt;/a&gt; · 19.10.2026 · erişim: 21.10.2026&lt;/li&gt;
  &lt;li&gt;ir.minimax.io — &lt;a href="https://ir.minimax.io/news-events/new-releases"&gt;"New Releases" duyuru listesi&lt;/a&gt; · 19.10.2026 · erişim: 21.10.2026&lt;/li&gt;
  &lt;li&gt;ollama.com — &lt;a href="https://ollama.com/library/minimax-m3.1-flash"&gt;"minimax-m3.1-flash" model sayfası&lt;/a&gt; · erişim: 21.10.2026&lt;/li&gt;
  &lt;li&gt;x.com/@ollama — &lt;a href="https://x.com/ollama/status/2112249324398858505"&gt;Resmî lansman göndermesi (1M ctx, US-hosted, zero data retention)&lt;/a&gt; · 19.10.2026 · erişim: 21.10.2026&lt;/li&gt;
  &lt;li&gt;x.com/@ollama — &lt;a href="https://x.com/ollama/status/2112358514228363362"&gt;"3 days free on Ollama's Cloud"&lt;/a&gt; · 20.10.2026 · erişim: 21.10.2026&lt;/li&gt;
  &lt;li&gt;ollama.com/blog — &lt;a href="https://ollama.com/blog"&gt;Blog arşivi (MiniMax M2 2025 yazısı; M3.1 yazısının yokluğu)&lt;/a&gt; · erişim: 21.10.2026&lt;/li&gt;
  &lt;li&gt;platform.minimax.io — &lt;a href="https://platform.minimax.io/docs/guides/text-generation"&gt;Text generation dokümantasyonu&lt;/a&gt; · erişim: 21.10.2026&lt;/li&gt;
  &lt;li&gt;x.com/@MiniMax_AI — &lt;a href="https://x.com/MiniMax_AI/status/2112281693642227753"&gt;Day0 Ollama desteği + Space Bunny doğrulaması&lt;/a&gt; · 19.10.2026 · erişim: 21.10.2026&lt;/li&gt;
  &lt;li&gt;agent.minimax.io — &lt;a href="https://agent.minimax.io/tools/m3.1-flash"&gt;MiniMax Agent ürün sayfası (M3.1 Flash GA duyurusu)&lt;/a&gt; · erişim: 21.10.2026&lt;/li&gt;
  &lt;li&gt;artificialanalysis.ai — &lt;a href="https://artificialanalysis.ai/models/minimax-m3-1-flash"&gt;"MiniMax-M3.1-Flash" model sayfası&lt;/a&gt; · erişim: 21.10.2026&lt;/li&gt;
  &lt;li&gt;x.com/@ArtificialAnlys — &lt;a href="https://x.com/ArtificialAnlys/status/2112211865128608002"&gt;Lansman benchmark analizi&lt;/a&gt; · erişim: 21.10.2026&lt;/li&gt;
  &lt;li&gt;marktechpost.com — &lt;a href="https://www.marktechpost.com/2026-10-19/minimax-m3-1-flash-release/"&gt;"MiniMax M3.1 Flash" (428B MoE, ~23B aktif iddiası, migrasyon notları)&lt;/a&gt; · 19.10.2026 · erişim: 21.10.2026&lt;/li&gt;
  &lt;li&gt;pc.watch.impress.co.jp — &lt;a href="https://pc.watch.impress.co.jp/docs/news/2149440.html"&gt;Space Bunny ifşası, lisans analizi, M3 Community License koşulları&lt;/a&gt; · erişim: 21.10.2026&lt;/li&gt;
  &lt;li&gt;faradaymachines.com — &lt;a href="https://www.faradaymachines.com/minimax-m3-1-flash-cost-per-task/"&gt;Lisans boşluğu ve task-başına-maliyet analizi&lt;/a&gt; · erişim: 21.10.2026&lt;/li&gt;
  &lt;li&gt;unite.ai — &lt;a href="https://www.unite.ai/minimax-launches-m3-1-flash-revealing-space-bunny-stealth-model/"&gt;"MiniMax Launches M3.1 Flash, Revealing Space Bunny Stealth Model"&lt;/a&gt; · erişim: 21.10.2026&lt;/li&gt;
  &lt;li&gt;daily.dev — &lt;a href="https://daily.dev/posts/minimax-m3-1-flash-is-cheap-free-this-week-and-everywhere-7uccejy7e"&gt;Ücretsiz hafta + karışık benchmark manzarası derlemesi&lt;/a&gt; · erişim: 21.10.2026&lt;/li&gt;
  &lt;li&gt;artificialanalysis.ai — &lt;a href="https://artificialanalysis.ai/models/comparisons/minimax-m3-1-flash-vs-glm-5-3"&gt;"M3.1-Flash vs GLM-5.3 (Max)" karşılaştırma sayfası&lt;/a&gt; · erişim: 21.10.2026&lt;/li&gt;
  &lt;li&gt;venturebeat.com — &lt;a href="https://venturebeat.com/technology/minimaxs-new-m3-1-flash-beats-deepseek-and-opus-5-on-a-key-business-benchmark-at-lower-cost-but-with-a-high-hallucination-rate"&gt;"MiniMax's new M3.1 Flash beats DeepSeek and Opus 5 on a key business benchmark at lower cost, but with a high hallucination rate"&lt;/a&gt; · erişim: 21.10.2026&lt;/li&gt;
&lt;/ol&gt;
&lt;!--/SLOT:KAYNAKCA--&gt;
&lt;!--SLOT:HASHTAG — düz metin şerit (rev.6 anayasası: stil yok); CEO bunları Blogger
     etiket (labels) alanına da taşır: ilki her zaman MuratKarakayaAkademi--&gt;
&lt;p&gt;Etiketler: #MuratKarakayaAkademi #MiniMax #M31Flash #Ollama #YapayZeka #AcikAgirlik&lt;/p&gt;
&lt;!--/SLOT:HASHTAG--&gt;
&lt;!--SLOT:FOOTER — IMZA BURADA (tek satır, düz metin):--&gt;
&lt;footer&gt;
  &lt;p&gt;Yayın tarihi: 22 Ekim 2026 · muratkarakaya.net&lt;/p&gt;
  &lt;p&gt;Murat Karakaya Akademi&lt;/p&gt;
&lt;/footer&gt;
&lt;!--/SLOT:FOOTER--&gt;
&lt;p&gt;&lt;em&gt;Read this article in English: &lt;a href="https://www.muratkarakaya.net/2026/10/minimaxs-m31-flash-sits-ahead-of.html"&gt;MiniMax's M3.1 Flash sits ahead of several open-weight rivals&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;</description><link>https://www.muratkarakaya.net/2026/10/minimax-ajan-modeli-m31-flash-act-hz_01756526152.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-7460936925081811657</guid><pubDate>Fri, 16 Oct 2026 19:25:26 +0000</pubDate><atom:updated>2026-10-16T22:25:26.667+03:00</atom:updated><title>YouTube AI Haftalık Trend Raporu (2026-10-09 → 2026-10-16)</title><description>&lt;!doctype html&gt;&lt;html lang="tr"&gt;&lt;head&gt;&lt;meta charset="utf-8"&gt;
&lt;meta name="viewport" content="width=device-width,initial-scale=1"&gt;
&lt;title&gt;YouTube AI Haftalık Trend Raporu&lt;/title&gt;&lt;style&gt;
:root{--bg:#0b0e12;--sf:#12161c;--sf2:#171c24;--ln:#222935;--ink:#e7eaee;--mu:#8d97a3;
--ac:#35d0ba;--am:#ffb454;--mono:ui-monospace,Consolas,monospace}
*{box-sizing:border-box;margin:0;padding:0}
body{background:var(--bg);color:var(--ink);font:15px/1.5 -apple-system,'Segoe UI',Roboto,sans-serif;padding:28px 24px 60px}
.wrap{max-width:1100px}
h1{font-size:32px;letter-spacing:.3px;text-align:center;margin-top:8px} .datebar{text-align:center;margin:12px 0 6px}
.dtc{display:inline-block;font:600 15px/1 var(--mono);color:#0b0e12;background:var(--ac);padding:7px 16px;border-radius:999px}
.meta{color:var(--mu);font-family:var(--mono);font-size:13.5px;margin-top:2px;text-align:center}
.dseg{cursor:pointer;transition:stroke-width .12s} .dseg:hover{stroke-width:36}
#tip{position:fixed;left:0;top:0;z-index:9;pointer-events:none;background:#1a2029;border:1px solid #2b3644;color:#e7eaee;font:12.5px/1.45 var(--mono);padding:7px 11px;border-radius:8px;opacity:0;transition:opacity .12s;max-width:340px;box-shadow:0 6px 20px rgba(0,0,0,.5)}
.meta b{color:var(--ac);font-weight:600}
nav{margin:14px 0 26px;display:flex;flex-wrap:wrap;gap:8px}
nav a{color:var(--mu);text-decoration:none;font-size:12px;padding:4px 10px;border:1px solid var(--ln);border-radius:999px;line-height:1.35} nav a b{color:var(--ac);font-family:var(--mono);font-weight:600;margin-left:4px}
nav a:hover{color:var(--ink);border-color:var(--ac)}
.card{background:var(--sf);border:1px solid var(--ln);border-radius:12px;padding:16px 18px}
.card h3,.stats h3{font-size:12px;text-transform:uppercase;letter-spacing:1.2px;color:var(--mu);margin-bottom:10px;font-weight:600}
.top{display:grid;grid-template-columns:2fr 1fr;gap:14px;margin-bottom:14px}
.champ{background:linear-gradient(135deg,#131a20,#10151b);border:1px solid #2b3644;border-radius:12px;padding:20px;text-decoration:none;color:var(--ink);display:block}
.champ:hover{border-color:var(--am)}
.ck{font-size:11px;letter-spacing:1.5px;text-transform:uppercase;color:var(--am)}
.cv{font:600 44px/1.1 var(--mono);color:var(--am);margin:6px 0}
.ct{font-size:17px;font-weight:600} .cs{color:var(--mu);font-size:13px;margin-top:4px}
.cd{color:#b8c1cb;font-size:12.5px;line-height:1.55;margin-top:10px;border-top:1px solid var(--ln);padding-top:10px}
.stats{background:var(--sf);border:1px solid var(--ln);border-radius:12px;padding:14px 18px;display:flex;flex-direction:column;justify-content:space-between}
.st{display:flex;justify-content:space-between;align-items:baseline;padding:5px 0;border-bottom:1px solid var(--ln)}
.st:last-child{border:0} .st b{font:600 19px var(--mono)} .st span{color:var(--mu);font-size:12.5px}
.credit{text-align:center;margin-top:14px;font:600 14px var(--sans);color:#c8d0d9;letter-spacing:.2px}
.credit b{color:var(--am);font-weight:800}
.rowg{display:grid;gap:14px;margin-bottom:14px}
.g23{grid-template-columns:2fr 1fr} .g11{grid-template-columns:1.2fr 1fr}
@media(max-width:820px){.top,.g23,.g11{grid-template-columns:1fr}}
svg{width:100%;height:auto;display:block}
.ax{font:11px var(--mono);fill:#8d97a3} .bn2{font:600 11px var(--mono);fill:#9aa3ad}
.dn{font:700 30px var(--mono);fill:#e7eaee}
.donut{max-width:220px;margin:0 auto}
.lgs{display:flex;flex-wrap:wrap;gap:6px 14px;margin-top:12px}
.lg{font-size:11.5px;color:var(--mu)} .lg i{display:inline-block;width:9px;height:9px;border-radius:2px;margin-right:5px} .lg b{color:#e7eaee;font-family:var(--mono)}
.cbar{display:grid;grid-template-columns:130px 1fr 90px;gap:10px;align-items:center;padding:7px 0;border-bottom:1px solid var(--ln)}
.cbar:last-child{border:0} .cn{font-size:12.5px;color:#b8c1cb;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
.ct2{background:var(--sf2);height:10px;border-radius:4px;overflow:hidden} .cf{display:block;height:100%;background:var(--ac);opacity:.85;border-radius:4px}
.cv2{font:600 12.5px var(--mono);text-align:right} .cv2 i{font-style:normal;color:var(--mu);font-size:10.5px;margin-left:6px}
h2.sec{font-size:13px;text-transform:uppercase;letter-spacing:1.4px;color:var(--mu);margin:26px 0 10px}
ol.heads{list-style:none;display:flex;flex-direction:column;gap:8px}
ol.heads li{display:flex;gap:14px;background:var(--sf);border:1px solid var(--ln);border-radius:10px;padding:13px 16px}
ol.heads .hr{font:700 20px var(--mono);color:var(--ac);min-width:26px}
ol.heads li.h2 .hr{color:#6fb9d6} ol.heads li.h3 .hr{color:#5d7fa3}
ol.heads li.h1{border-color:#2e4a45}
ol.heads p{font-size:14.5px} ol.heads li.h1 p{font-size:16px;font-weight:600} ol.heads li.h3 p{color:var(--mu)}
.bars{display:flex;flex-direction:column;gap:4px}
.bar{display:grid;grid-template-columns:190px 1fr 30px;gap:12px;align-items:center;text-decoration:none;color:var(--ink);padding:4px 0}
.bar:hover .bn{color:var(--ac)} .bn{font-size:12.5px;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
.bt{background:var(--sf2);border-radius:4px;height:10px;overflow:hidden} .bf{display:block;height:100%;opacity:.9;border-radius:4px}
.bv{font:600 12.5px var(--mono);text-align:right;color:var(--mu)}
@media(max-width:640px){.bar{grid-template-columns:130px 1fr 28px}}
.theme{margin-top:30px} .theme h2{font-size:17px;display:flex;align-items:baseline;gap:10px}
.tn{font:700 15px var(--mono)} .dot{width:9px;height:9px;border-radius:3px;display:inline-block}
.cnt{font:400 12.5px var(--mono);color:var(--mu);margin-left:auto}
.insight{color:#b8c1cb;font-size:13.5px;margin:6px 0 10px;padding-left:2px;max-width:860px}
.rows{border:1px solid var(--ln);border-radius:10px;overflow:hidden}
.row{display:grid;grid-template-columns:64px 56px 150px 1fr;gap:12px;align-items:baseline;padding:9px 14px;border-bottom:1px solid var(--ln);text-decoration:none;color:var(--ink);background:var(--sf)}
.row:last-child{border:0} .row:hover{background:var(--sf2)}
.vw{font:600 12.5px var(--mono);text-align:right}
.v1{color:var(--am)} .v2{color:var(--ac)} .v3{color:var(--mu)}
.dt{font:12px var(--mono);color:var(--mu)} .ch{font-size:12px;color:#7f9ab5;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
.tt{font-size:13.5px} .other .tt{color:var(--mu)}
footer{margin-top:34px;color:#5c6672;font:12px/1.7 var(--mono);border-top:1px solid var(--ln);padding-top:12px}
&lt;/style&gt;&lt;/head&gt;&lt;body&gt;&lt;div class="wrap"&gt;
&lt;h1&gt;YouTube AI Haftalık Trend Raporu&lt;/h1&gt;
&lt;div class="datebar"&gt;&lt;span class="dtc"&gt;2026-10-09 &amp;rarr; 2026-10-16&lt;/span&gt;&lt;/div&gt;
&lt;div class="credit"&gt;Bu veriler &lt;b&gt;Murat Karakaya Akademi&lt;/b&gt; tarafından derlenmiştir.&lt;/div&gt;

&lt;div class="meta"&gt;&lt;b&gt;25&lt;/b&gt; aktif kanal / 43 &amp;nbsp;|&amp;nbsp; &lt;b&gt;120&lt;/b&gt; video &amp;nbsp;|&amp;nbsp; &lt;b&gt;3.9M&lt;/b&gt; toplam izlenme&lt;/div&gt;
&lt;h2 class="sec"&gt;Haftanın 3 manşeti&lt;/h2&gt;&lt;ol class="heads"&gt;&lt;li class="h1"&gt;&lt;span class="hr"&gt;1&lt;/span&gt;&lt;p&gt;Gemini 4 Argon sahneye çıktı: Google&amp;#x27;ın yeni frontier modeli bu hafta 15 ayrı videoda test edildi, Claude Opus 5.5 ile kıyaslandı ve Antigravity üzerinden 2D konseptten 3D katedrale projeler üretti.&lt;/p&gt;&lt;/li&gt;&lt;li class="h2"&gt;&lt;span class="hr"&gt;2&lt;/span&gt;&lt;p&gt;Yerel AI donanımı haftaya damga vurdu: Kai&amp;#x27;nin 128GB VRAM rehberi 550 bin izlenmeyi geçerken, &amp;#x27;GPU satın almayın&amp;#x27; uyarıları ve artan donanım maliyetleri ikinci büyük gündem oldu.&lt;/p&gt;&lt;/li&gt;&lt;li class="h3"&gt;&lt;span class="hr"&gt;3&lt;/span&gt;&lt;p&gt;Meta sessiz ama büyük bir hamle yaptı: frontier sınıfındaki Muse Wonder ve 816B&amp;#x27;lik açık ağırlıklı Muse Spark, GPT-6 ve Claude ile kıyaslanarak haftanın en çok konuşulan model lansmanlarından biri oldu.&lt;/p&gt;&lt;/li&gt;&lt;/ol&gt;
&lt;h2 class="sec"&gt;Haftanın temaları&lt;/h2&gt;&lt;nav&gt;&lt;a href="#t1" style="font-size:15.3px"&gt;Gemini 4 Argon fırtınası&lt;b&gt;17&lt;/b&gt;&lt;/a&gt; &lt;a href="#t2" style="font-size:13.0px"&gt;Meta&amp;#x27;dan Muse sürprizi&lt;b&gt;9&lt;/b&gt;&lt;/a&gt; &lt;a href="#t3" style="font-size:12.1px"&gt;Güvenlik, ihlal ve güven tartışması&lt;b&gt;6&lt;/b&gt;&lt;/a&gt; &lt;a href="#t4" style="font-size:12.7px"&gt;OCR ve belge anlama&lt;b&gt;8&lt;/b&gt;&lt;/a&gt; &lt;a href="#t5" style="font-size:15.0px"&gt;OpenAI cephesi: ChatGPT, Codex, GPT-Live&lt;b&gt;16&lt;/b&gt;&lt;/a&gt; &lt;a href="#t6" style="font-size:14.4px"&gt;Küçük ve açık modeller hücumda&lt;b&gt;14&lt;/b&gt;&lt;/a&gt; &lt;a href="#t7" style="font-size:13.6px"&gt;Yerel donanım ve yerel çalıştırma&lt;b&gt;11&lt;/b&gt;&lt;/a&gt; &lt;a href="#t8" style="font-size:12.1px"&gt;Ses ve gerçek zamanlı çeviri&lt;b&gt;6&lt;/b&gt;&lt;/a&gt; &lt;a href="#t9" style="font-size:17.0px"&gt;Ajanlar, kodlama ve yazılım fabrikaları&lt;b&gt;23&lt;/b&gt;&lt;/a&gt; &lt;a href="#t10" style="font-size:11.6px"&gt;Görsel ve işitsel üretim&lt;b&gt;4&lt;/b&gt;&lt;/a&gt; &lt;a href="#t11" style="font-size:11.0px"&gt;Frontier modeller ve AI tartışmaları&lt;b&gt;2&lt;/b&gt;&lt;/a&gt; &lt;a href="#t12" style="font-size:11.6px"&gt;Öğrenme ve temeller&lt;b&gt;4&lt;/b&gt;&lt;/a&gt;&lt;/nav&gt;
&lt;div class="top"&gt;&lt;a class="champ" href="https://www.youtube.com/watch?v=NA9D49kQFuU" target="_blank" rel="noopener"&gt;&lt;div class="ck"&gt;Haftanın rekoru&lt;/div&gt;&lt;div class="cv"&gt;552K&lt;/div&gt;&lt;div class="ct"&gt;Every Ways to Get 128GB VRAM for Local AI at Full Context&lt;/div&gt;&lt;div class="cs"&gt;Kai &amp;middot; 10 Eki&lt;/div&gt;&lt;div class="cd"&gt;Kai&amp;#x27;nin haftanın şampiyon videosu, yerel AI modellerini tam bağlam uzunluğunda çalıştırmak için 128GB VRAM&amp;#x27;e ulaşmanın tüm yollarını anlatıyor; 550 bini aşan izlenmeyle haftanın açık ara en çok izlenen içeriği oldu.&lt;/div&gt;&lt;/a&gt;
&lt;div class="stats"&gt;&lt;h3&gt;Haftanın rakamları&lt;/h3&gt;&lt;div class="st"&gt;&lt;span&gt;Aktif kanal&lt;/span&gt;&lt;b&gt;25/43&lt;/b&gt;&lt;/div&gt;
&lt;div class="st"&gt;&lt;span&gt;Toplam video&lt;/span&gt;&lt;b&gt;120&lt;/b&gt;&lt;/div&gt;
&lt;div class="st"&gt;&lt;span&gt;Toplam izlenme&lt;/span&gt;&lt;b&gt;3.9M&lt;/b&gt;&lt;/div&gt;
&lt;div class="st"&gt;&lt;span&gt;Tema sayısı&lt;/span&gt;&lt;b&gt;12&lt;/b&gt;&lt;/div&gt;&lt;/div&gt;&lt;/div&gt;
&lt;div class="rowg g23"&gt;
&lt;div class="card"&gt;&lt;h3&gt;Haftalık ritim — günlük video sayısı (çubuk · sol eksen) + izlenme (eğri · sağ eksen)&lt;/h3&gt;&lt;svg viewBox="0 0 960 250" role="img"&gt;&lt;line x1="44" y1="26" x2="904" y2="26" stroke="#222935"/&gt;&lt;text x="36" y="30" text-anchor="end" class="ax"&gt;29&lt;/text&gt;&lt;line x1="44" y1="74" x2="904" y2="74" stroke="#222935"/&gt;&lt;text x="36" y="78" text-anchor="end" class="ax"&gt;22&lt;/text&gt;&lt;line x1="44" y1="121" x2="904" y2="121" stroke="#222935"/&gt;&lt;text x="36" y="125" text-anchor="end" class="ax"&gt;14&lt;/text&gt;&lt;line x1="44" y1="168" x2="904" y2="168" stroke="#222935"/&gt;&lt;text x="36" y="172" text-anchor="end" class="ax"&gt;7&lt;/text&gt;&lt;line x1="44" y1="216" x2="904" y2="216" stroke="#222935"/&gt;&lt;text x="36" y="220" text-anchor="end" class="ax"&gt;0&lt;/text&gt;&lt;rect x="63.3" y="196.3" width="68.8" height="19.7" rx="3" fill="#2b5f58"&gt;&lt;title&gt;3 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="97.8" y="191.3" text-anchor="middle" class="bn2"&gt;3&lt;/text&gt;&lt;text x="97.8" y="238" text-anchor="middle" class="ax"&gt;09 Eki&lt;/text&gt;&lt;rect x="170.8" y="170.1" width="68.8" height="45.9" rx="3" fill="#2b5f58"&gt;&lt;title&gt;7 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="205.2" y="165.1" text-anchor="middle" class="bn2"&gt;7&lt;/text&gt;&lt;text x="205.2" y="238" text-anchor="middle" class="ax"&gt;10 Eki&lt;/text&gt;&lt;rect x="278.4" y="196.3" width="68.8" height="19.7" rx="3" fill="#2b5f58"&gt;&lt;title&gt;3 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="312.8" y="191.3" text-anchor="middle" class="bn2"&gt;3&lt;/text&gt;&lt;text x="312.8" y="238" text-anchor="middle" class="ax"&gt;11 Eki&lt;/text&gt;&lt;rect x="385.9" y="150.5" width="68.8" height="65.5" rx="3" fill="#2b5f58"&gt;&lt;title&gt;10 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="420.2" y="145.5" text-anchor="middle" class="bn2"&gt;10&lt;/text&gt;&lt;text x="420.2" y="238" text-anchor="middle" class="ax"&gt;12 Eki&lt;/text&gt;&lt;rect x="493.4" y="26.0" width="68.8" height="190.0" rx="3" fill="#2b5f58"&gt;&lt;title&gt;29 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="527.8" y="21.0" text-anchor="middle" class="bn2"&gt;29&lt;/text&gt;&lt;text x="527.8" y="238" text-anchor="middle" class="ax"&gt;13 Eki&lt;/text&gt;&lt;rect x="600.9" y="65.3" width="68.8" height="150.7" rx="3" fill="#2b5f58"&gt;&lt;title&gt;23 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="635.2" y="60.3" text-anchor="middle" class="bn2"&gt;23&lt;/text&gt;&lt;text x="635.2" y="238" text-anchor="middle" class="ax"&gt;14 Eki&lt;/text&gt;&lt;rect x="708.4" y="52.2" width="68.8" height="163.8" rx="3" fill="#2b5f58"&gt;&lt;title&gt;25 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="742.8" y="47.2" text-anchor="middle" class="bn2"&gt;25&lt;/text&gt;&lt;text x="742.8" y="238" text-anchor="middle" class="ax"&gt;15 Eki&lt;/text&gt;&lt;rect x="815.9" y="85.0" width="68.8" height="131.0" rx="3" fill="#2b5f58"&gt;&lt;title&gt;20 video&lt;/title&gt;&lt;/rect&gt;&lt;text x="850.2" y="80.0" text-anchor="middle" class="bn2"&gt;20&lt;/text&gt;&lt;text x="850.2" y="238" text-anchor="middle" class="ax"&gt;16 Eki&lt;/text&gt;&lt;polyline fill="none" stroke="#ffb454" stroke-width="2" points="97.8,189.5 205.2,39.9 312.8,146.5 420.2,135.6 527.8,26.0 635.2,144.3 742.8,149.7 850.2,211.8"/&gt;&lt;circle cx="97.8" cy="189.5" r="3.5" fill="#ffb454"&gt;&lt;title&gt;150K izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="205.2" cy="39.9" r="3.5" fill="#ffb454"&gt;&lt;title&gt;998K izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="312.8" cy="146.5" r="3.5" fill="#ffb454"&gt;&lt;title&gt;394K izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="420.2" cy="135.6" r="3.5" fill="#ffb454"&gt;&lt;title&gt;456K izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="527.8" cy="26.0" r="3.5" fill="#ffb454"&gt;&lt;title&gt;1.1M izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="635.2" cy="144.3" r="3.5" fill="#ffb454"&gt;&lt;title&gt;407K izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="742.8" cy="149.7" r="3.5" fill="#ffb454"&gt;&lt;title&gt;376K izlenme&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="850.2" cy="211.8" r="3.5" fill="#ffb454"&gt;&lt;title&gt;24K izlenme&lt;/title&gt;&lt;/circle&gt;&lt;text x="912" y="220" class="ax" fill="#a8814a"&gt;0&lt;/text&gt;&lt;text x="912" y="125" class="ax" fill="#a8814a"&gt;539K&lt;/text&gt;&lt;text x="912" y="30" class="ax" fill="#a8814a"&gt;1.1M&lt;/text&gt;&lt;/svg&gt;&lt;/div&gt;
&lt;div class="card"&gt;&lt;h3&gt;Tema dağılımı&lt;/h3&gt;&lt;svg viewBox="0 0 220 220" class="donut"&gt;&lt;path class="dseg" data-tip="Gemini 4 Argon fırtınası — 17 video (%14)" d="M110.0,30.0 A80,80 0 0 1 171.9,59.4" stroke="#35d0ba" stroke-width="30" fill="none"&gt;&lt;title&gt;Gemini 4 Argon fırtınası: 17 (%14)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Meta&amp;#x27;dan Muse sürprizi — 9 video (%8)" d="M172.2,59.7 A80,80 0 0 1 188.2,93.2" stroke="#ffb454" stroke-width="30" fill="none"&gt;&lt;title&gt;Meta&amp;#x27;dan Muse sürprizi: 9 (%8)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Güvenlik, ihlal ve güven tartışması — 6 video (%5)" d="M188.3,93.4 A80,80 0 0 1 189.6,118.2" stroke="#f47174" stroke-width="30" fill="none"&gt;&lt;title&gt;Güvenlik, ihlal ve güven tartışması: 6 (%5)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="OCR ve belge anlama — 8 video (%7)" d="M189.6,118.4 A80,80 0 0 1 179.4,149.9" stroke="#7aa2f7" stroke-width="30" fill="none"&gt;&lt;title&gt;OCR ve belge anlama: 8 (%7)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="OpenAI cephesi: ChatGPT, Codex, GPT-Live — 16 video (%13)" d="M179.3,150.0 A80,80 0 0 1 127.0,188.2" stroke="#9ece6a" stroke-width="30" fill="none"&gt;&lt;title&gt;OpenAI cephesi: ChatGPT, Codex, GPT-Live: 16 (%13)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Küçük ve açık modeller hücumda — 14 video (%12)" d="M126.6,188.3 A80,80 0 0 1 70.3,179.4" stroke="#bb9af7" stroke-width="30" fill="none"&gt;&lt;title&gt;Küçük ve açık modeller hücumda: 14 (%12)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Yerel donanım ve yerel çalıştırma — 11 video (%9)" d="M70.0,179.3 A80,80 0 0 1 38.8,146.5" stroke="#7dcfff" stroke-width="30" fill="none"&gt;&lt;title&gt;Yerel donanım ve yerel çalıştırma: 11 (%9)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Ses ve gerçek zamanlı çeviri — 6 video (%5)" d="M38.7,146.3 A80,80 0 0 1 31.0,122.6" stroke="#e0af68" stroke-width="30" fill="none"&gt;&lt;title&gt;Ses ve gerçek zamanlı çeviri: 6 (%5)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Ajanlar, kodlama ve yazılım fabrikaları — 23 video (%19)" d="M31.0,122.5 A80,80 0 0 1 69.6,41.0" stroke="#f7768e" stroke-width="30" fill="none"&gt;&lt;title&gt;Ajanlar, kodlama ve yazılım fabrikaları: 23 (%19)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Görsel ve işitsel üretim — 4 video (%3)" d="M70.0,40.7 A80,80 0 0 1 85.2,33.9" stroke="#4ec9b0" stroke-width="30" fill="none"&gt;&lt;title&gt;Görsel ve işitsel üretim: 4 (%3)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Frontier modeller ve AI tartışmaları — 2 video (%2)" d="M85.3,33.9 A80,80 0 0 1 93.3,31.8" stroke="#c0caf5" stroke-width="30" fill="none"&gt;&lt;title&gt;Frontier modeller ve AI tartışmaları: 2 (%2)&lt;/title&gt;&lt;/path&gt;&lt;path class="dseg" data-tip="Öğrenme ve temeller — 4 video (%3)" d="M93.4,31.7 A80,80 0 0 1 109.9,30.0" stroke="#d19a66" stroke-width="30" fill="none"&gt;&lt;title&gt;Öğrenme ve temeller: 4 (%3)&lt;/title&gt;&lt;/path&gt;&lt;circle cx="110" cy="110" r="64" fill="#0b0e12"/&gt;&lt;text x="110" y="106" text-anchor="middle" class="dn"&gt;120&lt;/text&gt;&lt;text x="110" y="126" text-anchor="middle" class="ax"&gt;video&lt;/text&gt;&lt;/svg&gt;&lt;div class="lgs"&gt;&lt;span class="lg"&gt;&lt;i style="background:#35d0ba"&gt;&lt;/i&gt;Gemini 4 Argon fırtınası &lt;b&gt;17&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#ffb454"&gt;&lt;/i&gt;Meta&amp;#x27;dan Muse sürprizi &lt;b&gt;9&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#f47174"&gt;&lt;/i&gt;Güvenlik, ihlal ve güven tartışması &lt;b&gt;6&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#7aa2f7"&gt;&lt;/i&gt;OCR ve belge anlama &lt;b&gt;8&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#9ece6a"&gt;&lt;/i&gt;OpenAI cephesi: ChatGPT, Codex, GPT-Live &lt;b&gt;16&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#bb9af7"&gt;&lt;/i&gt;Küçük ve açık modeller hücumda &lt;b&gt;14&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#7dcfff"&gt;&lt;/i&gt;Yerel donanım ve yerel çalıştırma &lt;b&gt;11&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#e0af68"&gt;&lt;/i&gt;Ses ve gerçek zamanlı çeviri &lt;b&gt;6&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#f7768e"&gt;&lt;/i&gt;Ajanlar, kodlama ve yazılım fabrikaları &lt;b&gt;23&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#4ec9b0"&gt;&lt;/i&gt;Görsel ve işitsel üretim &lt;b&gt;4&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#c0caf5"&gt;&lt;/i&gt;Frontier modeller ve AI tartışmaları &lt;b&gt;2&lt;/b&gt;&lt;/span&gt;&lt;span class="lg"&gt;&lt;i style="background:#d19a66"&gt;&lt;/i&gt;Öğrenme ve temeller &lt;b&gt;4&lt;/b&gt;&lt;/span&gt;&lt;/div&gt;&lt;/div&gt;&lt;/div&gt;
&lt;div class="card" style="margin-bottom:14px"&gt;&lt;h3&gt;Trend zaman çizelgesi — her nokta bir video · x: yayın anı · y: log izlenme · renk: tema · üzerine gel: başlık&lt;/h3&gt;&lt;svg viewBox="0 0 960 300" role="img"&gt;&lt;line x1="46" y1="264.0" x2="946" y2="264.0" stroke="#222935"/&gt;&lt;text x="38" y="268" text-anchor="end" class="ax"&gt;100&lt;/text&gt;&lt;line x1="46" y1="197.7" x2="946" y2="197.7" stroke="#222935"/&gt;&lt;text x="38" y="202" text-anchor="end" class="ax"&gt;1K&lt;/text&gt;&lt;line x1="46" y1="131.5" x2="946" y2="131.5" stroke="#222935"/&gt;&lt;text x="38" y="135" text-anchor="end" class="ax"&gt;10K&lt;/text&gt;&lt;line x1="46" y1="65.2" x2="946" y2="65.2" stroke="#222935"/&gt;&lt;text x="38" y="69" text-anchor="end" class="ax"&gt;100K&lt;/text&gt;&lt;line x1="102.2" y1="16" x2="102.2" y2="264" stroke="#1a2029"/&gt;&lt;text x="102.2" y="288" text-anchor="middle" class="ax"&gt;09 Eki&lt;/text&gt;&lt;line x1="214.8" y1="16" x2="214.8" y2="264" stroke="#1a2029"/&gt;&lt;text x="214.8" y="288" text-anchor="middle" class="ax"&gt;10 Eki&lt;/text&gt;&lt;line x1="327.2" y1="16" x2="327.2" y2="264" stroke="#1a2029"/&gt;&lt;text x="327.2" y="288" text-anchor="middle" class="ax"&gt;11 Eki&lt;/text&gt;&lt;line x1="439.8" y1="16" x2="439.8" y2="264" stroke="#1a2029"/&gt;&lt;text x="439.8" y="288" text-anchor="middle" class="ax"&gt;12 Eki&lt;/text&gt;&lt;line x1="552.2" y1="16" x2="552.2" y2="264" stroke="#1a2029"/&gt;&lt;text x="552.2" y="288" text-anchor="middle" class="ax"&gt;13 Eki&lt;/text&gt;&lt;line x1="664.8" y1="16" x2="664.8" y2="264" stroke="#1a2029"/&gt;&lt;text x="664.8" y="288" text-anchor="middle" class="ax"&gt;14 Eki&lt;/text&gt;&lt;line x1="777.2" y1="16" x2="777.2" y2="264" stroke="#1a2029"/&gt;&lt;text x="777.2" y="288" text-anchor="middle" class="ax"&gt;15 Eki&lt;/text&gt;&lt;line x1="889.8" y1="16" x2="889.8" y2="264" stroke="#1a2029"/&gt;&lt;text x="889.8" y="288" text-anchor="middle" class="ax"&gt;16 Eki&lt;/text&gt;&lt;circle cx="950.7" cy="166.2" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;We’ve Been Giving AI Agents Access Wrong. - Prompt Engineering (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="711.6" cy="112.1" r="4.5" fill="#35d0ba" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Argon Built All of This - Prompt Engineering (19K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="627.2" cy="58.5" r="4.5" fill="#35d0ba" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Google is BACK! (Gemini 4 Argon - Fully Tested!) - Prompt Engineering (126K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="524.1" cy="121.9" r="4.5" fill="#ffb454" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Meta’s Muse Wonder - Fable Class Model - Prompt Engineering (14K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="388.2" cy="104.1" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Google&amp;#x27;s GPT-3 Moment for Databases &amp;amp; Spreadsheets (TabFM) - Prompt Engineering (26K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="303.8" cy="150.8" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;What Is a Software Factory? - Prompt Engineering (5K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="964.8" cy="236.5" r="4.5" fill="#7aa2f7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;This Tiny 0.8B AI Reads Documents Better Than OCR? - Prompt Engineer 48 (160)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="960.1" cy="240.0" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;FLUX 3 Image Klein Hands-On is CRAZY - Prompt Engineer 48 (130)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="941.3" cy="235.3" r="4.5" fill="#7dcfff" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Fish Audio Drama 3: Can You Run It Yourself? - Prompt Engineer 48 (171)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="922.6" cy="228.8" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Claude Can Now Make AI Videos By Itself (Higgsfield MCP) - Prompt Engineer 48 (240)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="889.8" cy="209.8" r="4.5" fill="#bb9af7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;This 150M Model Claims #1. I Tested It. - Prompt Engineer 48 (557)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="880.4" cy="220.8" r="4.5" fill="#bb9af7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Jeva - the Jev Competitor: This 4B Model Never Writes a Word. I Tested It - Prompt Engineer 48 (348)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="852.2" cy="200.1" r="4.5" fill="#7dcfff" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;LlamaCPP on Native Windows | This is HUGE - Prompt Engineer 48 (820)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="828.8" cy="200.1" r="4.5" fill="#7dcfff" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;NVIDIA&amp;#x27;s Sana: 1024px Image in 0.4 Seconds? I Tested It - Prompt Engineer 48 (822)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="810.1" cy="211.3" r="4.5" fill="#bb9af7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;This Tiny 2B AI vs Claude Code&amp;#x27;s Token Bill (Real Test) - Prompt Engineer 48 (524)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="791.3" cy="212.9" r="4.5" fill="#bb9af7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Amos 1 Small: 69.5% Accurate, Zero Tokens, 5 GB VRAM - Prompt Engineer 48 (490)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="758.5" cy="197.1" r="4.5" fill="#35d0ba" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Argon vs Opus 5.5: I Gave Both the Same Repo - Prompt Engineer 48 (921)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="739.8" cy="206.7" r="4.5" fill="#e0af68" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;I Built a Voice Agent You Can Interrupt (Murf Falcon 2, Real Latency Numbers) - Prompt Engineer 48 (633)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="692.9" cy="219.0" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Can You Spot the $0.15 Legal AI? (GLM-Legal Test) - Prompt Engineer 48 (377)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="631.9" cy="212.5" r="4.5" fill="#7aa2f7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Baidu&amp;#x27;s Unlimited OCR on an 8GB Laptop: 25 Tests - Prompt Engineer 48 (498)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="617.9" cy="183.0" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;A PS5 Game Running Natively on a PC? I Checked the Repo - Prompt Engineer 48 (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="627.2" cy="84.1" r="4.5" fill="#35d0ba" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Argon - Out and Tested!! - Sam Witteveen (52K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="960.1" cy="148.0" r="4.5" fill="#bb9af7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Qwen 3.8 27B Just Hit 200 Token/s on Mac M5 (Lithos Metal) - Kai (6K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="847.6" cy="86.9" r="4.5" fill="#7dcfff" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Local AI Hardware Just Got Way More Expensive Than You Think - Kai (47K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="735.1" cy="74.5" r="4.5" fill="#7dcfff" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Why Your Mac Is Slow at Local AI (And It&amp;#x27;s Not the Chip) - Kai (72K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="622.6" cy="90.2" r="4.5" fill="#bb9af7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Can 9 year old GPU beats today&amp;#x27;s latest GPUs? - Kai (42K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="510.1" cy="53.0" r="4.5" fill="#bb9af7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Don&amp;#x27;t Buy a GPU Instead Build This 32GB Free AI Server Instead - Kai (153K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="397.6" cy="84.2" r="4.5" fill="#7dcfff" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Don&amp;#x27;t Buy a RTX Spark Laptop For Local AI Until You See This - Kai (51K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="285.1" cy="16.0" r="4.5" fill="#7dcfff" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Every Ways to Get 128GB VRAM for Local AI at Full Context - Kai (552K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="627.2" cy="120.6" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;I Built an App with AI. These Tools Helped Me Understand It - The Gray Cat (14K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="931.9" cy="179.7" r="4.5" fill="#bb9af7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Jev Can Now SEE (And It&amp;#x27;s Absurdly Fast) - The AI Automators (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="842.9" cy="108.0" r="4.5" fill="#bb9af7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Jev Can Now SEE (And It&amp;#x27;s Absurdly Fast) - The AI Automators (22K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="739.8" cy="68.9" r="4.5" fill="#35d0ba" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;I Tested Gemini 4 Argon vs Opus 5.5 (No Hype) - The AI Automators (88K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="716.3" cy="147.3" r="4.5" fill="#ffb454" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;I tested Muse Wonder Full against GPT-6 and Claude. - The AI Automators (6K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="622.6" cy="150.0" r="4.5" fill="#ffb454" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Meta Just Dropped Muse Wonder. (I Tested it vs GPT-6 and Claude). - The AI Automators (5K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="486.6" cy="167.3" r="4.5" fill="#7dcfff" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Google just made AI search 100% offline - The AI Automators (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="885.1" cy="194.1" r="4.5" fill="#d19a66" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;How to Use Substack for Beginners - Howfinity (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="955.4" cy="198.1" r="4.5" fill="#35d0ba" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Argon built this cyberpunk city - Matthew Berman (888)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="866.3" cy="149.7" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;From Punch Cards to AI Software Factories - Matthew Berman (5K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="669.4" cy="64.8" r="4.5" fill="#35d0ba" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Google just went FRONTIER (Gemini 4 Argon) - Matthew Berman (101K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="650.7" cy="69.2" r="4.5" fill="#f47174" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;OpenAI Employee Arrested Over Recording Device - Matthew Berman (87K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="552.2" cy="59.2" r="4.5" fill="#ffb454" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Meta just dropped something insane... (Muse Wonder) - Matthew Berman (123K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="514.8" cy="121.3" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;I&amp;#x27;m Rebuilding Super Mario World With AI - Matthew Berman (14K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="449.1" cy="136.9" r="4.5" fill="#c0caf5" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Human Creativity Still Matters - Matthew Berman (8K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="299.1" cy="38.0" r="4.5" fill="#c0caf5" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Everything has been solved - Matthew Berman (257K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="214.8" cy="59.1" r="4.5" fill="#4ec9b0" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;AI Is Recreating Photoshop and Other Adobe Apps - Matthew Berman (123K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="974.1" cy="199.6" r="4.5" fill="#9ece6a" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Put ChatGPT to work on shared team tasks - OpenAI (838)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="871.0" cy="74.1" r="4.5" fill="#9ece6a" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Build more personal sites with Codex - OpenAI (73K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="861.6" cy="57.4" r="4.5" fill="#9ece6a" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;GPT-Live-1.1 has new voices and better accents. - OpenAI (131K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="842.9" cy="152.5" r="4.5" fill="#f47174" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Weeks to Hours: SentinelOne on Cybersecurity Research with OpenAI - OpenAI (5K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="842.9" cy="113.2" r="4.5" fill="#f47174" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Weeks to Hours: SentinelOne on Cybersecurity Research with OpenAI - OpenAI (19K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="744.4" cy="142.2" r="4.5" fill="#9ece6a" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Cooking up a catering business with ChatGPT - OpenAI (7K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="706.9" cy="145.6" r="4.5" fill="#9ece6a" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;How Oracle Uses Codex to Help Business Users Get Answers - OpenAI (6K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="650.7" cy="78.1" r="4.5" fill="#9ece6a" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Work with your team in ChatGPT Space - OpenAI (64K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="636.6" cy="142.9" r="4.5" fill="#f47174" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Block on Finding and Fixing Security Risks with OpenAI - OpenAI (7K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="631.9" cy="106.5" r="4.5" fill="#f47174" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Block on Finding and Fixing Security Risks with OpenAI - OpenAI (24K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="622.6" cy="134.4" r="4.5" fill="#9ece6a" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Ride out this sticker storm in style - OpenAI (9K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="514.8" cy="69.2" r="4.5" fill="#9ece6a" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Code review is no longer the bottleneck - OpenAI (87K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="313.2" cy="99.4" r="4.5" fill="#9ece6a" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;How Oracle Uses Codex to Help Business Users Get Answers - OpenAI (30K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="191.3" cy="57.5" r="4.5" fill="#9ece6a" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Set up your dot in the ChatGPT mobile app - OpenAI (130K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="950.7" cy="223.7" r="4.5" fill="#bb9af7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;NVIDIA JUST DROPPED Nemotron 3.5 Super VL… Can Qwen 27B Beat It? - AI WITH Rithesh (305)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="922.6" cy="239.8" r="4.5" fill="#bb9af7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Liquid AI d1-Ultra vs Jev… WHO WINS? | Battle of AI Decision Models! - AI WITH Rithesh (132)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="922.6" cy="216.3" r="4.5" fill="#ffb454" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Meta&amp;#x27;s 816B open model Muse Spark Open Tested - AI WITH Rithesh (425)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="913.2" cy="234.4" r="4.5" fill="#ffb454" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Meta Muse Spark Open Is WILD… 816B Parameters!  198 Real Tests! Open Weights! - AI WITH Rithesh (180)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="814.8" cy="216.8" r="4.5" fill="#7aa2f7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Indra OCR Is WILD… 1M Parameters, 25 Pages/Second on My MacBook! - AI WITH Rithesh (415)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="786.6" cy="194.9" r="4.5" fill="#9ece6a" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Ultramarine 1 Flash Is WILD… Claude-Level AI at 60× Lower Cost! - AI WITH Rithesh (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="716.3" cy="131.4" r="4.5" fill="#35d0ba" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Argon vs Claude Opus 5.5… CHEAPER, FASTER, BETTER? - AI WITH Rithesh (10K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="711.6" cy="200.5" r="4.5" fill="#35d0ba" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Argon created this (one prompt each) - AI WITH Rithesh (808)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="702.2" cy="177.2" r="4.5" fill="#35d0ba" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Argon did this - AI WITH Rithesh (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="702.2" cy="143.6" r="4.5" fill="#bb9af7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Liquid AI LFM2.5-24B-A2B Is WILD… I Tested It Locally! - AI WITH Rithesh (6K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="688.2" cy="193.3" r="4.5" fill="#35d0ba" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Argon Is INSANE… Wild Demos From Real Users! - AI WITH Rithesh (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="678.8" cy="198.6" r="4.5" fill="#35d0ba" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Argon Is INSANE… Incredible Demos &amp;amp; Hidden Issues! - AI WITH Rithesh (869)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="599.1" cy="229.2" r="4.5" fill="#ffb454" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Meta Muse Wonder BENCHMAXXED? Incredible Demos and Hidden Issues - AI WITH Rithesh (235)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="589.8" cy="192.9" r="4.5" fill="#ffb454" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Meta just dropped a frontier model &#129327; Muse Wonder - AI WITH Rithesh (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="575.7" cy="187.6" r="4.5" fill="#7aa2f7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;MiniCPM-V 4.7 Is WILD… 35B Vision AI on 24GB RAM! Tested Locally! - AI WITH Rithesh (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="842.9" cy="132.0" r="4.5" fill="#4ec9b0" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;ComfyUI Nodes Explained: The Basics for Beginners - pixaroma (10K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="622.6" cy="103.5" r="4.5" fill="#4ec9b0" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;50 Free ComfyUI Automation Workflows (Ep38) - pixaroma (26K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="964.8" cy="218.4" r="4.5" fill="#d19a66" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Give Me 42 Min, I Will Make PyTorch Indexing Click Forever - Zachary Huang (387)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="622.6" cy="79.2" r="4.5" fill="#d19a66" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Give me 31 minutes, I will make PyTorch click forever - Zachary Huang (61K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="308.5" cy="120.7" r="4.5" fill="#d19a66" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Give me 24 minutes, I will make Neural Networks click forever - Zachary Huang (14K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="955.4" cy="145.5" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;My AI Dark Factory Dream Is Dead (Here&amp;#x27;s What Actually Works) - Cole Medin (6K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="777.2" cy="109.8" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;How I Keep Up with AI Without Drowning in It (Free Skill) - Cole Medin (21K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="510.1" cy="109.8" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Is Learning to Build Software Still Worth It with AI? (My Honest Take) - Cole Medin (21K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="889.8" cy="179.6" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Never Used the GitHub Copilot App? Build a Game - Burke Holland (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="753.8" cy="137.0" r="4.5" fill="#bb9af7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;RTX 5090 vs M5 Ultra: Which AI Powerhouse Wins? - Hugging Face (8K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="753.8" cy="165.1" r="4.5" fill="#9ece6a" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;New Model: Reflection Beam is a new Open Weight Model - Hugging Face (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="721.0" cy="264.0" r="4.5" fill="#7dcfff" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Local AI 202: Get Massive Speed-ups with Speculative Decoding - Hugging Face (0)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="613.2" cy="200.8" r="4.5" fill="#7dcfff" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;I Turned Structured Output Into a Game &#127918; - Hugging Face (800)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="566.3" cy="120.3" r="4.5" fill="#bb9af7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;How to Build Your Own AI Agent (Like PewDiePie) - Hugging Face (15K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="458.5" cy="122.6" r="4.5" fill="#9ece6a" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Build a Business Dashboard with AI in MINUTES (for dummies...) - Hugging Face (14K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="589.8" cy="33.8" r="4.5" fill="#7dcfff" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;OpenAI’s New AI Is Doing Things We Barely Understand - Two Minute Papers (297K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="402.2" cy="32.0" r="4.5" fill="#7aa2f7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;DeepSeek Solved AI’s Memory Problem - Two Minute Papers (317K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="960.1" cy="203.5" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Why Nobody Has Solved Continual Learning - LangChain (718)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="852.2" cy="181.5" r="4.5" fill="#f47174" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Vanta and LangSmith: Assessing Compliance Evidence Like an Auditor - LangChain (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="838.2" cy="158.0" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;How Stripe Built Kai - LangChain (4K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="838.2" cy="191.9" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Why You Think the Frontier Model Is Always Better (It&amp;#x27;s Psychology) - LangChain (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="735.1" cy="151.5" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;What is Deep Agents? - LangChain (5K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="641.3" cy="139.9" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;What is an Agent Harness? - LangChain (7K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="594.4" cy="203.9" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;POV: You’re arriving at Interrupt London, The Agent Conference by LangChain. - LangChain (708)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="491.3" cy="166.2" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Why &amp;quot;best model&amp;quot; is the wrong question now - LangChain (3K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="205.4" cy="128.0" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;What is a Skill? - LangChain (11K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="810.1" cy="192.9" r="4.5" fill="#4ec9b0" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;14c - Zero-Shot Design of Drug-Binding Proteins - DigitalSreeni (1K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="922.6" cy="256.2" r="4.5" fill="#e0af68" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Design a Voice by Talking | Dialt #Shorts - Trelis Research (31)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="810.1" cy="220.3" r="4.5" fill="#e0af68" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Two Ways to Switch Languages with Voice Assistants - Trelis Research (356)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="697.6" cy="205.9" r="4.5" fill="#9ece6a" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Multiplayer Coding Agents: Our Experience with Amp - Trelis Research (652)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="599.1" cy="216.8" r="4.5" fill="#e0af68" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Design a Voice by Talking: Interactive Voice Design in Dialt - Trelis Research (415)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="749.1" cy="152.8" r="4.5" fill="#9ece6a" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Give your team AI skills you can trust - DeepLearningAI (5K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="631.9" cy="216.3" r="4.5" fill="#9ece6a" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Billion Inferences a Day ML Platform | Coinbase + Anyscale - Anyscale (424)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="627.2" cy="219.2" r="4.5" fill="#7aa2f7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;OpenDocRouter: Use the API and SDK - LlamaIndex (374)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="622.6" cy="227.3" r="4.5" fill="#7aa2f7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;OpenDocRouter: Picking your Parsing Model - LlamaIndex (258)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="622.6" cy="225.2" r="4.5" fill="#7aa2f7" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;OpenDocRouter: Parsing your First Document - LlamaIndex (285)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="294.4" cy="118.2" r="4.5" fill="#f7768e" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;What Companies Hire Freelance Engineers For in 2026 - Dave Ebbelaar (16K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="542.9" cy="55.5" r="4.5" fill="#ffb454" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;AI Labs Warn Race Makes Breach ‘Inevitable’ - AI Explained (140K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="964.8" cy="204.6" r="4.5" fill="#35d0ba" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;&#127925; Turning motion into music with Gemini - Google for Developers (688)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="885.1" cy="173.6" r="4.5" fill="#e0af68" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;&#128214; Explore custom voice narration with Gemini 3.8 Flash TTS - Google for Developers (2K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="866.3" cy="131.1" r="4.5" fill="#35d0ba" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Turning raw hardware sensor data into music with Gemini - Google for Developers (10K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="852.2" cy="114.9" r="4.5" fill="#35d0ba" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Argon in Antigravity - Google for Developers (18K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="744.4" cy="78.4" r="4.5" fill="#35d0ba" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Argon: What’s new and what’s next - Google for Developers (63K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="641.3" cy="62.4" r="4.5" fill="#35d0ba" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Gemini 4 Argon and Google Antigravity: From 2D concept to 3D cathedral - Google for Developers (110K)&lt;/title&gt;&lt;/circle&gt;&lt;circle cx="210.1" cy="135.1" r="4.5" fill="#e0af68" fill-opacity=".85" stroke="#0b0e12" stroke-width="1"&gt;&lt;title&gt;Order in any language with Gemini 3.5 Live Translate - Google for Developers (9K)&lt;/title&gt;&lt;/circle&gt;&lt;/svg&gt;&lt;/div&gt;
&lt;div class="rowg g11"&gt;
&lt;div class="card"&gt;&lt;h3&gt;Tema yoğunluğu&lt;/h3&gt;&lt;div class="bars"&gt;&lt;a class="bar" href="#t1"&gt;&lt;span class="bn"&gt;Gemini 4 Argon fırtınası&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:74%;background:#35d0ba"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;17&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t2"&gt;&lt;span class="bn"&gt;Meta&amp;#x27;dan Muse sürprizi&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:39%;background:#ffb454"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;9&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t3"&gt;&lt;span class="bn"&gt;Güvenlik, ihlal ve güven tartışması&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:26%;background:#f47174"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;6&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t4"&gt;&lt;span class="bn"&gt;OCR ve belge anlama&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:35%;background:#7aa2f7"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;8&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t5"&gt;&lt;span class="bn"&gt;OpenAI cephesi: ChatGPT, Codex, GPT-Live&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:70%;background:#9ece6a"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;16&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t6"&gt;&lt;span class="bn"&gt;Küçük ve açık modeller hücumda&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:61%;background:#bb9af7"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;14&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t7"&gt;&lt;span class="bn"&gt;Yerel donanım ve yerel çalıştırma&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:48%;background:#7dcfff"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;11&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t8"&gt;&lt;span class="bn"&gt;Ses ve gerçek zamanlı çeviri&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:26%;background:#e0af68"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;6&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t9"&gt;&lt;span class="bn"&gt;Ajanlar, kodlama ve yazılım fabrikaları&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:100%;background:#f7768e"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;23&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t10"&gt;&lt;span class="bn"&gt;Görsel ve işitsel üretim&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:17%;background:#4ec9b0"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;4&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t11"&gt;&lt;span class="bn"&gt;Frontier modeller ve AI tartışmaları&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:9%;background:#c0caf5"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;2&lt;/span&gt;&lt;/a&gt;&lt;a class="bar" href="#t12"&gt;&lt;span class="bn"&gt;Öğrenme ve temeller&lt;/span&gt;&lt;span class="bt"&gt;&lt;span class="bf" style="width:17%;background:#d19a66"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="bv"&gt;4&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/div&gt;
&lt;div class="card"&gt;&lt;h3&gt;En çok izlenen kanallar&lt;/h3&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Kai&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:100%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;922K&lt;i&gt;7 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Matthew Berman&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:78%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;720K&lt;i&gt;9 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Two Minute Papers&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:67%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;614K&lt;i&gt;2 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;OpenAI&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:64%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;592K&lt;i&gt;14 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Google for Developers&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:23%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;213K&lt;i&gt;7 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;div class="cbar"&gt;&lt;span class="cn"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="ct2"&gt;&lt;span class="cf" style="width:21%"&gt;&lt;/span&gt;&lt;/span&gt;&lt;span class="cv2"&gt;193K&lt;i&gt;6 v&lt;/i&gt;&lt;/span&gt;&lt;/div&gt;&lt;/div&gt;&lt;/div&gt;
&lt;section class="theme" id="t1"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#35d0ba"&gt;01&lt;/span&gt;&lt;i class="dot" style="background:#35d0ba"&gt;&lt;/i&gt;Gemini 4 Argon fırtınası&lt;span class="cnt"&gt;17 video &amp;middot; 604K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Google, Gemini 4 Argon ile uzun bir aradan sonra yeniden frontier yarışının merkezine döndü. Kanal genelinde test, Opus 5.5 kıyaslaması ve Antigravity demosu üst üste geldi; haftanın en kalabalık teması bu.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=yksrd5uNxcw" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;126K&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;Google is BACK! (Gemini 4 Argon - Fully Tested!)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=OYp5j2RihPE" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;110K&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Argon and Google Antigravity: From 2D concept to 3D cathedral&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=M8iqahFogtA" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;101K&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Google just went FRONTIER (Gemini 4 Argon)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=uBQaEXjNUe4" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;88K&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;The AI Automators&lt;/span&gt;&lt;span class="tt"&gt;I Tested Gemini 4 Argon vs Opus 5.5 (No Hype)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Jh6mgf9M06M" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;63K&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Argon: What’s new and what’s next&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=4iK_UFzW6Zw" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;52K&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;Sam Witteveen&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Argon - Out and Tested!!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=YVV9csOi770" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;19K&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Argon Built All of This&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=nSwmcs37Jd4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;18K&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Argon in Antigravity&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=i3X7Fv0Z3Uc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;10K&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;Turning raw hardware sensor data into music with Gemini&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=t1hragr_wwc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;10K&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Argon vs Claude Opus 5.5… CHEAPER, FASTER, BETTER?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=28Rnvufvgos" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Argon did this&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=PvJ_Iw94VJM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Argon Is INSANE… Wild Demos From Real Users!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=jHaJfbZpB_Q" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;921&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Argon vs Opus 5.5: I Gave Both the Same Repo&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=vdU8mYN_-hE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;888&lt;/span&gt;&lt;span class="dt"&gt;16 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Argon built this cyberpunk city&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=tosvF51mEXU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;869&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Argon Is INSANE… Incredible Demos &amp;amp; Hidden Issues!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=jS-aP4rB-HM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;808&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Gemini 4 Argon created this (one prompt each)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=-nmcYcO-EE8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;688&lt;/span&gt;&lt;span class="dt"&gt;16 Eki&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;&#127925; Turning motion into music with Gemini&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t2"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#ffb454"&gt;02&lt;/span&gt;&lt;i class="dot" style="background:#ffb454"&gt;&lt;/i&gt;Meta&amp;#x27;dan Muse sürprizi&lt;span class="cnt"&gt;9 video &amp;middot; 290K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Meta, Muse Wonder ile &amp;#x27;fable sınıfı&amp;#x27; iddiasını ortaya koyarken 816B parametreli açık ağırlıklı Muse Spark ile açık model cephesini de zorladı. Test videoları GPT-6 ve Claude ile kafa kafaya sonuçlar raporluyor.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=C4Sq3Rot8FU" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;140K&lt;/span&gt;&lt;span class="dt"&gt;12 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI Explained&lt;/span&gt;&lt;span class="tt"&gt;AI Labs Warn Race Makes Breach ‘Inevitable’&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=rpLrkJ5NNEI" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;123K&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Meta just dropped something insane... (Muse Wonder)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=JqeqEqh_KeM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;14K&lt;/span&gt;&lt;span class="dt"&gt;12 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;Meta’s Muse Wonder - Fable Class Model&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=DvngEf053b8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;6K&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;The AI Automators&lt;/span&gt;&lt;span class="tt"&gt;I tested Muse Wonder Full against GPT-6 and Claude.&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=tYf3A9thVKQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;5K&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;The AI Automators&lt;/span&gt;&lt;span class="tt"&gt;Meta Just Dropped Muse Wonder. (I Tested it vs GPT-6 and Claude).&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=EIh8sSaEZvA" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Meta just dropped a frontier model &#129327; Muse Wonder&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=87JjbkkqORI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;425&lt;/span&gt;&lt;span class="dt"&gt;16 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Meta&amp;#x27;s 816B open model Muse Spark Open Tested&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Am-ElhZsOtc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;235&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Meta Muse Wonder BENCHMAXXED? Incredible Demos and Hidden Issues&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=eIlgpxlYaOs" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;180&lt;/span&gt;&lt;span class="dt"&gt;16 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Meta Muse Spark Open Is WILD… 816B Parameters!  198 Real Tests! Open Weights!&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t3"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#f47174"&gt;03&lt;/span&gt;&lt;i class="dot" style="background:#f47174"&gt;&lt;/i&gt;Güvenlik, ihlal ve güven tartışması&lt;span class="cnt"&gt;6 video &amp;middot; 142K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;AI lab&amp;#x27;ları araştırma yarışının güvenlik ihlalini &amp;#x27;kaçınılmaz&amp;#x27; kıldığını açıkça tartışmaya başladı; SentinelOne gibi şirketlerin haftalar süren araştırmalarını saatlere indirmesi hız-güvenlik gerilimini büyütüyor.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=CIlqLzoJiU8" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;87K&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;OpenAI Employee Arrested Over Recording Device&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=N6K_pxaAh0w" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;24K&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Block on Finding and Fixing Security Risks with OpenAI&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=BOTAeAU5ExQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;19K&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Weeks to Hours: SentinelOne on Cybersecurity Research with OpenAI&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=9aDuSpVoB-Y" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;7K&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Block on Finding and Fixing Security Risks with OpenAI&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=3IVo-MNwk2I" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;5K&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Weeks to Hours: SentinelOne on Cybersecurity Research with OpenAI&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=_MsL_RIztVc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;Vanta and LangSmith: Assessing Compliance Evidence Like an Auditor&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t4"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#7aa2f7"&gt;04&lt;/span&gt;&lt;i class="dot" style="background:#7aa2f7"&gt;&lt;/i&gt;OCR ve belge anlama&lt;span class="cnt"&gt;8 video &amp;middot; 320K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Baidu&amp;#x27;nun sınırsız OCR&amp;#x27;ı, 1M parametreyle saniyede 25 sayfa okuyan Indra ve OpenDocRouter serisiyle belge anlama, küçük modellerin en somut sahasına dönüştü.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=4fLVq6HJsKc" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;317K&lt;/span&gt;&lt;span class="dt"&gt;11 Eki&lt;/span&gt;&lt;span class="ch"&gt;Two Minute Papers&lt;/span&gt;&lt;span class="tt"&gt;DeepSeek Solved AI’s Memory Problem&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=B95HCsccDpE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;MiniCPM-V 4.7 Is WILD… 35B Vision AI on 24GB RAM! Tested Locally!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=M-Utfqo2Dhg" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;498&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Baidu&amp;#x27;s Unlimited OCR on an 8GB Laptop: 25 Tests&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=wXvJeofdAFQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;415&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Indra OCR Is WILD… 1M Parameters, 25 Pages/Second on My MacBook!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=AInKdz7vCQY" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;374&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;LlamaIndex&lt;/span&gt;&lt;span class="tt"&gt;OpenDocRouter: Use the API and SDK&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=CFxxiFvJdcY" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;285&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;LlamaIndex&lt;/span&gt;&lt;span class="tt"&gt;OpenDocRouter: Parsing your First Document&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ybrvt4260SM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;258&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;LlamaIndex&lt;/span&gt;&lt;span class="tt"&gt;OpenDocRouter: Picking your Parsing Model&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Uj0oxKlbENQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;160&lt;/span&gt;&lt;span class="dt"&gt;16 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;This Tiny 0.8B AI Reads Documents Better Than OCR?&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t5"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#9ece6a"&gt;05&lt;/span&gt;&lt;i class="dot" style="background:#9ece6a"&gt;&lt;/i&gt;OpenAI cephesi: ChatGPT, Codex, GPT-Live&lt;span class="cnt"&gt;16 video &amp;middot; 561K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;OpenAI ürün tarafında takım çalışmasına yükleniyor: ChatGPT Space, paylaşımlı görevler ve Codex ile Oracle gibi kurumsal müşterilerin iş kullanımları öne çıkan tanıtımlar.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=GyO8Ao7yKJc" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;131K&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;GPT-Live-1.1 has new voices and better accents.&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=p5KxBTUncoE" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;130K&lt;/span&gt;&lt;span class="dt"&gt;09 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Set up your dot in the ChatGPT mobile app&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=IZXv6a78NDw" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;87K&lt;/span&gt;&lt;span class="dt"&gt;12 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Code review is no longer the bottleneck&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=KJ_7KPKlO_Q" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;73K&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Build more personal sites with Codex&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=G-9Y2xNiDKQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;64K&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Work with your team in ChatGPT Space&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Ju-JyxDgV7g" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;30K&lt;/span&gt;&lt;span class="dt"&gt;10 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;How Oracle Uses Codex to Help Business Users Get Answers&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=44RXU8Cxoas" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;14K&lt;/span&gt;&lt;span class="dt"&gt;12 Eki&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;Build a Business Dashboard with AI in MINUTES (for dummies...)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=TD0bImyg_IM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;9K&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Ride out this sticker storm in style&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=PZxbaIp8PzE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;7K&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Cooking up a catering business with ChatGPT&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=DTcK95TAZ-k" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;6K&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;How Oracle Uses Codex to Help Business Users Get Answers&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=2IuO_80wB0s" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;5K&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;DeepLearningAI&lt;/span&gt;&lt;span class="tt"&gt;Give your team AI skills you can trust&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=dGhUFqeITbY" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;New Model: Reflection Beam is a new Open Weight Model&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=CcRr7dJy4gI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Ultramarine 1 Flash Is WILD… Claude-Level AI at 60× Lower Cost!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=OMxEUdIgsQc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;838&lt;/span&gt;&lt;span class="dt"&gt;16 Eki&lt;/span&gt;&lt;span class="ch"&gt;OpenAI&lt;/span&gt;&lt;span class="tt"&gt;Put ChatGPT to work on shared team tasks&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=76IUEiDMiVE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;652&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Multiplayer Coding Agents: Our Experience with Amp&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=wfLNdx3kvbU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;424&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;Anyscale&lt;/span&gt;&lt;span class="tt"&gt;Billion Inferences a Day ML Platform | Coinbase + Anyscale&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t6"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#bb9af7"&gt;06&lt;/span&gt;&lt;i class="dot" style="background:#bb9af7"&gt;&lt;/i&gt;Küçük ve açık modeller hücumda&lt;span class="cnt"&gt;14 video &amp;middot; 256K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;DeepSeek&amp;#x27;in bellek problemi çözümü 316 bin izlenmeye ulaşırken; 150M&amp;#x27;den 816B&amp;#x27;ye uzanan açık model dalgası, &amp;#x27;en iyi model&amp;#x27; sorusunun yerel ve ucuz modeller lehine anlam değiştirdiğini gösteriyor.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=IMVjEsKX5UQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;153K&lt;/span&gt;&lt;span class="dt"&gt;12 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Don&amp;#x27;t Buy a GPU Instead Build This 32GB Free AI Server Instead&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=eckOsa0kBkY" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;42K&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Can 9 year old GPU beats today&amp;#x27;s latest GPUs?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=V-fTrJqkG7k" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;22K&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;The AI Automators&lt;/span&gt;&lt;span class="tt"&gt;Jev Can Now SEE (And It&amp;#x27;s Absurdly Fast)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=EuN8WCWQf7g" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;15K&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;How to Build Your Own AI Agent (Like PewDiePie)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=qaP4pymNrEc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;8K&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;RTX 5090 vs M5 Ultra: Which AI Powerhouse Wins?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=kKPWsfXIZwQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;6K&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Liquid AI LFM2.5-24B-A2B Is WILD… I Tested It Locally!&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=L7noXCBSObc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;6K&lt;/span&gt;&lt;span class="dt"&gt;16 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Qwen 3.8 27B Just Hit 200 Token/s on Mac M5 (Lithos Metal)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=M2bkqSH69mE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;16 Eki&lt;/span&gt;&lt;span class="ch"&gt;The AI Automators&lt;/span&gt;&lt;span class="tt"&gt;Jev Can Now SEE (And It&amp;#x27;s Absurdly Fast)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=rJFXtjKynRU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;557&lt;/span&gt;&lt;span class="dt"&gt;16 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;This 150M Model Claims #1. I Tested It.&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=nXs9uQXa4fM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;524&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;This Tiny 2B AI vs Claude Code&amp;#x27;s Token Bill (Real Test)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=F62juaKg4rk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;490&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Amos 1 Small: 69.5% Accurate, Zero Tokens, 5 GB VRAM&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=b9CWEvxhXP4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;348&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Jeva - the Jev Competitor: This 4B Model Never Writes a Word. I Tested It&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=n8KwOkjTUWk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;305&lt;/span&gt;&lt;span class="dt"&gt;16 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;NVIDIA JUST DROPPED Nemotron 3.5 Super VL… Can Qwen 27B Beat It?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=mVuTVqIYwD0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;132&lt;/span&gt;&lt;span class="dt"&gt;16 Eki&lt;/span&gt;&lt;span class="ch"&gt;AI WITH Rithesh&lt;/span&gt;&lt;span class="tt"&gt;Liquid AI d1-Ultra vs Jev… WHO WINS? | Battle of AI Decision Models!&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t7"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#7dcfff"&gt;07&lt;/span&gt;&lt;i class="dot" style="background:#7dcfff"&gt;&lt;/i&gt;Yerel donanım ve yerel çalıştırma&lt;span class="cnt"&gt;11 video &amp;middot; 1.0M izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;128GB VRAM rehberi ve &amp;#x27;GPU alma, bu sunucuyu kur&amp;#x27; videoları yerel AI maliyet tartışmasını alevlendirdi; RTX Spark ve Mac kıyaslamaları ile donanımın sanılandan pahalılaştığı vurgusu öne çıktı.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=NA9D49kQFuU" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;552K&lt;/span&gt;&lt;span class="dt"&gt;10 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Every Ways to Get 128GB VRAM for Local AI at Full Context&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=z_sXLlLVUls" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;297K&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;Two Minute Papers&lt;/span&gt;&lt;span class="tt"&gt;OpenAI’s New AI Is Doing Things We Barely Understand&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=x_qmqqp6d7s" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;72K&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Why Your Mac Is Slow at Local AI (And It&amp;#x27;s Not the Chip)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=aQKGejbv-Rs" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;51K&lt;/span&gt;&lt;span class="dt"&gt;11 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Don&amp;#x27;t Buy a RTX Spark Laptop For Local AI Until You See This&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Y226-wwfCSw" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;47K&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;Kai&lt;/span&gt;&lt;span class="tt"&gt;Local AI Hardware Just Got Way More Expensive Than You Think&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=e7LiBfs9hOs" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;12 Eki&lt;/span&gt;&lt;span class="ch"&gt;The AI Automators&lt;/span&gt;&lt;span class="tt"&gt;Google just made AI search 100% offline&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=jKKO5aHY7hk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;822&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;NVIDIA&amp;#x27;s Sana: 1024px Image in 0.4 Seconds? I Tested It&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=kw3jiSJP86M" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;820&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;LlamaCPP on Native Windows | This is HUGE&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=LNy__Ge8OLQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;800&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;I Turned Structured Output Into a Game &#127918;&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=YKnlFU22j_o" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;171&lt;/span&gt;&lt;span class="dt"&gt;16 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Fish Audio Drama 3: Can You Run It Yourself?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=H4azmWcqO5g" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;0&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;Hugging Face&lt;/span&gt;&lt;span class="tt"&gt;Local AI 202: Get Massive Speed-ups with Speculative Decoding&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t8"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#e0af68"&gt;08&lt;/span&gt;&lt;i class="dot" style="background:#e0af68"&gt;&lt;/i&gt;Ses ve gerçek zamanlı çeviri&lt;span class="cnt"&gt;6 video &amp;middot; 12K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;Konuşarak ses tasarımı (Dialt), kesilebilen ses ajanları ve Gemini Live Translate ile ses, ajanların ana arayüzü olmaya devam ediyor.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=IBVmenOroPs" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;9K&lt;/span&gt;&lt;span class="dt"&gt;09 Eki&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;Order in any language with Gemini 3.5 Live Translate&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=qJplHFjyv2A" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;Google for Developers&lt;/span&gt;&lt;span class="tt"&gt;&#128214; Explore custom voice narration with Gemini 3.8 Flash TTS&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=VLcPY8QKnd0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;633&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;I Built a Voice Agent You Can Interrupt (Murf Falcon 2, Real Latency Numbers)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=xKPVUa-whpQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;415&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Design a Voice by Talking: Interactive Voice Design in Dialt&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=GTtGVgAjZns" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;356&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Two Ways to Switch Languages with Voice Assistants&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=5ArbD2pNN2c" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;31&lt;/span&gt;&lt;span class="dt"&gt;16 Eki&lt;/span&gt;&lt;span class="ch"&gt;Trelis Research&lt;/span&gt;&lt;span class="tt"&gt;Design a Voice by Talking | Dialt #Shorts&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t9"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#f7768e"&gt;09&lt;/span&gt;&lt;i class="dot" style="background:#f7768e"&gt;&lt;/i&gt;Ajanlar, kodlama ve yazılım fabrikaları&lt;span class="cnt"&gt;23 video &amp;middot; 168K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;&amp;#x27;Yazılım fabrikası&amp;#x27; kavramı, dark-factory hayalinin çöküş itirafları ve &amp;#x27;kod incelemesi artık darboğaz değil&amp;#x27; tezleriyle olgunlaşıyor; ajan altyapıları (harness, skill, erişim modeli) yeni tartışma başlığı.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=5YJGDElbCDY" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;26K&lt;/span&gt;&lt;span class="dt"&gt;11 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;Google&amp;#x27;s GPT-3 Moment for Databases &amp;amp; Spreadsheets (TabFM)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=PMOnd8T7Py0" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;21K&lt;/span&gt;&lt;span class="dt"&gt;12 Eki&lt;/span&gt;&lt;span class="ch"&gt;Cole Medin&lt;/span&gt;&lt;span class="tt"&gt;Is Learning to Build Software Still Worth It with AI? (My Honest Take)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=2i_ZyqDmtio" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;21K&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;Cole Medin&lt;/span&gt;&lt;span class="tt"&gt;How I Keep Up with AI Without Drowning in It (Free Skill)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=1HtydHmdhrQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;16K&lt;/span&gt;&lt;span class="dt"&gt;10 Eki&lt;/span&gt;&lt;span class="ch"&gt;Dave Ebbelaar&lt;/span&gt;&lt;span class="tt"&gt;What Companies Hire Freelance Engineers For in 2026&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=9DT8y0dnga8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;14K&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;The Gray Cat&lt;/span&gt;&lt;span class="tt"&gt;I Built an App with AI. These Tools Helped Me Understand It&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=FTaPMTB-LxI" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;14K&lt;/span&gt;&lt;span class="dt"&gt;12 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;I&amp;#x27;m Rebuilding Super Mario World With AI&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=bv1qgkbWhms" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;11K&lt;/span&gt;&lt;span class="dt"&gt;09 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;What is a Skill?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=R0ckdFjUF5U" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;7K&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;What is an Agent Harness?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=GQatuTWSRqY" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;6K&lt;/span&gt;&lt;span class="dt"&gt;16 Eki&lt;/span&gt;&lt;span class="ch"&gt;Cole Medin&lt;/span&gt;&lt;span class="tt"&gt;My AI Dark Factory Dream Is Dead (Here&amp;#x27;s What Actually Works)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=6VDjFXM1FbM" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;5K&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;From Punch Cards to AI Software Factories&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ycZaPLeJb30" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;5K&lt;/span&gt;&lt;span class="dt"&gt;10 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;What Is a Software Factory?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=8gCam2-iGLc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;5K&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;What is Deep Agents?&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=rduH9wnXto8" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;4K&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;How Stripe Built Kai&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=rv-X-Qvd7nk" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;16 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineering&lt;/span&gt;&lt;span class="tt"&gt;We’ve Been Giving AI Agents Access Wrong.&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=n6NJ37qJ7WQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;3K&lt;/span&gt;&lt;span class="dt"&gt;12 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;Why &amp;quot;best model&amp;quot; is the wrong question now&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=JKTTB1j2azw" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;16 Eki&lt;/span&gt;&lt;span class="ch"&gt;Burke Holland&lt;/span&gt;&lt;span class="tt"&gt;Never Used the GitHub Copilot App? Build a Game&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=z5ccyYiwOyo" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;2K&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;A PS5 Game Running Natively on a PC? I Checked the Repo&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=Lf1gAY4ZO3k" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;Why You Think the Frontier Model Is Always Better (It&amp;#x27;s Psychology)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=whnDiwcL4HE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;718&lt;/span&gt;&lt;span class="dt"&gt;16 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;Why Nobody Has Solved Continual Learning&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=eZ_TrdYnHRY" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;708&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;LangChain&lt;/span&gt;&lt;span class="tt"&gt;POV: You’re arriving at Interrupt London, The Agent Conference by LangChain.&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=ow6stdOcm2A" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;377&lt;/span&gt;&lt;span class="dt"&gt;14 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Can You Spot the $0.15 Legal AI? (GLM-Legal Test)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=HSIKzx6eUhg" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;240&lt;/span&gt;&lt;span class="dt"&gt;16 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;Claude Can Now Make AI Videos By Itself (Higgsfield MCP)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=4zOCI_xqezc" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;130&lt;/span&gt;&lt;span class="dt"&gt;16 Eki&lt;/span&gt;&lt;span class="ch"&gt;Prompt Engineer 48&lt;/span&gt;&lt;span class="tt"&gt;FLUX 3 Image Klein Hands-On is CRAZY&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t10"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#4ec9b0"&gt;10&lt;/span&gt;&lt;i class="dot" style="background:#4ec9b0"&gt;&lt;/i&gt;Görsel ve işitsel üretim&lt;span class="cnt"&gt;4 video &amp;middot; 160K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;NVIDIA Sana&amp;#x27;nın 0.4 saniyede 1024px görüntü üretimi ve Adobe uygulamalarını yeniden yazan AI&amp;#x27;lar, üretim araçlarının yeniden tanımlandığına işaret ediyor.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=TqxHhuOve2Q" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;123K&lt;/span&gt;&lt;span class="dt"&gt;10 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;AI Is Recreating Photoshop and Other Adobe Apps&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=av-xDOxS1rs" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;26K&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;pixaroma&lt;/span&gt;&lt;span class="tt"&gt;50 Free ComfyUI Automation Workflows (Ep38)&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=8l8nYfU81nA" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;10K&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;pixaroma&lt;/span&gt;&lt;span class="tt"&gt;ComfyUI Nodes Explained: The Basics for Beginners&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=CJRuCpw1SnU" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;DigitalSreeni&lt;/span&gt;&lt;span class="tt"&gt;14c - Zero-Shot Design of Drug-Binding Proteins&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t11"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#c0caf5"&gt;11&lt;/span&gt;&lt;i class="dot" style="background:#c0caf5"&gt;&lt;/i&gt;Frontier modeller ve AI tartışmaları&lt;span class="cnt"&gt;2 video &amp;middot; 265K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;&amp;#x27;Her şey çözüldü&amp;#x27; iddiaları ile &amp;#x27;sürekli öğrenme hâlâ çözülmedi&amp;#x27; itirafları aynı haftada yan yana; frontier modellerin gerçekten üstün mü yoksa algı mı olduğu sorgulanıyor.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=cX2kD2yQf88" target="_blank" rel="noopener"&gt;&lt;span class="vw v1"&gt;257K&lt;/span&gt;&lt;span class="dt"&gt;10 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Everything has been solved&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=PODwi90Ne8k" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;8K&lt;/span&gt;&lt;span class="dt"&gt;12 Eki&lt;/span&gt;&lt;span class="ch"&gt;Matthew Berman&lt;/span&gt;&lt;span class="tt"&gt;Human Creativity Still Matters&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;&lt;section class="theme" id="t12"&gt;&lt;div class="th-head"&gt;&lt;h2&gt;&lt;span class="tn" style="color:#d19a66"&gt;12&lt;/span&gt;&lt;i class="dot" style="background:#d19a66"&gt;&lt;/i&gt;Öğrenme ve temeller&lt;span class="cnt"&gt;4 video &amp;middot; 77K izlenme&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;&lt;p class="insight"&gt;PyTorch ve sinir ağları &amp;#x27;44 dakikada oturtma&amp;#x27; videoları yüksek izlenme almaya devam ederken, AI çağında yazılım öğrenmenin hâlâ değerli olup olmadığı dürüstçe tartışılıyor.&lt;/p&gt;&lt;div class="rows"&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=K1ksiPHfgVs" target="_blank" rel="noopener"&gt;&lt;span class="vw v2"&gt;61K&lt;/span&gt;&lt;span class="dt"&gt;13 Eki&lt;/span&gt;&lt;span class="ch"&gt;Zachary Huang&lt;/span&gt;&lt;span class="tt"&gt;Give me 31 minutes, I will make PyTorch click forever&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=wq3gFCU8bfE" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;14K&lt;/span&gt;&lt;span class="dt"&gt;10 Eki&lt;/span&gt;&lt;span class="ch"&gt;Zachary Huang&lt;/span&gt;&lt;span class="tt"&gt;Give me 24 minutes, I will make Neural Networks click forever&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=DRxhKJ4-MVQ" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;1K&lt;/span&gt;&lt;span class="dt"&gt;15 Eki&lt;/span&gt;&lt;span class="ch"&gt;Howfinity&lt;/span&gt;&lt;span class="tt"&gt;How to Use Substack for Beginners&lt;/span&gt;&lt;/a&gt;&lt;a class="row" href="https://www.youtube.com/watch?v=uI_KNZlAYU4" target="_blank" rel="noopener"&gt;&lt;span class="vw v3"&gt;387&lt;/span&gt;&lt;span class="dt"&gt;16 Eki&lt;/span&gt;&lt;span class="ch"&gt;Zachary Huang&lt;/span&gt;&lt;span class="tt"&gt;Give Me 42 Min, I Will Make PyTorch Indexing Click Forever&lt;/span&gt;&lt;/a&gt;&lt;/div&gt;&lt;/section&gt;
&lt;footer&gt;Bu rapor &lt;b style="color:#b8c1cb"&gt;Murat Karakaya Akademi&lt;/b&gt; tarafından derlenmiştir · Veri: YouTube RSS (canlı, API key yok) · pencere: 2026-10-09 .. 2026-10-16 (UTC, çalışma anından hesaplanır) · kanal başı max 15 video · PromptEngineer48 ve RitheshSreenivasan kanallarının RSS beslemesi 15 videoyla sınırlı; bu kanalların haftanın ilk günlerindeki videoları görülememiş olabilir. · render.py @ 16.10.2026 19:15 UTC&lt;/footer&gt;
&lt;/div&gt;&lt;/body&gt;&lt;/html&gt;</description><link>https://www.muratkarakaya.net/2026/10/youtube-ai-haftalk-trend-raporu-2026-10_0362366858.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-6157076101226871145</guid><pubDate>Sun, 27 Sep 2026 12:32:29 +0000</pubDate><atom:updated>2026-09-27T15:32:29.830+03:00</atom:updated><category domain="http://www.blogger.com/atom/ns#">apay Zeka</category><category domain="http://www.blogger.com/atom/ns#">Eleştirel Düşünme</category><category domain="http://www.blogger.com/atom/ns#">Jev</category><category domain="http://www.blogger.com/atom/ns#">Murat Karakaya Akademi</category><category domain="http://www.blogger.com/atom/ns#">OpenClaw</category><category domain="http://www.blogger.com/atom/ns#">Sosyal Medya</category><category domain="http://www.blogger.com/atom/ns#">Teknoloji</category><category domain="http://www.blogger.com/atom/ns#">Yapay Zeka Balonu</category><category domain="http://www.blogger.com/atom/ns#">Üretken Yapay Zeka</category><title>BU BALONU KİM ŞİŞİRİYOR?</title><description>&lt;!-- Blogger: Bu dosyanın tamamını yazı düzenleyicisinin HTML görünümüne yapıştırın. --&gt;
&lt;style&gt;
.mk-ai-essay{width:100%;max-width:100%;box-sizing:border-box;color:#243247;font-family:Arial,Helvetica,sans-serif;font-size:17px;line-height:1.85;overflow-wrap:anywhere;text-align:left}
.mk-ai-essay *{box-sizing:border-box}.mk-ai-essay p{margin:0 0 20px}.mk-ai-essay a{color:#146b85;text-decoration:underline;text-underline-offset:3px}.mk-ai-essay .mk-hero{padding:32px 26px;background:#14283e;color:#fff;border-radius:16px;margin:0 0 28px;border-top:6px solid #e7ac54}.mk-ai-essay .mk-eyebrow{font-size:12px;letter-spacing:2px;color:#f0c78c;margin:0 0 14px}.mk-ai-essay h2{font-size:29px;line-height:1.3;margin:0 0 18px;color:inherit}.mk-ai-essay h3{font-size:23px;line-height:1.4;color:#14283e;margin:36px 0 16px}.mk-ai-essay .mk-deck{font-size:18px;line-height:1.65;color:#e2eaf2;margin-bottom:18px}.mk-ai-essay .mk-meta{font-size:13px;color:#bdcddd;margin:0}.mk-ai-essay .mk-lead{font-size:20px;line-height:1.7;color:#14283e}.mk-ai-essay .mk-callout{margin:26px 0;padding:22px 24px;background:#f2f6f9;border-left:4px solid #247f90;border-radius:0 10px 10px 0}.mk-ai-essay .mk-callout p:last-child{margin-bottom:0}.mk-ai-essay .mk-quote{font-size:22px;line-height:1.6;color:#14283e}.mk-ai-essay .mk-note{font-size:14px;line-height:1.7;color:#536477}.mk-ai-essay .mk-step{padding:18px 20px;border:1px solid #dce5ed;border-radius:10px;margin:12px 0;background:#fff}.mk-ai-essay .mk-step p{margin:0}.mk-ai-essay .mk-step-title{display:block;color:#17677a;font-size:18px;margin:0 0 6px}.mk-ai-essay .mk-table-wrap{width:100%;overflow-x:auto;margin:22px 0}.mk-ai-essay table{width:100%;border-collapse:collapse;font-size:15px;line-height:1.65;table-layout:fixed}.mk-ai-essay th{background:#14283e;color:white;text-align:left;padding:14px}.mk-ai-essay td{padding:14px;border-bottom:1px solid #dae3eb;vertical-align:top}.mk-ai-essay tr:nth-child(even) td{background:#f5f8fa}.mk-ai-essay .mk-end{background:#fff5e6;border:1px solid #efd3a5;border-radius:12px;padding:24px;margin:30px 0}.mk-ai-essay .mk-end p:last-child{margin:0}.mk-ai-essay .mk-sources{border-top:1px solid #dae3eb;padding-top:20px;margin-top:34px;font-size:14px;line-height:1.7}.mk-ai-essay .mk-sources ol{padding-left:22px}.mk-ai-essay .mk-sources li{margin-bottom:10px}.mk-ai-essay .mk-tags{font-size:14px;color:#17677a;margin-top:24px;padding-top:18px;border-top:1px solid #dae3eb}@media(max-width:520px){.mk-ai-essay{font-size:16px}.mk-ai-essay .mk-hero{padding:25px 18px}.mk-ai-essay h2{font-size:25px}.mk-ai-essay h3{font-size:21px}.mk-ai-essay .mk-callout,.mk-ai-essay .mk-end{padding:18px}.mk-ai-essay .mk-quote{font-size:20px}.mk-ai-essay th,.mk-ai-essay td{padding:10px;font-size:14px}}
&lt;/style&gt;
&lt;article class="mk-ai-essay" lang="tr"&gt;
  &lt;header class="mk-hero"&gt;
    &lt;p class="mk-eyebrow"&gt;YAPAY ZEKÂ · SOSYAL MEDYA · ELEŞTİREL BAKIŞ&lt;/p&gt;
    &lt;h2&gt;Yapay Zekâ Balonunu Kim Şişiriyor? OpenClaw’dan Sıradaki “Büyük Devrime”&lt;/h2&gt;
    &lt;p class="mk-deck"&gt;“Hayatımı değiştirdi” videoları, Mac mini tavsiyeleri, bulut sunucuları ve indirim kuponları… Peki, ilk günkü heyecandan sonra geriye ne kaldı?&lt;/p&gt;
    
  &lt;/header&gt;

  &lt;p class="mk-lead"&gt;&#127880; Yapay zekâ alanında çok hızlı ve köklü değişiklikler oluyor. Ama bunların birçoğu da balon gibi şişiriliyor, sonra da patlıyor. Benim merak ettiğim, bu balonları kimin şişirdiği ve patladıktan sonra kimin dönüp arkasına baktığı.&lt;/p&gt;
  &lt;p&gt;&#128227; Özellikle YouTube veya X gibi sosyal medya ortamlarında, tek derdi çok sayıda tıklama almakmış gibi davranan sözde “yapay zekâ uzmanları” var. Yeni bir büyük dil modeli, araç, uygulama veya teknik mi çıktı? Daha ne işe yaradığını, hangi koşullarda başarılı olduğunu ve nerede zorlandığını anlamadan videolar hazır: “Her şey değişti!”, “Artık hiçbir şey eskisi gibi olmayacak!”, “Bunu kullanmayan geride kalacak!”&lt;/p&gt;
  &lt;p&gt;&#129300; Bir süre sonra insan şunu sormadan edemiyor: Madem her hafta her şey değişiyor, geçen haftaki büyük devrime ne oldu?&lt;/p&gt;

  &lt;h3&gt;&#129438; OpenClaw: Daha adını öğrenemeden hayatımız değişmişti&lt;/h3&gt;
  &lt;p&gt;OpenClaw etrafında oluşan tanıtım furyası, bu konudaki gözlemimi çok iyi anlatıyor. Clawdbot, ardından Moltbot, sonra OpenClaw… Daha aracın adını öğrenemeden bir sonraki ismine geçtik. :) Projenin kendi duyurusu da bu isim yolculuğunu anlatıyor; ilk isim değişikliğinin arkasında Anthropic’in marka konusundaki itirazı bulunuyor. &lt;a href="https://openclaw.ai/blog/introducing-openclaw" target="_blank" rel="noopener"&gt;[1]&lt;/a&gt;&lt;/p&gt;
  &lt;p&gt;Peter Steinberger’in başlattığı bu açık kaynaklı kişisel yapay zekâ asistanı, mesajlaşma uygulamaları üzerinden çeşitli işler yaptırma fikriyle dikkat çekti. Bu fikir elbette ilgi çekici. Ama sosyal medyada karşılaştığım anlatı, bir aracın olanaklarını tanıtmaktan çok daha ileri gidiyordu: Hayatlarını değiştirmişti, işlerini devralmıştı, hepimizin bir an önce bu devrime katılması gerekiyordu.&lt;/p&gt;
  &lt;p&gt;Steinberger’in yapay zekâ kodlama ajanlarıyla çalışma biçimi de bu hikâyenin bir parçası oldu. “Okumadığım kodu yayımlıyorum” açıklaması dikkat çekiciydi. Ancak röportajında planlama, mimari kararlar, testler ve geri bildirim süreçlerini de anlatıyordu. Bu ayrıntılar kaybolunca geriye kolay tüketilen bir masal kalıyor: Bir kişi yapay zekâya bir şey söyledi ve ortaya dünyayı değiştiren bir yazılım çıktı. Oysa aynı röportaj, deneyimli bir geliştiricinin mühendislik tercihlerini de içeriyor. &lt;a href="https://newsletter.pragmaticengineer.com/p/the-creator-of-clawd-i-ship-code" target="_blank" rel="noopener"&gt;[2]&lt;/a&gt;&lt;/p&gt;

  &lt;h3&gt;&#128722; “Büyük devrim”in hemen altında satın alma bağlantısı&lt;/h3&gt;
  &lt;p&gt;İş yalnızca aracı övmekle de kalmadı. İzlediğim içeriklerde dönüp dolaşıp iki öneri karşıma çıkıyordu: Mac mini alın veya bulutta bir sunucu kiralayıp orada çalıştırın. OpenClaw çevresindeki Mac mini heyecanı dönemin teknoloji haberlerine de yansıdı. Bulut sağlayıcıları ise OpenClaw için hazır kurulum seçenekleri sundu. &lt;a href="https://9to5mac.com/2026/01/27/clawdbot-sparks-mac-mini-memes-as-anthropic-forces-name-change/" target="_blank" rel="noopener"&gt;[3]&lt;/a&gt; &lt;a href="https://www.digitalocean.com/community/tutorials/how-to-run-openclaw" target="_blank" rel="noopener"&gt;[4]&lt;/a&gt;&lt;/p&gt;
  &lt;p&gt;Bazı tanıtımlarda sunucu kiralama önerisinin yanında içerik üreticisine ait bağlantı ve indirim kuponu da vardı. Örneğin “RIP Mac Mini!” başlıklı bir OpenClaw videosunun açıklamasında, Hostinger VPS için içerik üreticisinin adıyla bir kupon kodu sunuluyordu. Yani teknik bir öneriyle ticari yönlendirme aynı içerikte buluşuyordu. &lt;a href="https://www.youtube.com/watch?v=N7jXAQLjkX4" target="_blank" rel="noopener"&gt;[5]&lt;/a&gt;&lt;/p&gt;
  &lt;p&gt;Bir içerik üreticisinin sponsorluk alması veya indirim kodu paylaşması tek başına sorun değil. Benim itirazım, satın alma tavsiyesinin dayanağının belirsiz kalmasına. İzleyiciye ne kadar süre denenmiş bir çözüm öneriliyor? Hangi ihtiyacı karşılıyor? Yeni donanım veya abonelik gerçekten gerekli mi? Denemenin sonunda beklenen fayda çıkmazsa bu da aynı açıklıkla anlatılıyor mu?&lt;/p&gt;
  &lt;div class="mk-callout"&gt;
    &lt;p class="mk-quote"&gt;&#128172; “Bu araç hayatınızı değiştirecek!” diyen birinin, birkaç ay sonra “Peki, benim hayatımda neyi değiştirdi?” sorusuna da cevap vermesini beklemek fazla mı?&lt;/p&gt;
  &lt;/div&gt;
&lt;div class="separator" style="clear: both;"&gt;&lt;a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhM2i2AlDZeprfgnerhfwosDJ5LKuAH-SpfOdHR5W8eSizc7MOeUPmtnoLqlxiB63mhrXJi_HuHmyVMrHSoMwr2ZK5UczsZeEGfYUKSjw_MV6Q37psoEdAWIR-j-kc5qqrh-Nun6E-95Wcrno3nqr2WHq77RidgN9eLreFwGTGljbeEGucHJd0q-XJK2CEd/s1265/chrome_Ckye2YRunG.png" style="display: block; padding: 1em 0; text-align: center; "&gt;&lt;img alt="" border="0" width="400" data-original-height="1030" data-original-width="1265" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhM2i2AlDZeprfgnerhfwosDJ5LKuAH-SpfOdHR5W8eSizc7MOeUPmtnoLqlxiB63mhrXJi_HuHmyVMrHSoMwr2ZK5UczsZeEGfYUKSjw_MV6Q37psoEdAWIR-j-kc5qqrh-Nun6E-95Wcrno3nqr2WHq77RidgN9eLreFwGTGljbeEGucHJd0q-XJK2CEd/s400/chrome_Ckye2YRunG.png"/&gt;&lt;/a&gt;&lt;/div&gt;
   &lt;p class="mk-note"&gt;&#128202; Google Trends grafiği, seçilen dönem ve bölge içindeki göreli arama ilgisini gösterir; kullanıcı sayısını veya toplam video sayısını ölçmez. Buradaki gözlem, ilginin zirveye göre belirgin biçimde gerilemesidir.&lt;/p&gt;
  &lt;h3&gt;&#128201; Patlayan şey, şişirilen ilgi balonu&lt;/h3&gt;
  &lt;p&gt;OpenClaw için baktığım Google Trends ekranında, dünya genelinde son bir yıldaki arama ilgisi önce hızla yükseliyor, ardından zirvesinin oldukça altına geriliyor. İlk günlerdeki tanıtım furyasını düşündüğümde bu görüntü bana çok tanıdık geliyor: Büyük bir heyecan, peş peşe içerikler ve ardından giderek azalan görünürlük.&lt;/p&gt;
 
  &lt;p&gt;Benim “balon” dediğim tam da etrafında oluşturulan bu ilgi ve beklenti. Aracın yeni sürümü çıkabilir, geliştirilmesi sürebilir, birileri ondan fayda sağlamaya devam edebilir. Ama onu her gün büyük bir devrim diye anlatanların ilgisi nereye gitti? Asıl merak ettiğim bu.&lt;/p&gt;
  &lt;p&gt;Yeni gelişmeleri, yeni kullanım senaryolarını, uzun süreli deneyimlerini neden aynı heyecanla görmüyoruz? O gün “Artık bütün işlerimi buna yaptırıyorum” diyenlerin bugün hangi işlerini ona yaptırdığını öğrenmek isterdim. Benim karşıma, ilk günlerdeki tanıtım yoğunluğuyla kıyaslanabilecek bir deneyim paylaşımı çıkmıyor.&lt;/p&gt;

  &lt;h3&gt;&#128257; Her “devrim”de aynı içerik döngüsü&lt;/h3&gt;
  &lt;p&gt;Farklı araçlarda tekrar tekrar karşılaştığım bir yayın alışkanlığı var. Ürün değişiyor, video başlığı ve şaşkın yüz ifadesi çoğu zaman aynı kalıyor.&lt;/p&gt;
  &lt;div class="mk-step"&gt;&lt;span class="mk-step-title"&gt;01 · Duyuru gelir&lt;/span&gt;&lt;p&gt;Yeni bir model veya araç tanıtılır. Üreticinin seçtiği etkileyici örnekler ve performans iddiaları dolaşıma girer.&lt;/p&gt;&lt;/div&gt;
  &lt;div class="mk-step"&gt;&lt;span class="mk-step-title"&gt;02 · İddia büyür&lt;/span&gt;&lt;p&gt;“Belirli bir işte yararlı olabilir” cümlesi, birkaç paylaşım sonra “Bütün işleri değiştirecek” başlığına dönüşür.&lt;/p&gt;&lt;/div&gt;
  &lt;div class="mk-step"&gt;&lt;span class="mk-step-title"&gt;03 · Aynı anlatı çoğalır&lt;/span&gt;&lt;p&gt;Aynı demo, aynı örnek, aynı iddia farklı kanallarda karşımıza çıkar. Tek bir kaynaktan yayılan heyecan, çok sayıda bağımsız doğrulama varmış gibi görünebilir.&lt;/p&gt;&lt;/div&gt;
  &lt;div class="mk-step"&gt;&lt;span class="mk-step-title"&gt;04 · Satın alma önerisi gelir&lt;/span&gt;&lt;p&gt;Donanım, abonelik, sunucu, kurs veya kupon bağlantısı anlatıya eklenir. İzleyici, ihtiyacını değerlendirmeden harekete geçmeye teşvik edilir.&lt;/p&gt;&lt;/div&gt;
  &lt;div class="mk-step"&gt;&lt;span class="mk-step-title"&gt;05 · Sıradaki devrime geçilir&lt;/span&gt;&lt;p&gt;Önceki iddiaların sonucunu anlatmak için pek zaman kalmaz. Çünkü tanıtılmayı bekleyen yeni bir “her şeyi değiştiren” araç vardır.&lt;/p&gt;&lt;/div&gt;

  &lt;h3&gt;&#129436; Papağan gibi tekrar etmek, uzmanlık mı?&lt;/h3&gt;
  &lt;p&gt;Bir başka kanalda veya mecrada yayımlanan içeriği düşünmeden, incelemeden, denemeden ve yorumlamadan aktarmak kolay. Özellikle de hazır bir demo ve etkileyici bir başlık varsa. Ama aynı iddiayı Türkçeye çevirmek, ona birkaç şaşkınlık ifadesi eklemek ve yayımlamak ne kadar uzmanlık sayılır?&lt;/p&gt;
  &lt;p&gt;Uzmanlık iddiası, bilginin üzerine bir şey koymayı gerektirir. Hangi koşullarda işe yarıyor? Hangi durumda başarısız oluyor? Karşılaştırılan yöntemler aynı işi mi yapıyor? Gösterilen sonuç, seçilmiş tek bir başarılı örnek mi, yoksa tekrarlanabilen bir performans mı? Bu sorulara zaman ayırmadan yapılan tanıtım, izleyicinin karar vermesine ne kadar yardımcı olabilir?&lt;/p&gt;
  &lt;p&gt;Her içerik üreticisinden akademik makale beklemiyorum. Ama “İlk kez deniyorum”, “Bu üreticinin iddiası”, “Bunu kendi işimde henüz test etmedim” demek zor olmamalı. Denemediğini denemiş gibi, anlamadığını çözmüş gibi ve sınırlı bir örneği herkes için geçerliymiş gibi sunmamak da işin bir parçası.&lt;/p&gt;

  &lt;h3&gt;&#129514; İlk günkü demo yerine üçüncü ayın bilançosu&lt;/h3&gt;
  &lt;p&gt;Bence en değerli içeriklerden biri, bir aracın birkaç ay sonra yeniden değerlendirilmesi olurdu. Kurulum tamamlandıktan, ilk heyecan geçtikten ve günlük işlerin sıradan sorunları başladıktan sonra ne kaldı?&lt;/p&gt;
  &lt;div class="mk-table-wrap"&gt;
    &lt;table&gt;
      &lt;thead&gt;&lt;tr&gt;&lt;th scope="col"&gt;Tanıtımda duyduğumuz&lt;/th&gt;&lt;th scope="col"&gt;Devamında öğrenmek istediğimiz&lt;/th&gt;&lt;/tr&gt;&lt;/thead&gt;
      &lt;tbody&gt;
        &lt;tr&gt;&lt;td&gt;“Hayatımı değiştirdi.”&lt;/td&gt;&lt;td&gt;Hangi işi, ne sıklıkla ve ne kadar süredir yapıyor?&lt;/td&gt;&lt;/tr&gt;
        &lt;tr&gt;&lt;td&gt;“Bana saatler kazandırıyor.”&lt;/td&gt;&lt;td&gt;Kurulum, bakım ve hata düzeltme süreleri hesaba katıldığında ne kadar zaman kalıyor?&lt;/td&gt;&lt;/tr&gt;
        &lt;tr&gt;&lt;td&gt;“Çok ucuz çalışıyor.”&lt;/td&gt;&lt;td&gt;Model kullanımı, sunucu, abonelik ve donanım dâhil toplam maliyet ne?&lt;/td&gt;&lt;/tr&gt;
        &lt;tr&gt;&lt;td&gt;“Her şeyi kendi yapıyor.”&lt;/td&gt;&lt;td&gt;Kaç işte insan kontrolü veya müdahalesi gerekiyor?&lt;/td&gt;&lt;/tr&gt;
        &lt;tr&gt;&lt;td&gt;“Mutlaka almalısınız.”&lt;/td&gt;&lt;td&gt;Kim için gerekli? Kim mevcut araçlarıyla aynı ihtiyacı karşılayabilir?&lt;/td&gt;&lt;/tr&gt;
      &lt;/tbody&gt;
    &lt;/table&gt;
  &lt;/div&gt;
  &lt;p&gt;Hatta bazen en faydalı devam videosu şu olabilir: “Size önermiştim ama ben kullanmayı bıraktım. İşte nedenleri.” Böyle bir cümle, bana göre içerik üreticisinin güvenilirliğini azaltmaz. Tam tersine, deneyimini izlenme kaygısından daha fazla önemsediğini gösterir.&lt;/p&gt;

  &lt;h3&gt;&#128184; Balon patlayınca fatura kimde kalıyor?&lt;/h3&gt;
  &lt;p&gt;Bir sonraki videoya geçmek kolay. Ama izleyicinin aldığı cihaz, başlattığı abonelik, kiraladığı sunucu ve kurulum için harcadığı zaman yerinde duruyor. Bunların her biri, karşılığında işe yarayan bir sonuç alınıyorsa anlamlı olabilir. Sorun, ihtiyacın önüne “Bu devrimi kaçırmamalıyım” duygusunun geçmesi.&lt;/p&gt;
  &lt;p&gt;Bir başka maliyet de beklentiler. Her yeni araca olağanüstü yetenekler atfedildiğinde, gerçek kullanımda yaşanan sıradan bir başarısızlık bile büyük hayal kırıklığına dönüşebiliyor. Böylece işe yarayabilecek bir teknolojiyi anlamak için gerekli sabır da kayboluyor. Önce gereğinden fazla yüceltiyor, sonra beklediğimiz mucize gelmeyince hızla başka bir araca koşuyoruz.&lt;/p&gt;
  &lt;p&gt;Yapay zekâ balonu için çok şey söylenebilir. Ama bence bu balonu şişiren en önemli etkenlerden biri, bir başka mecradaki heyecanı sorgulamadan çoğaltan ve önceki tavsiyelerinin sonucunu takip etmeyen “Yapay Zekâ Uzmanları”.&lt;/p&gt;
&lt;div class="separator" style="clear: both;"&gt;&lt;a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEi816Z7gBVgKnJvtM3aoHElCpFGhBQG5lbboPY28XZJ32sfouEQDKZzGahn5EiR7__YN76CiO0o8PN-DlNXmmwp8mJaHhm7A3MyIjlv_GkNblnw4KJDu-QCiU1xfXQEjOdgT29Sd54vxjvhaM1VKYe4xGBUuPKt59UsC5HwJVBA4AQhJHMkU56n_YmlvLuE/s1243/chrome_t5WcUpveA2.png" style="display: block; padding: 1em 0; text-align: center; "&gt;&lt;img alt="" border="0" width="400" data-original-height="1032" data-original-width="1243" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEi816Z7gBVgKnJvtM3aoHElCpFGhBQG5lbboPY28XZJ32sfouEQDKZzGahn5EiR7__YN76CiO0o8PN-DlNXmmwp8mJaHhm7A3MyIjlv_GkNblnw4KJDu-QCiU1xfXQEjOdgT29Sd54vxjvhaM1VKYe4xGBUuPKt59UsC5HwJVBA4AQhJHMkU56n_YmlvLuE/s400/chrome_t5WcUpveA2.png"/&gt;&lt;/a&gt;&lt;/div&gt;
  &lt;h3&gt;&#127917; Sıradaki başlık: “LLM’leri bitiren Jev”&lt;/h3&gt;
  &lt;p&gt;Şimdi de Jev konuşuluyor. TypeSafe’ın resmî belgelerinde Jev; yazılımlar için yapılandırılmış kararlar üreten, seçenekler arasından seçim yapabilen ve olasılıklar döndüren bir model olarak tanımlanıyor. Aynı belgeler, onun metin üretmediğini, kod yazmadığını ve sohbet etmediğini açıkça belirtiyor. &lt;a href="https://docs.typesafe.ai/introduction/coding-agents" target="_blank" rel="noopener"&gt;[6]&lt;/a&gt;&lt;/p&gt;
  &lt;p&gt;Bu yaklaşım belirli uygulamalar için yararlı olabilir ve incelenmeyi hak eder. Ama buradan “LLM’ler bitti” sonucuna nasıl sıçrıyoruz? Belirli bir karar görevinde hız veya maliyet avantajı göstermek ile büyük dil modellerinin bütün kullanım alanlarını devralmak arasında epey mesafe var.&lt;/p&gt;
  &lt;p&gt;Yeni bir tekniğin değerini anlamak için önce hangi problemi çözdüğüne bakmak gerekiyor. Daha ilk gün onu bütün bir alanın sonu ilan ettiğimizde, teknik tartışma yine tanıtım gürültüsünün içinde kayboluyor.&lt;/p&gt;

  &lt;div class="mk-end"&gt;
    &lt;p&gt;&#128578; Günahları boyunlarına! Ama izleyenlerin zamanı, parası ve beklentileri de bu balonlarla birlikte uçup gidiyor.&lt;/p&gt;
    &lt;p&gt;&#128172; Siz hâlâ OpenClaw kullanıyor musunuz? Kullanıyorsanız hangi işinizde fayda sağlıyor? Bıraktıysanız neden bıraktınız? İlk günkü heyecandan sonra kalan gerçek deneyimleri merak ediyorum.&lt;/p&gt;
    &lt;p class="mk-quote"&gt;&#127880; Neyse, OpenClaw’ı bırakalım; sıradaki “büyük devrime” geçelim… “LLM’leri bitiren” Jev’i duydunuz mu? :)&lt;/p&gt;
  &lt;/div&gt;

  &lt;footer class="mk-sources"&gt;
    &lt;p&gt;&#128218; Kaynaklar ve okuma bağlantıları&lt;/p&gt;
    &lt;ol&gt;
      &lt;li&gt;&lt;a href="https://openclaw.ai/blog/introducing-openclaw" target="_blank" rel="noopener"&gt;OpenClaw: Introducing OpenClaw&lt;/a&gt; — Projenin isim geçmişi ve tanıtımı.&lt;/li&gt;
      &lt;li&gt;&lt;a href="https://newsletter.pragmaticengineer.com/p/the-creator-of-clawd-i-ship-code" target="_blank" rel="noopener"&gt;The Pragmatic Engineer: Peter Steinberger röportajı&lt;/a&gt; — Kodlama ajanları, planlama, mimari ve test yaklaşımı.&lt;/li&gt;
      &lt;li&gt;&lt;a href="https://9to5mac.com/2026/01/27/clawdbot-sparks-mac-mini-memes-as-anthropic-forces-name-change/" target="_blank" rel="noopener"&gt;9to5Mac: Clawdbot ve Mac mini heyecanı&lt;/a&gt; — Dönemin donanım gündemi.&lt;/li&gt;
      &lt;li&gt;&lt;a href="https://www.digitalocean.com/community/tutorials/how-to-run-openclaw" target="_blank" rel="noopener"&gt;DigitalOcean: OpenClaw kurulumu&lt;/a&gt; — Bulutta hazır kurulum örneği.&lt;/li&gt;
      &lt;li&gt;&lt;a href="https://www.youtube.com/watch?v=N7jXAQLjkX4" target="_blank" rel="noopener"&gt;YouTube: SUPER OpenClaw — RIP Mac Mini!&lt;/a&gt; — VPS ve kupon koduyla tanıtım örneği.&lt;/li&gt;
      &lt;li&gt;&lt;a href="https://docs.typesafe.ai/introduction/coding-agents" target="_blank" rel="noopener"&gt;TypeSafe: Jev with coding agents&lt;/a&gt; — Jev’in işlevi ve sohbet/kod üretiminden farkı.&lt;/li&gt;
    &lt;/ol&gt;
    &lt;p class="mk-note"&gt;Bu yazı, kaynaklarla desteklenen ürün bilgileri ile yazarın sosyal medya içeriklerine ilişkin gözlem ve yorumlarını bir araya getirir. Arama ilgisi değerlendirmesi, yazının hazırlanmasında kullanılan “OpenClaw · Dünya geneli · Geçen yıl · Google Arama” ayarlı Google Trends ekran görüntüsüne dayanır. YouTube içeriklerinin toplamına veya içerik üreticilerinin aktif kullanım oranına ilişkin sayısal bir iddia ileri sürülmemektedir.&lt;/p&gt;
  &lt;/footer&gt;
  &lt;p class="mk-tags"&gt;&#127991;️ #YapayZeka #OpenClaw #Jev #ÜretkenYapayZeka #YapayZekaBalonu #EleştirelDüşünme #SosyalMedya #Teknoloji #MuratKarakayaAkademi&lt;/p&gt;
&lt;/article&gt;</description><link>https://www.muratkarakaya.net/2026/09/bu-balonu-kim-sisiriyor.html</link><author>noreply@blogger.com (Murat Karakaya)</author><media:thumbnail xmlns:media="http://search.yahoo.com/mrss/" height="72" url="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEhM2i2AlDZeprfgnerhfwosDJ5LKuAH-SpfOdHR5W8eSizc7MOeUPmtnoLqlxiB63mhrXJi_HuHmyVMrHSoMwr2ZK5UczsZeEGfYUKSjw_MV6Q37psoEdAWIR-j-kc5qqrh-Nun6E-95Wcrno3nqr2WHq77RidgN9eLreFwGTGljbeEGucHJd0q-XJK2CEd/s72-c/chrome_Ckye2YRunG.png" width="72"/><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-2842949510606064684</guid><pubDate>Sat, 26 Sep 2026 20:42:16 +0000</pubDate><atom:updated>2026-09-26T23:42:16.384+03:00</atom:updated><category domain="http://www.blogger.com/atom/ns#">AGI</category><category domain="http://www.blogger.com/atom/ns#">AI Governance</category><category domain="http://www.blogger.com/atom/ns#">AI Regulation</category><category domain="http://www.blogger.com/atom/ns#">AI Safety</category><category domain="http://www.blogger.com/atom/ns#">AI Sovereignty</category><category domain="http://www.blogger.com/atom/ns#">Artificial Intelligence</category><category domain="http://www.blogger.com/atom/ns#">Frontier AI</category><category domain="http://www.blogger.com/atom/ns#">Murat Karakaya Academy</category><category domain="http://www.blogger.com/atom/ns#">Open Source AI</category><category domain="http://www.blogger.com/atom/ns#">Open Weights</category><category domain="http://www.blogger.com/atom/ns#">Tech Policy</category><title>SOMEBODY STOP US: Who Will Control Artificial Intelligence?</title><description>&lt;div style="width:100%;max-width:900px;margin:0 auto;padding:24px 16px;box-sizing:border-box;font-family:Arial,Helvetica,sans-serif;line-height:1.75;color:#202124;background:#ffffff;"&gt;

  &lt;h1 style="font-size:34px;line-height:1.2;margin:0 0 12px 0;color:#111827;"&gt;
    SOMEBODY STOP US: Who Will Control Artificial Intelligence?
  &lt;/h1&gt;

  &lt;p style="font-size:18px;color:#4b5563;margin:0 0 28px 0;"&gt;
    Is artificial intelligence really slipping out of control, or are we moving toward a different kind of danger: a future in which only a handful of governments and technology giants are able to develop the most advanced AI systems?
  &lt;/p&gt;

  &lt;div style="border-left:5px solid #111827;background:#f4f6f8;padding:20px 22px;margin:26px 0;box-sizing:border-box;"&gt;
    &lt;p style="margin:0;font-size:20px;font-weight:600;line-height:1.6;"&gt;
      For me, the central question is no longer simply, “How dangerous could artificial intelligence become?”&lt;br&gt;&lt;br&gt;
      The more important question is:&lt;br&gt;
      “Who will control artificial intelligence, and could that control mechanism eventually turn into a monopoly that decides who is allowed to develop AI?”
    &lt;/p&gt;
  &lt;/div&gt;

  &lt;h2 style="font-size:26px;margin:38px 0 14px 0;color:#111827;"&gt;2023: The First Serious Alarm Bells&lt;/h2&gt;

  &lt;p&gt;
    Today’s debate is not entirely new. The year 2023 marked an important turning point in discussions about the future of artificial intelligence.
  &lt;/p&gt;

  &lt;p&gt;
    Geoffrey Hinton, one of the pioneers of modern AI, left Google in 2023 and began speaking more openly about the risks of artificial intelligence. He warned that AI systems were developing much faster than he had expected and that increasingly capable systems could create serious risks in the future.
  &lt;/p&gt;

  &lt;p&gt;
    In the same year, an open letter published by the Future of Life Institute made an even more extraordinary proposal.
  &lt;/p&gt;

  &lt;p&gt;
    Among its signatories were Elon Musk, Yoshua Bengio, Steve Wozniak and many researchers and technology leaders.
  &lt;/p&gt;

  &lt;div style="background:#fff7ed;border:1px solid #fed7aa;border-radius:8px;padding:18px 20px;margin:22px 0;box-sizing:border-box;"&gt;
    &lt;p style="margin:0;font-size:18px;"&gt;
      The letter called for at least a six-month pause in the training of AI systems more powerful than GPT-4.
    &lt;/p&gt;
  &lt;/div&gt;

  &lt;p&gt;
    At the time, many people considered the proposal unnecessarily dramatic.
  &lt;/p&gt;

  &lt;p&gt;
    The dominant logic of the technology industry was exactly the opposite: build larger models, use more computing power, collect more data, and push performance further.
  &lt;/p&gt;

  &lt;p&gt;
    Three years later, however, the picture looks quite different.
  &lt;/p&gt;

  &lt;h2 style="font-size:26px;margin:38px 0 14px 0;color:#111827;"&gt;2026: This Time, the AI Companies Themselves Are Sounding the Alarm&lt;/h2&gt;

  &lt;p&gt;
    Today, concerns about AI safety are no longer being raised only by academics, activists or technology critics.
  &lt;/p&gt;

  &lt;p&gt;
    Some of the companies leading the AI race are themselves calling for stronger safety and governance mechanisms.
  &lt;/p&gt;

  &lt;p&gt;
    OpenAI has supported mandatory, capability-based national AI safety rules. Anthropic has argued that a technology developing as rapidly and affecting society as broadly as artificial intelligence should not be governed by industry alone, and that governments also need enforceable rules. Google DeepMind, through its Frontier Safety Framework, has established critical capability thresholds for areas including cybersecurity, biological risk, autonomy, manipulation and loss of control.
  &lt;/p&gt;

  &lt;p&gt;
    In other words, the debate is no longer purely theoretical.
  &lt;/p&gt;

  &lt;p&gt;
    The companies developing the most capable AI systems are themselves evaluating scenarios in which future models could enable powerful cyberattacks, biological threats, large-scale manipulation or serious control problems.
  &lt;/p&gt;

  &lt;p&gt;
    And the concern is no longer limited to malicious humans using AI.
  &lt;/p&gt;

  &lt;p&gt;
    Increasing attention is also being paid to highly autonomous systems that may pursue unintended objectives, circumvent human supervision or behave in unexpected ways.
  &lt;/p&gt;

  &lt;div style="border-left:5px solid #b91c1c;background:#fef2f2;padding:20px 22px;margin:26px 0;box-sizing:border-box;"&gt;
    &lt;p style="margin:0;"&gt;
      Therefore, it would be a mistake to dismiss the entire AI safety debate as manufactured fear. There are real technical and societal risks that deserve serious attention.
    &lt;/p&gt;
  &lt;/div&gt;

  &lt;h2 style="font-size:26px;margin:38px 0 14px 0;color:#111827;"&gt;But Another Race Is Taking Place at the Same Time&lt;/h2&gt;

  &lt;p&gt;
    This is where the second part of the problem begins.
  &lt;/p&gt;

  &lt;p&gt;
    Artificial intelligence is no longer merely a scientific or commercial technology.
  &lt;/p&gt;

  &lt;p&gt;
    It has become a strategic technology with implications for economic power, cybersecurity, defense, intelligence, scientific research and geopolitical influence.
  &lt;/p&gt;

  &lt;p&gt;
    This leads to another powerful argument, particularly in the United States:
  &lt;/p&gt;

  &lt;div style="background:#eff6ff;border:1px solid #bfdbfe;border-radius:8px;padding:18px 20px;margin:22px 0;box-sizing:border-box;"&gt;
    &lt;p style="margin:0;font-size:18px;"&gt;
      “If we slow down, there is no guarantee that our competitors will slow down with us.”
    &lt;/p&gt;
  &lt;/div&gt;

  &lt;p&gt;
    The fact that the U.S. AI Action Plan prominently frames the issue in terms of “Winning the AI Race” illustrates how central this logic has become.
  &lt;/p&gt;

  &lt;p&gt;
    The plan emphasizes accelerating innovation, expanding AI infrastructure and maintaining American leadership in artificial intelligence.
  &lt;/p&gt;

  &lt;p&gt;
    Particularly in the context of competition with China, AI policy is increasingly beginning to resemble national security policy rather than conventional innovation policy.
  &lt;/p&gt;

  &lt;p&gt;
    This creates a remarkably difficult equation:
  &lt;/p&gt;

  &lt;table style="width:100%;border-collapse:collapse;margin:24px 0;font-size:16px;"&gt;
    &lt;tr&gt;
      &lt;td style="width:50%;padding:16px;border:1px solid #d1d5db;background:#f9fafb;vertical-align:top;"&gt;
        &lt;strong&gt;On one side:&lt;/strong&gt;&lt;br&gt;
        More capable AI models could create serious risks. Testing, oversight, safety standards and, in some cases, restrictions may be necessary.
      &lt;/td&gt;
      &lt;td style="width:50%;padding:16px;border:1px solid #d1d5db;background:#f9fafb;vertical-align:top;"&gt;
        &lt;strong&gt;On the other side:&lt;/strong&gt;&lt;br&gt;
        If one country or company slows down, there is no guarantee that its competitors will do the same.
      &lt;/td&gt;
    &lt;/tr&gt;
  &lt;/table&gt;

  &lt;p&gt;
    My concern begins precisely where these two arguments intersect.
  &lt;/p&gt;

  &lt;h2 style="font-size:26px;margin:38px 0 14px 0;color:#111827;"&gt;Could Safety Regulation Become a Barrier to Entry?&lt;/h2&gt;

  &lt;p&gt;
    I also believe that artificial intelligence needs meaningful control mechanisms.
  &lt;/p&gt;

  &lt;p&gt;
    But I strongly disagree with the assumption that “AI needs control” must automatically lead to the conclusion that control should rest in the hands of a few large governments and a few major corporations.
  &lt;/p&gt;

  &lt;p&gt;
    Regulation creates a very important structural problem:
  &lt;/p&gt;

  &lt;p&gt;
    A safety requirement that is relatively easy for a trillion-dollar company to satisfy can become an impossible barrier for a startup, university laboratory or independent research group.
  &lt;/p&gt;

  &lt;p&gt;
    Imagine that developing a frontier model eventually requires extremely expensive licenses, mandatory certification, multimillion-dollar safety infrastructure, audits that can only be conducted by a small number of approved organizations, or special government authorization.
  &lt;/p&gt;

  &lt;p&gt;
    On paper, the rules might apply equally to everyone.
  &lt;/p&gt;

  &lt;p&gt;
    In practice, however, only a very small number of organizations might still be able to develop advanced AI.
  &lt;/p&gt;

  &lt;div style="border:1px solid #d1d5db;border-radius:8px;padding:20px;margin:24px 0;box-sizing:border-box;"&gt;
    &lt;p style="margin:0;font-weight:600;font-size:18px;"&gt;
      Would such regulation reduce competition?&lt;br&gt;
      Would it make it harder for new companies to emerge?&lt;br&gt;
      Would it restrict frontier AI research at universities?&lt;br&gt;
      Would it ultimately strengthen the position of the companies already dominating the field?
    &lt;/p&gt;
  &lt;/div&gt;

  &lt;p&gt;
    I believe these questions deserve much more attention.
  &lt;/p&gt;

  &lt;h2 style="font-size:26px;margin:38px 0 14px 0;color:#111827;"&gt;Why the Nuclear Technology Analogy Keeps Appearing&lt;/h2&gt;

  &lt;p&gt;
    Artificial intelligence is frequently compared with nuclear technology.
  &lt;/p&gt;

  &lt;p&gt;
    The analogy has obvious limitations.
  &lt;/p&gt;

  &lt;p&gt;
    An AI model is not a nuclear reactor. Software can be copied, modified and deployed in many different parts of the world.
  &lt;/p&gt;

  &lt;p&gt;
    Yet when we look specifically at frontier AI, one important similarity is becoming increasingly difficult to ignore.
  &lt;/p&gt;

  &lt;p&gt;
    Developing the most advanced systems requires rapidly increasing amounts of computing power, energy, data-center infrastructure, advanced semiconductors, highly specialized human capital and enormous financial resources.
  &lt;/p&gt;

  &lt;p&gt;
    If heavy regulatory requirements are added on top of these existing barriers, frontier AI development may naturally become concentrated in the hands of fewer and fewer actors.
  &lt;/p&gt;

  &lt;p&gt;
    And that creates an uncomfortable relationship between safety regulation and the centralization of technological power.
  &lt;/p&gt;

  &lt;h2 style="font-size:26px;margin:38px 0 14px 0;color:#111827;"&gt;Why Open-Weight AI Is at the Center of This Debate&lt;/h2&gt;

  &lt;p&gt;
    One of the most important battlegrounds in this discussion will be open-source AI and, more precisely, &lt;em&gt;open-weight&lt;/em&gt; models whose model weights are available for others to access and use.
  &lt;/p&gt;

  &lt;p&gt;
    Open models offer important advantages.
  &lt;/p&gt;

  &lt;p&gt;
    Universities can conduct independent research. Startups can develop products without becoming completely dependent on large API providers. Organizations can run models on their own infrastructure without sending sensitive data to third parties. Countries can develop systems tailored to their own languages, cultures and strategic needs.
  &lt;/p&gt;

  &lt;p&gt;
    At the same time, openness can also make it easier to remove safety mechanisms or modify highly capable models for malicious purposes.
  &lt;/p&gt;

  &lt;p&gt;
    Therefore, there is no simple answer to the question of whether AI should be open or closed.
  &lt;/p&gt;

  &lt;p&gt;
    But one distinction is essential.
  &lt;/p&gt;

  &lt;div style="background:#f0fdf4;border:1px solid #bbf7d0;border-radius:8px;padding:20px;margin:24px 0;box-sizing:border-box;"&gt;
    &lt;p style="margin:0;"&gt;
      Discussing the risks of open models is one thing.&lt;br&gt;&lt;br&gt;
      Using those risks as a justification for creating an AI ecosystem in which everyone becomes dependent on a handful of closed platforms is something entirely different.
    &lt;/p&gt;
  &lt;/div&gt;

  &lt;h2 style="font-size:26px;margin:38px 0 14px 0;color:#111827;"&gt;But There Is Also Important Counter-Evidence&lt;/h2&gt;

  &lt;p&gt;
    If we want to evaluate this issue seriously, we should not only look for evidence that supports our concerns.
  &lt;/p&gt;

  &lt;p&gt;
    We should also look for evidence that challenges them.
  &lt;/p&gt;

  &lt;p&gt;
    At present, there is no clear evidence that the major AI companies are collectively pursuing a strategy to eliminate open AI.
  &lt;/p&gt;

  &lt;p&gt;
    Anthropic CEO Dario Amodei has stated that his company does not advocate banning open-weight models and has described open models without dangerous capabilities as a public good.
  &lt;/p&gt;

  &lt;p&gt;
    More importantly, the current U.S. AI Action Plan explicitly includes policies aimed at supporting open-source and open-weight AI.
  &lt;/p&gt;

  &lt;p&gt;
    The plan recognizes the value of open models for startups, academic research, organizations working with sensitive data, and the broader international influence of the U.S. technology ecosystem.
  &lt;/p&gt;

  &lt;p&gt;
    Therefore, it would not be justified to claim that there is already a proven plan to shut down open-source AI.
  &lt;/p&gt;

  &lt;p&gt;
    My concern is different.
  &lt;/p&gt;

  &lt;p&gt;
    We might reach the same outcome even without any secret plan.
  &lt;/p&gt;

  &lt;div style="border-left:5px solid #7c3aed;background:#f5f3ff;padding:20px 22px;margin:26px 0;box-sizing:border-box;"&gt;
    &lt;p style="margin:0;font-size:19px;"&gt;
      Poorly designed incentives, excessively expensive safety obligations, concentration of computing power and national-security concerns could gradually create a system in which only a small number of actors are capable of developing frontier AI — even if nobody originally intended to create such a monopoly.
    &lt;/p&gt;
  &lt;/div&gt;

  &lt;h2 style="font-size:26px;margin:38px 0 14px 0;color:#111827;"&gt;The Next Stage: AI Sovereignty&lt;/h2&gt;

  &lt;p&gt;
    For me, this issue extends far beyond competition between companies.
  &lt;/p&gt;

  &lt;p&gt;
    The larger question concerns countries.
  &lt;/p&gt;

  &lt;p&gt;
    In the future, having strong AI researchers may not be enough to develop the most capable AI systems.
  &lt;/p&gt;

  &lt;p&gt;
    Countries may also need access to advanced GPUs, massive data centers, energy infrastructure, leading-edge semiconductor technologies, significant capital and international supply chains.
  &lt;/p&gt;

  &lt;p&gt;
    If international licensing systems, export controls or authorization mechanisms for frontier AI are added to this structure, technological sovereignty will become an even more important issue.
  &lt;/p&gt;

  &lt;p&gt;
    At that point, we may have to ask:
  &lt;/p&gt;

  &lt;div style="background:#111827;color:#ffffff;border-radius:10px;padding:24px;margin:26px 0;box-sizing:border-box;"&gt;
    &lt;p style="margin:0;font-size:21px;line-height:1.6;"&gt;
      If a country needs another country's permission to develop one of the most important general-purpose technologies of the future, can it really be considered technologically independent?
    &lt;/p&gt;
  &lt;/div&gt;

  &lt;p&gt;
    This question will become increasingly strategic for every country that wants to be not merely a user of artificial intelligence, but also a producer of it — including Türkiye.
  &lt;/p&gt;

  &lt;h2 style="font-size:26px;margin:38px 0 14px 0;color:#111827;"&gt;So Is the Solution “No Regulation at All”?&lt;/h2&gt;

  &lt;p&gt;
    No.
  &lt;/p&gt;

  &lt;p&gt;
    I believe both extremes in the AI debate are problematic.
  &lt;/p&gt;

  &lt;p&gt;
    The first extreme is to allow technology to advance without meaningful oversight and simply expect the market to solve serious safety problems.
  &lt;/p&gt;

  &lt;p&gt;
    The second extreme is to use safety concerns to create a system in which the ability to develop advanced AI is effectively controlled by a handful of companies and governments.
  &lt;/p&gt;

  &lt;p&gt;
    What we need is a governance model somewhere outside these two extremes.
  &lt;/p&gt;

  &lt;p&gt;
    Regulation should be based on measurable capabilities and risks rather than company names or model size alone.
  &lt;/p&gt;

  &lt;p&gt;
    Audit criteria should be transparent.
  &lt;/p&gt;

  &lt;p&gt;
    Universities and independent researchers should retain the ability to conduct meaningful AI safety research.
  &lt;/p&gt;

  &lt;p&gt;
    Compliance costs should not automatically push startups and smaller research organizations out of the field.
  &lt;/p&gt;

  &lt;p&gt;
    Not every open model should be treated as belonging to the same risk category.
  &lt;/p&gt;

  &lt;p&gt;
    A small language model and a frontier system capable of advanced biological design or sophisticated cyber operations should not necessarily be governed by identical rules.
  &lt;/p&gt;

  &lt;p&gt;
    And perhaps most importantly:
  &lt;/p&gt;

  &lt;p&gt;
    The rules governing the future of artificial intelligence should not be written only by the AI companies themselves.
  &lt;/p&gt;

  &lt;p&gt;
    When the companies being regulated also become the primary architects of the regulation, a natural conflict of interest may emerge.
  &lt;/p&gt;

  &lt;h2 style="font-size:26px;margin:38px 0 14px 0;color:#111827;"&gt;The Real Issue May Be Bigger Than Safety&lt;/h2&gt;

  &lt;p&gt;
    The developments of the past three years have made one thing increasingly clear to me.
  &lt;/p&gt;

  &lt;p&gt;
    AI safety is a real issue and deserves serious attention.
  &lt;/p&gt;

  &lt;p&gt;
    But from now on, it will not be enough to ask whether AI models themselves are safe.
  &lt;/p&gt;

  &lt;p&gt;
    We also need to ask how economic, technological and political power over artificial intelligence is distributed.
  &lt;/p&gt;

  &lt;p&gt;
    Because an AI ecosystem that is safe but completely controlled by a very small number of actors may not be an ideal outcome for humanity either.
  &lt;/p&gt;

  &lt;p&gt;
    I believe one of the most important technology-policy debates of the coming years will therefore be:
  &lt;/p&gt;

  &lt;div style="border:2px solid #111827;padding:24px;margin:28px 0;box-sizing:border-box;text-align:center;"&gt;
    &lt;p style="font-size:24px;line-height:1.5;font-weight:700;margin:0;"&gt;
      How should we control artificial intelligence?
    &lt;/p&gt;
    &lt;p style="font-size:24px;line-height:1.5;font-weight:700;margin:12px 0 0 0;"&gt;
      And more importantly:&lt;br&gt;
      Who will control those who control AI?
    &lt;/p&gt;
  &lt;/div&gt;

  &lt;p&gt;
    My concern, therefore, is not that we should avoid controlling artificial intelligence.
  &lt;/p&gt;

  &lt;p&gt;
    Quite the opposite.
  &lt;/p&gt;

  &lt;p&gt;
    We should build serious, scientific, transparent and democratically accountable mechanisms for governing advanced AI.
  &lt;/p&gt;

  &lt;p&gt;
    But while trying to make AI safer, we should be careful not to turn one of the most powerful technologies in history into the permanent privilege of a few governments and a few giant corporations.
  &lt;/p&gt;

  &lt;p&gt;
    Because one day, the biggest problem may not be that artificial intelligence has escaped our control.
  &lt;/p&gt;

  &lt;p&gt;
    It may be that control over artificial intelligence has become far too concentrated.
  &lt;/p&gt;

  &lt;hr style="border:0;border-top:1px solid #d1d5db;margin:38px 0 24px 0;"&gt;

  &lt;p style="font-size:16px;color:#374151;"&gt;
    &lt;strong&gt;What do you think?&lt;/strong&gt;&lt;br&gt;
    Do frontier AI systems require stronger global oversight? Or could the mechanisms created to ensure safety eventually produce an even greater risk by concentrating technological power in the hands of a few countries and corporations?
  &lt;/p&gt;

  &lt;p style="margin-top:28px;font-size:16px;line-height:1.8;"&gt;
    #ArtificialIntelligence #AISafety #AIGovernance #OpenSourceAI #OpenWeights #FrontierAI #AGI #TechPolicy #AIRegulation #AISovereignty #OpenAI #Anthropic #DeepMind #MuratKarakayaAkademi
  &lt;/p&gt;

&lt;/div&gt;
</description><link>https://www.muratkarakaya.net/2026/09/somebody-stop-us-who-will-control.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-8879837101229473294</guid><pubDate>Sat, 26 Sep 2026 20:40:18 +0000</pubDate><atom:updated>2026-09-26T23:40:18.641+03:00</atom:updated><category domain="http://www.blogger.com/atom/ns#">ABD Çin Rekabeti</category><category domain="http://www.blogger.com/atom/ns#">AGI</category><category domain="http://www.blogger.com/atom/ns#">AI Governance</category><category domain="http://www.blogger.com/atom/ns#">AI Safety</category><category domain="http://www.blogger.com/atom/ns#">Frontier AI</category><category domain="http://www.blogger.com/atom/ns#">Murat Karakaya Akademi</category><category domain="http://www.blogger.com/atom/ns#">Open Source AI</category><category domain="http://www.blogger.com/atom/ns#">Open Weight</category><category domain="http://www.blogger.com/atom/ns#">Teknoloji Politikası</category><category domain="http://www.blogger.com/atom/ns#">Yapay Zeka</category><category domain="http://www.blogger.com/atom/ns#">Yapay Zeka Güvenliği</category><title>BİRİ BİZİ DURDURSUN: Yapay Zekâyı Kim Kontrol Edecek?</title><description>&lt;div style="width:100%;max-width:900px;margin:0 auto;padding:24px 16px;box-sizing:border-box;font-family:Arial,Helvetica,sans-serif;line-height:1.75;color:#202124;background:#ffffff;"&gt;

  &lt;h1 style="font-size:34px;line-height:1.2;margin:0 0 12px 0;color:#111827;"&gt;
    BİRİ BİZİ DURDURSUN: Yapay Zekâyı Kim Kontrol Edecek?
  &lt;/h1&gt;

  &lt;p style="font-size:18px;color:#4b5563;margin:0 0 28px 0;"&gt;
    Yapay zekâ gerçekten kontrolden mi çıkıyor, yoksa “güvenlik” tartışmasının sonunda çok daha farklı bir tehlikeyle mi karşılaşacağız: Yapay zekâyı geliştirme gücünün birkaç devlet ve birkaç büyük şirketin elinde toplanması?
  &lt;/p&gt;

  &lt;div style="border-left:5px solid #111827;background:#f4f6f8;padding:20px 22px;margin:26px 0;box-sizing:border-box;"&gt;
    &lt;p style="margin:0;font-size:20px;font-weight:600;line-height:1.6;"&gt;
      Benim için asıl soru artık yalnızca “Yapay zekâ ne kadar tehlikeli olabilir?” değil.&lt;br&gt;&lt;br&gt;
      Asıl soru şu:&lt;br&gt;
      “Yapay zekâyı kim kontrol edecek ve bu kontrol mekanizması bir süre sonra kimin yapay zekâ geliştirebileceğine karar veren bir tekele dönüşecek mi?”
    &lt;/p&gt;
  &lt;/div&gt;

  &lt;h2 style="font-size:26px;margin:38px 0 14px 0;color:#111827;"&gt;2023: İlk ciddi alarm zilleri&lt;/h2&gt;
&lt;div class="separator" style="clear: both;"&gt;&lt;a href="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEh3Om6MPvKUzkJM-mnCbgK7OnaKajShh9CEQjTirUyQ-iiFyWDzs31C9JNkQ4euG_iNGAjwLVepL4jz8lyHNK2gpYNZeNJP7DNjPoyumvBbhb_kXAXy6rkI_PtTGdli34-hx9QVJxF2OpmKxXK_lUJzryOXyp4PryuFpaWhL_yjfG30F_UTOIYUkOhOdxdt/s800/1789474328481.jpeg" style="display: block; padding: 1em 0; text-align: center; "&gt;&lt;img alt="" border="0" width="320" data-original-height="800" data-original-width="800" src="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEh3Om6MPvKUzkJM-mnCbgK7OnaKajShh9CEQjTirUyQ-iiFyWDzs31C9JNkQ4euG_iNGAjwLVepL4jz8lyHNK2gpYNZeNJP7DNjPoyumvBbhb_kXAXy6rkI_PtTGdli34-hx9QVJxF2OpmKxXK_lUJzryOXyp4PryuFpaWhL_yjfG30F_UTOIYUkOhOdxdt/s320/1789474328481.jpeg"/&gt;&lt;/a&gt;&lt;/div&gt;
  &lt;p&gt;
    Bugünkü tartışma aslında yeni değil. 2023 yılı, yapay zekânın geleceği konusunda önemli bir kırılma noktasıydı.
  &lt;/p&gt;

  &lt;p&gt;
    Yapay zekânın öncülerinden Geoffrey Hinton, 2023'te Google'dan ayrıldıktan sonra yapay zekânın beklediğinden çok daha hızlı ilerlediğini ve bu sistemlerin gelecekte ciddi riskler yaratabileceğini açık biçimde dile getirmeye başladı.
  &lt;/p&gt;

  &lt;p&gt;
    Aynı yıl Future of Life Institute tarafından yayımlanan ve Elon Musk, Yoshua Bengio, Steve Wozniak gibi isimlerin de imzaladığı açık mektup çok daha sıra dışı bir çağrı yaptı:
  &lt;/p&gt;

  &lt;div style="background:#fff7ed;border:1px solid #fed7aa;border-radius:8px;padding:18px 20px;margin:22px 0;box-sizing:border-box;"&gt;
    &lt;p style="margin:0;font-size:18px;"&gt;
      GPT-4'ten daha güçlü yapay zekâ sistemlerinin eğitiminin en az altı ay süreyle durdurulması.
    &lt;/p&gt;
  &lt;/div&gt;

  &lt;p&gt;
    O gün bu çağrı bazılarına fazlasıyla dramatik gelmişti. Çünkü teknoloji dünyasının genel refleksi tam tersiydi: Daha büyük modeller, daha fazla hesaplama gücü, daha fazla veri ve daha yüksek performans.
  &lt;/p&gt;

  &lt;p&gt;
    Fakat aradan üç yıl geçtiğinde ilginç bir tablo ortaya çıktı.
  &lt;/p&gt;

  &lt;h2 style="font-size:26px;margin:38px 0 14px 0;color:#111827;"&gt;2026: Bu kez alarmı yapay zekâ şirketlerinin kendileri çalıyor&lt;/h2&gt;

  &lt;p&gt;
    Bugün artık yalnızca akademisyenler veya teknoloji eleştirmenleri değil, yapay zekâ yarışının tam merkezindeki şirketler de çok daha güçlü güvenlik ve yönetişim mekanizmalarından söz ediyor.
  &lt;/p&gt;

  &lt;p&gt;
    OpenAI, 2026 yılında ulusal düzeyde zorunlu ve yetenek-temelli yapay zekâ güvenliği düzenlemelerini desteklediğini açıkladı. Anthropic, yapay zekâ gibi hızla ilerleyen ve çok geniş etki alanına sahip bir teknolojinin yalnızca sektör tarafından yönetilemeyeceğini, devletlerin de bağlayıcı kurallar koyması gerektiğini savunuyor. Google DeepMind ise Frontier Safety Framework kapsamında biyogüvenlik, siber güvenlik, otonomi, manipülasyon ve kontrol kaybı gibi alanlarda kritik yetenek eşikleri tanımlıyor.
  &lt;/p&gt;

  &lt;p&gt;
    Yani artık tartışma teorik olmaktan çıkıyor.
  &lt;/p&gt;

  &lt;p&gt;
    Şirketler kendi geliştirdikleri sistemlerin gelecekte çok ciddi siber saldırı, biyolojik tehdit, kitlesel manipülasyon veya kontrol problemi yaratabilecek yeteneklere ulaşabileceğini bizzat kendi güvenlik dokümanlarında değerlendiriyor.
  &lt;/p&gt;

  &lt;p&gt;
    Üstelik mesele yalnızca kötü niyetli insanların yapay zekâyı kullanması da değil. Giderek daha fazla araştırma, yüksek derecede otonom sistemlerin hedeflerinden sapması, insan denetimini atlatması veya beklenmeyen davranışlar geliştirmesi gibi senaryoları da inceliyor.
  &lt;/p&gt;

  &lt;div style="border-left:5px solid #b91c1c;background:#fef2f2;padding:20px 22px;margin:26px 0;box-sizing:border-box;"&gt;
    &lt;p style="margin:0;"&gt;
      Dolayısıyla “yapay zekâ güvenliği” meselesini tamamen uydurulmuş bir korku olarak görmek doğru olmaz. Ortada gerçekten üzerinde çalışılması gereken teknik ve toplumsal riskler var.
    &lt;/p&gt;
  &lt;/div&gt;

  &lt;h2 style="font-size:26px;margin:38px 0 14px 0;color:#111827;"&gt;Fakat aynı anda başka bir yarış sürüyor&lt;/h2&gt;

  &lt;p&gt;
    Sorunun ikinci tarafı burada başlıyor.
  &lt;/p&gt;

  &lt;p&gt;
    Yapay zekâ artık yalnızca bilimsel veya ticari bir teknoloji değil. Aynı zamanda ekonomik güç, siber güvenlik, savunma, istihbarat, bilimsel araştırma ve jeopolitik üstünlük açısından stratejik bir teknoloji.
  &lt;/p&gt;

  &lt;p&gt;
    Bu nedenle ABD'nin yaklaşımında başka bir güçlü argüman daha bulunuyor:
  &lt;/p&gt;

  &lt;div style="background:#eff6ff;border:1px solid #bfdbfe;border-radius:8px;padding:18px 20px;margin:22px 0;box-sizing:border-box;"&gt;
    &lt;p style="margin:0;font-size:18px;"&gt;
      “Biz yavaşlarsak, rakiplerimiz yavaşlamak zorunda değil.”
    &lt;/p&gt;
  &lt;/div&gt;

  &lt;p&gt;
    ABD'nin resmi AI Action Plan'ının adının bile “Winning the AI Race” olması bu yaklaşımın ne kadar merkezi olduğunu gösteriyor. Plan; inovasyonun hızlandırılmasını, yapay zekâ altyapısının büyütülmesini ve ABD'nin küresel yapay zekâ liderliğinin korunmasını temel stratejik hedefler arasında görüyor.
  &lt;/p&gt;

  &lt;p&gt;
    Özellikle Çin ile rekabet söz konusu olduğunda teknoloji politikası giderek klasik bir inovasyon politikasından çok ulusal güvenlik politikasına benzemeye başlıyor.
  &lt;/p&gt;

  &lt;p&gt;
    Böylece ortaya son derece zor bir denklem çıkıyor:
  &lt;/p&gt;

  &lt;table style="width:100%;border-collapse:collapse;margin:24px 0;font-size:16px;"&gt;
    &lt;tr&gt;
      &lt;td style="width:50%;padding:16px;border:1px solid #d1d5db;background:#f9fafb;vertical-align:top;"&gt;
        &lt;strong&gt;Bir tarafta:&lt;/strong&gt;&lt;br&gt;
        Daha güçlü modeller ciddi riskler yaratabilir. Denetim, test, güvenlik standartları ve bazı durumlarda kısıtlamalar gerekebilir.
      &lt;/td&gt;
      &lt;td style="width:50%;padding:16px;border:1px solid #d1d5db;background:#f9fafb;vertical-align:top;"&gt;
        &lt;strong&gt;Diğer tarafta:&lt;/strong&gt;&lt;br&gt;
        Bir ülke veya şirket yavaşladığında rakiplerinin de yavaşlayacağının hiçbir garantisi yok.
      &lt;/td&gt;
    &lt;/tr&gt;
  &lt;/table&gt;

  &lt;p&gt;
    İşte benim kaygım bu iki argümanın kesiştiği noktada başlıyor.
  &lt;/p&gt;

  &lt;h2 style="font-size:26px;margin:38px 0 14px 0;color:#111827;"&gt;Güvenlik mekanizması zamanla bir giriş bariyerine dönüşebilir mi?&lt;/h2&gt;

  &lt;p&gt;
    Yapay zekâ için bir kontrol mekanizmasına ihtiyaç olduğunu ben de düşünüyorum.
  &lt;/p&gt;

  &lt;p&gt;
    Ancak “kontrol gerekli” önermesinden otomatik olarak “kontrol birkaç büyük devletin ve birkaç büyük şirketin elinde olmalı” sonucunun çıkmasını son derece tehlikeli buluyorum.
  &lt;/p&gt;

  &lt;p&gt;
    Çünkü düzenleme tasarımında çok kritik bir sorun vardır:
  &lt;/p&gt;

  &lt;p&gt;
    Büyük şirketlerin kolayca yerine getirebildiği bir güvenlik zorunluluğu, küçük şirketler, üniversiteler ve bağımsız araştırmacılar için aşılması imkânsız bir maliyet oluşturabilir.
  &lt;/p&gt;

  &lt;p&gt;
    Örneğin bir frontier model geliştirmek için çok yüksek lisans ücretleri, zorunlu sertifikasyon süreçleri, milyonlarca dolarlık güvenlik altyapısı, yalnızca belirli kuruluşların gerçekleştirebildiği denetimler veya devlet tarafından verilen özel izinler gerekli hale gelirse, kâğıt üzerinde herkes için geçerli olan bir güvenlik rejimi pratikte yalnızca birkaç kuruluşun faaliyet gösterebildiği bir pazara dönüşebilir.
  &lt;/p&gt;

  &lt;div style="border:1px solid #d1d5db;border-radius:8px;padding:20px;margin:24px 0;box-sizing:border-box;"&gt;
    &lt;p style="margin:0;font-weight:600;font-size:18px;"&gt;
      Bu durumda düzenleme rekabeti azaltmaz mı?&lt;br&gt;
      Yeni şirketlerin ortaya çıkmasını zorlaştırmaz mı?&lt;br&gt;
      Üniversitelerin ve bağımsız araştırmacıların frontier AI çalışmalarını sınırlandırmaz mı?&lt;br&gt;
      Sonuçta mevcut büyük oyuncuların konumunu daha da güçlendirmez mi?
    &lt;/p&gt;
  &lt;/div&gt;

  &lt;p&gt;
    İşte üzerinde durmamız gereken asıl meselelerden biri bence bu.
  &lt;/p&gt;

  &lt;h2 style="font-size:26px;margin:38px 0 14px 0;color:#111827;"&gt;Nükleer teknoloji benzetmesi neden akla geliyor?&lt;/h2&gt;

  &lt;p&gt;
    Yapay zekâ tartışmalarında nükleer teknoloji benzetmesi sık sık karşımıza çıkıyor.
  &lt;/p&gt;

  &lt;p&gt;
    Bu benzetmenin sınırları var. Yapay zekâ bir nükleer reaktör değil. Bir yazılım modeli kopyalanabilir, dünyanın farklı yerlerinde çalıştırılabilir ve çok daha geniş bir geliştirici topluluğu tarafından değiştirilebilir.
  &lt;/p&gt;

  &lt;p&gt;
    Fakat frontier AI açısından önemli bir benzerlik giderek belirginleşiyor:
  &lt;/p&gt;

  &lt;p&gt;
    En ileri sistemleri geliştirebilmek için gereken hesaplama gücü, enerji, veri merkezleri, gelişmiş yarı iletkenler, uzman insan kaynağı ve sermaye miktarı hızla büyüyor.
  &lt;/p&gt;

  &lt;p&gt;
    Buna bir de ağır düzenleyici yük eklenirse, frontier AI geliştirme kapasitesi doğal olarak giderek daha az sayıda aktörün elinde toplanabilir.
  &lt;/p&gt;

  &lt;p&gt;
    Ve tam burada güvenlik için oluşturulan sistem ile teknolojik gücün merkezileşmesi arasında rahatsız edici bir ilişki ortaya çıkabilir.
  &lt;/p&gt;

  &lt;h2 style="font-size:26px;margin:38px 0 14px 0;color:#111827;"&gt;Açık ağırlıklı yapay zekâ neden bu tartışmanın merkezinde?&lt;/h2&gt;

  &lt;p&gt;
    Bu tartışmanın en kritik cephelerinden biri açık kaynak ve özellikle &lt;em&gt;open-weight&lt;/em&gt;, yani model ağırlıkları erişilebilir yapay zekâ modelleri olacak.
  &lt;/p&gt;

  &lt;p&gt;
    Açık modellerin çok önemli avantajları var.
  &lt;/p&gt;

  &lt;p&gt;
    Üniversiteler araştırma yapabiliyor. Küçük şirketler dev API sağlayıcılarına tamamen bağımlı olmadan ürün geliştirebiliyor. Kurumlar hassas verilerini üçüncü taraf sistemlerine göndermeden kendi altyapılarında model çalıştırabiliyor. Ülkeler kendi dilleri, kültürleri ve ihtiyaçları için modeller geliştirebiliyor.
  &lt;/p&gt;

  &lt;p&gt;
    Diğer taraftan aynı açıklık, çok yetenekli bir model söz konusu olduğunda bazı güvenlik mekanizmalarının kaldırılmasını veya modelin kötü niyetli amaçlarla değiştirilmesini de kolaylaştırabilir.
  &lt;/p&gt;

  &lt;p&gt;
    Dolayısıyla “açık mı kapalı mı?” sorusunun basit bir cevabı yok.
  &lt;/p&gt;

  &lt;p&gt;
    Fakat burada önemli bir ayrım yapmalıyız:
  &lt;/p&gt;

  &lt;div style="background:#f0fdf4;border:1px solid #bbf7d0;border-radius:8px;padding:20px;margin:24px 0;box-sizing:border-box;"&gt;
    &lt;p style="margin:0;"&gt;
      Açık modellerin risklerinin tartışılması başka bir şeydir.&lt;br&gt;&lt;br&gt;
      Bu riskler gerekçe gösterilerek açık model geliştirme ekosisteminin tamamının birkaç kapalı platforma bağımlı hale getirilmesi başka bir şeydir.
    &lt;/p&gt;
  &lt;/div&gt;

  &lt;h2 style="font-size:26px;margin:38px 0 14px 0;color:#111827;"&gt;Ama mevcut tabloya bakınca önemli bir karşı kanıt da var&lt;/h2&gt;

  &lt;p&gt;
    Burada kendi kaygımızı doğrulayacak kanıtları aramak kadar, onu yanlışlayabilecek kanıtlara da bakmak gerekiyor.
  &lt;/p&gt;

  &lt;p&gt;
    Bugünkü veriler büyük yapay zekâ şirketlerinin ortaklaşa “açık yapay zekâyı ortadan kaldıralım” şeklinde bir politika yürüttüğünü göstermiyor.
  &lt;/p&gt;

  &lt;p&gt;
    Örneğin Anthropic CEO'su Dario Amodei 2026'da şirketinin açık ağırlıklı modellerin yasaklanmasını savunmadığını açıkça belirtti ve tehlikeli yeteneklere sahip olmayan açık modelleri bir kamu yararı olarak tanımladı.
  &lt;/p&gt;

  &lt;p&gt;
    Daha da önemlisi, ABD'nin mevcut AI Action Plan'ı da açık kaynak ve açık ağırlıklı modelleri destekleyen ayrı bir politika bölümü içeriyor. Belgede açık modellerin startup'lar, akademik araştırma, hassas verilerle çalışan kurumlar ve ABD'nin teknolojik etkisi açısından değerli olduğu açık biçimde ifade ediliyor.
  &lt;/p&gt;

  &lt;p&gt;
    Bu nedenle bugün itibarıyla “plan çoktan yapılmış, açık kaynak kapatılacak” demek için elimizde yeterli kanıt yok.
  &lt;/p&gt;

  &lt;p&gt;
    Benim endişem biraz farklı.
  &lt;/p&gt;

  &lt;p&gt;
    Böyle bir sonuca ulaşmak için gizli bir planın var olması bile gerekmeyebilir.
  &lt;/p&gt;

  &lt;div style="border-left:5px solid #7c3aed;background:#f5f3ff;padding:20px 22px;margin:26px 0;box-sizing:border-box;"&gt;
    &lt;p style="margin:0;font-size:19px;"&gt;
      Yanlış tasarlanmış teşvikler, aşırı pahalı güvenlik yükümlülükleri, hesaplama gücünün birkaç şirkette yoğunlaşması ve ulusal güvenlik kaygıları; kimse baştan böyle bir tekel tasarlamasa bile bizi fiilen aynı noktaya götürebilir.
    &lt;/p&gt;
  &lt;/div&gt;

  &lt;h2 style="font-size:26px;margin:38px 0 14px 0;color:#111827;"&gt;Bir sonraki aşama: Yapay zekâ egemenliği&lt;/h2&gt;

  &lt;p&gt;
    Benim için mesele yalnızca şirket rekabeti de değil.
  &lt;/p&gt;

  &lt;p&gt;
    Daha büyük soru ülkeler açısından ortaya çıkıyor.
  &lt;/p&gt;

  &lt;p&gt;
    Gelecekte en güçlü yapay zekâ sistemlerini geliştirebilmek için yalnızca model geliştirme bilgisine sahip olmak yetmeyebilir.
  &lt;/p&gt;

  &lt;p&gt;
    Gelişmiş GPU'lara, dev veri merkezlerine, enerji altyapısına, ileri yarı iletken teknolojilerine, büyük miktarda sermayeye ve uluslararası tedarik zincirlerine erişim gerekecek.
  &lt;/p&gt;

  &lt;p&gt;
    Bunların üzerine uluslararası lisanslama, ihracat kontrolü veya frontier AI geliştirme izinleri gibi mekanizmalar eklendiğinde teknoloji egemenliği sorusu daha da önemli hale gelebilir.
  &lt;/p&gt;

  &lt;p&gt;
    O zaman şu soruyu sormak zorunda kalabiliriz:
  &lt;/p&gt;

  &lt;div style="background:#111827;color:#ffffff;border-radius:10px;padding:24px;margin:26px 0;box-sizing:border-box;"&gt;
    &lt;p style="margin:0;font-size:21px;line-height:1.6;"&gt;
      Bir ülke geleceğin en önemli genel amaçlı teknolojilerinden birini geliştirebilmek için başka bir ülkenin iznine ihtiyaç duyarsa, gerçekten teknolojik olarak bağımsız sayılabilir mi?
    &lt;/p&gt;
  &lt;/div&gt;

  &lt;p&gt;
    Bu soru Türkiye dahil olmak üzere yapay zekâda yalnızca kullanıcı değil üretici olmak isteyen bütün ülkeler için giderek daha stratejik hale gelecektir.
  &lt;/p&gt;

  &lt;h2 style="font-size:26px;margin:38px 0 14px 0;color:#111827;"&gt;O halde çözüm “hiç düzenleme olmasın” mı?&lt;/h2&gt;

  &lt;p&gt;
    Hayır.
  &lt;/p&gt;

  &lt;p&gt;
    Bence yapay zekâ tartışmasında iki uç yaklaşım da problemli.
  &lt;/p&gt;

  &lt;p&gt;
    Birinci uç, teknolojinin kendi kendine ilerlemesine izin verip ciddi riskleri tamamen piyasanın çözmesini beklemek.
  &lt;/p&gt;

  &lt;p&gt;
    İkinci uç ise güvenlik gerekçesiyle yapay zekâ geliştirme yetkisini fiilen birkaç şirkete ve birkaç devlete bırakmak.
  &lt;/p&gt;

  &lt;p&gt;
    İhtiyacımız olan şey bu ikisinin dışında bir yönetişim modeli.
  &lt;/p&gt;

  &lt;p&gt;
    Düzenlemeler şirket isimlerine veya yalnızca model büyüklüğüne değil, ölçülebilir yetenek ve risklere dayanmalı. Denetim kriterleri şeffaf olmalı. Üniversitelerin ve bağımsız araştırmacıların güvenlik araştırması yapabilmesi korunmalı. Küçük şirketleri otomatik olarak piyasanın dışına itecek maliyetler yaratılmamalı.
  &lt;/p&gt;

  &lt;p&gt;
    Her açık model aynı risk sınıfında değerlendirilmemeli. Küçük bir dil modeli ile biyolojik tasarım veya ileri siber saldırı kapasitesi gösterebilen frontier modelin aynı kurallara tabi olması mantıklı değil.
  &lt;/p&gt;

  &lt;p&gt;
    Ve belki de en önemlisi:
  &lt;/p&gt;

  &lt;p&gt;
    Yapay zekânın geleceğini belirleyen kurallar yalnızca yapay zekâ şirketleri tarafından yazılmamalı.
  &lt;/p&gt;

  &lt;p&gt;
    Çünkü düzenlemenin öznesi olan şirketlerin aynı zamanda düzenlemenin nasıl olması gerektiğini belirleyen başlıca aktörler haline gelmesi doğal bir çıkar çatışması yaratabilir.
  &lt;/p&gt;

  &lt;h2 style="font-size:26px;margin:38px 0 14px 0;color:#111827;"&gt;Asıl mesele güvenlikten daha büyük olabilir&lt;/h2&gt;

  &lt;p&gt;
    Son üç yılda yaşanan gelişmeler bana bir şeyi açık biçimde gösteriyor:
  &lt;/p&gt;

  &lt;p&gt;
    Yapay zekânın güvenliği gerçekten konuşmamız gereken bir konu.
  &lt;/p&gt;

  &lt;p&gt;
    Ancak bundan sonra yalnızca modellerin güvenli olup olmadığını tartışmak yeterli olmayacak.
  &lt;/p&gt;

  &lt;p&gt;
    Aynı zamanda yapay zekâ üzerindeki ekonomik, teknik ve politik gücün nasıl dağıtıldığını da konuşmak zorundayız.
  &lt;/p&gt;

  &lt;p&gt;
    Çünkü güvenli fakat birkaç aktörün tamamen kontrol ettiği bir yapay zekâ ekosistemi de insanlık için ideal bir sonuç olmayabilir.
  &lt;/p&gt;

  &lt;p&gt;
    Bence önümüzdeki yılların en önemli teknoloji politikası tartışmalarından biri tam olarak bu olacak:
  &lt;/p&gt;

  &lt;div style="border:2px solid #111827;padding:24px;margin:28px 0;box-sizing:border-box;text-align:center;"&gt;
    &lt;p style="font-size:24px;line-height:1.5;font-weight:700;margin:0;"&gt;
      Yapay zekâyı nasıl kontrol edeceğiz?
    &lt;/p&gt;
    &lt;p style="font-size:24px;line-height:1.5;font-weight:700;margin:12px 0 0 0;"&gt;
      Ve daha önemlisi:&lt;br&gt;
      Kontrol edenleri kim kontrol edecek?
    &lt;/p&gt;
  &lt;/div&gt;

  &lt;p&gt;
    Bu nedenle benim kaygım “yapay zekâyı kontrol etmeyelim” değil.
  &lt;/p&gt;

  &lt;p&gt;
    Tam tersine.
  &lt;/p&gt;

  &lt;p&gt;
    Yapay zekâ üzerinde ciddi, bilimsel, şeffaf ve demokratik denetim mekanizmaları oluşturalım.
  &lt;/p&gt;

  &lt;p&gt;
    Fakat bunu yaparken güvenliği sağlamak adına dünyanın en güçlü teknolojilerinden birini birkaç devletin ve birkaç dev şirketin kalıcı ayrıcalığı haline getirmeyelim.
  &lt;/p&gt;

  &lt;p&gt;
    Çünkü bir gün asıl problem yapay zekânın kontrolden çıkması değil, yapay zekânın kontrolünün fazlasıyla küçük bir grubun elinde toplanması olabilir.
  &lt;/p&gt;

  &lt;hr style="border:0;border-top:1px solid #d1d5db;margin:38px 0 24px 0;"&gt;

  &lt;p style="font-size:16px;color:#374151;"&gt;
    &lt;strong&gt;Siz ne düşünüyorsunuz?&lt;/strong&gt;&lt;br&gt;
    Frontier yapay zekâ sistemleri için daha güçlü bir küresel kontrol mekanizması mı gerekiyor? Yoksa bu mekanizmaların zamanla teknolojik gücü birkaç ülke ve birkaç şirketin elinde toplaması daha büyük bir risk mi?
  &lt;/p&gt;

  &lt;p style="margin-top:28px;font-size:16px;line-height:1.8;"&gt;
    #YapayZeka #ArtificialIntelligence #AISafety #AIGovernance #OpenSourceAI #OpenWeights #FrontierAI #AGI #TeknolojiPolitikası #YapayZekaGuvenligi #MuratKarakayaAkademi
  &lt;/p&gt;

&lt;/div&gt;
</description><link>https://www.muratkarakaya.net/2026/09/biri-bizi-durdursun-yapay-zekay-kim.html</link><author>noreply@blogger.com (Murat Karakaya)</author><media:thumbnail xmlns:media="http://search.yahoo.com/mrss/" height="72" url="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEh3Om6MPvKUzkJM-mnCbgK7OnaKajShh9CEQjTirUyQ-iiFyWDzs31C9JNkQ4euG_iNGAjwLVepL4jz8lyHNK2gpYNZeNJP7DNjPoyumvBbhb_kXAXy6rkI_PtTGdli34-hx9QVJxF2OpmKxXK_lUJzryOXyp4PryuFpaWhL_yjfG30F_UTOIYUkOhOdxdt/s72-c/1789474328481.jpeg" width="72"/><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-8732088612179014080</guid><pubDate>Sat, 26 Sep 2026 20:31:26 +0000</pubDate><atom:updated>2026-09-26T23:31:26.195+03:00</atom:updated><category domain="http://www.blogger.com/atom/ns#">AI</category><category domain="http://www.blogger.com/atom/ns#">Career Strategy</category><category domain="http://www.blogger.com/atom/ns#">Early Specialization</category><category domain="http://www.blogger.com/atom/ns#">LLM</category><category domain="http://www.blogger.com/atom/ns#">Meta Learning</category><category domain="http://www.blogger.com/atom/ns#">RAG</category><category domain="http://www.blogger.com/atom/ns#">Software</category><category domain="http://www.blogger.com/atom/ns#">Technology</category><title>The "Early Specialization" Fallacy in AI Careers: Why New Grads Should Embrace a Broad Profile</title><description>&lt;div style="font-family: 'Georgia', serif; line-height: 1.8; color: #2c3e50; max-width: 800px; margin: auto;"&gt;

  &lt;p style="font-size: 1.15em; color: #555;"&gt;Recently, I received an email from a highly talented senior student whose portfolio I reviewed. It perfectly summarized the immense anxiety felt by almost everyone stepping into today's artificial intelligence ecosystem. She wrote:&lt;/p&gt;

  &lt;blockquote style="background-color: #f8f9fa; border-left: 6px solid #2c3e50; margin: 2em 0; padding: 1.5em; font-style: italic; border-radius: 4px; box-shadow: 2px 2px 8px rgba(0,0,0,0.05);"&gt;
    &lt;p style="margin: 0; font-size: 1.1em; color: #34495e;"&gt;"I want to get into the AI field. I've experimented with a lot of different things—local LLM integration, RAG (Retrieval-Augmented Generation), Edge AI projects... But when I look at my CV, I feel scattered. Should I narrow my portfolio and specialize in a single area before getting a job?"&lt;/p&gt;
  &lt;/blockquote&gt;

  &lt;p&gt;This question is the clearest example of a fallacy inherited from traditional software engineering teachings hitting a brick wall in the &lt;strong&gt;"Age of AI"&lt;/strong&gt;. In the traditional world, the "T-shaped person" (deep in one area, broad in others) was the ideal. But when it comes to AI, the rules of the game change entirely.&lt;/p&gt;

  &lt;h2 style="font-family: 'Helvetica Neue', Helvetica, Arial, sans-serif; color: #1a252f; font-size: 1.6em; margin-top: 40px; margin-bottom: 20px; border-bottom: 2px solid #ecf0f1; padding-bottom: 10px;"&gt;The Illusion of "Maturity" Destroyed by Speed&lt;/h2&gt;
  
  &lt;p&gt;Almost all the topics we lose sleep over today, wondering &lt;em&gt;"should I specialize in this?"&lt;/em&gt; (Transformer architectures, Agentic RAG systems, vector databases, SLMs), are products of the last 5 to 6 years. More importantly, none of these technologies are fully mature yet.&lt;/p&gt;
  
  &lt;p&gt;Right now, the lifespan of a technique in the industry is measured in months. Yesterday, &lt;em&gt;Prompt Engineering&lt;/em&gt; was an area of expertise; today, libraries like &lt;em&gt;DSPy&lt;/em&gt; make it algorithmic. Yesterday, massive cloud models were essential; today, 8-billion parameter models running locally via &lt;em&gt;Apple MLX&lt;/em&gt; or &lt;em&gt;Ollama&lt;/em&gt; work wonders.&lt;/p&gt;

  &lt;p&gt;The morning after you declare yourself an "expert" in a framework or specific architecture, a brand new paper or open-source code is published that renders that architecture completely obsolete. In such a hyper-dynamic landscape, someone without industry experience making a decision for &lt;strong&gt;premature specialization&lt;/strong&gt; is essentially cutting off the branch they are sitting on.&lt;/p&gt;

  &lt;h2 style="font-family: 'Helvetica Neue', Helvetica, Arial, sans-serif; color: #1a252f; font-size: 1.6em; margin-top: 40px; margin-bottom: 20px; border-bottom: 2px solid #ecf0f1; padding-bottom: 10px;"&gt;The True Expectation of Startups and Companies&lt;/h2&gt;
  
  &lt;p&gt;When you enter the industry's kitchen, you see that reality is much more pragmatic. Most companies and agile startups aren't looking for laboratory perfection. They are looking for profiles that can &lt;strong&gt;integrate&lt;/strong&gt; these rapidly changing new tools, models, and libraries into existing business processes without wasting time.&lt;/p&gt;
  
  &lt;p&gt;Therefore, employers prefer a &lt;strong&gt;"technology integrator"&lt;/strong&gt;—someone who has dirtied their hands with RAG systems, tried running models on edge devices (Edge AI) when necessary, and experienced those constraints firsthand—over a "narrow specialist" who has academically deepened in a single algorithm but remains blind to the rest of the system. That broad spectrum you see as being "scattered" is actually proof of your &lt;em&gt;“adaptation muscle,”&lt;/em&gt; which is your biggest competitive advantage.&lt;/p&gt;

  &lt;h2 style="font-family: 'Helvetica Neue', Helvetica, Arial, sans-serif; color: #1a252f; font-size: 1.6em; margin-top: 40px; margin-bottom: 20px; border-bottom: 2px solid #ecf0f1; padding-bottom: 10px;"&gt;Where Does True Depth Begin?&lt;/h2&gt;
  
  &lt;p&gt;So, will there be no specialization at all? Of course there will be. But not through "choices" made from the outside before getting a job.&lt;/p&gt;
  
  &lt;p&gt;True specialization begins when you sit at the desk. You deepen when you try to build a RAG system on your company's 50 million lines of dirty corporate data and your system starts "hallucinating." You specialize in that area when you are forced into model quantization to increase inference speed. The needs of the project naturally and inevitably pull you into a deep well in a specific area. Sometimes this focus comes at the cost of staying away from other hot trends, but the depth you acquire is now a &lt;strong&gt;"useful"&lt;/strong&gt; depth; not a hypothetical one.&lt;/p&gt;

  &lt;h2 style="font-family: 'Helvetica Neue', Helvetica, Arial, sans-serif; color: #1a252f; font-size: 1.6em; margin-top: 40px; margin-bottom: 20px; border-bottom: 2px solid #ecf0f1; padding-bottom: 10px;"&gt;My Advice to New Graduates&lt;/h2&gt;
  
  &lt;p&gt;In summary; do not fear the diversity in your portfolio. Instead of becoming a disciple of a single topic in the AI field, build plenty of prototypes with different techniques. Make mistakes, see the limits. Master the breadth of the ecosystem so that tomorrow, when a problem arises at the company you work for, you have the vision to say, &lt;em&gt;"It would be more effective if we solved this with that newly released agent architecture, rather than LangChain."&lt;/em&gt;&lt;/p&gt;
  
  &lt;p&gt;Let the epaulet called specialization settle on your shoulders naturally while you are wrestling with real-world problems.&lt;/p&gt;

  &lt;div style="background-color: #f0f7fb; border-left: 4px solid #3498db; padding: 20px; margin-top: 45px; font-family: 'Helvetica Neue', Helvetica, Arial, sans-serif;"&gt;
    &lt;p style="margin: 0; font-weight: bold; color: #2c3e50; font-size: 1.1em;"&gt;What do you think?&lt;/p&gt;
    &lt;p style="margin: 10px 0 0 0; color: #555;"&gt;How is this dizzying speed in the AI ecosystem affecting your career planning? Specialize in a narrow field, or become an integrator with a broad perspective? Let's discuss your ideas in the comments.&lt;/p&gt;
  &lt;/div&gt;

  &lt;div style="margin-top: 40px; font-family: 'Helvetica Neue', Helvetica, Arial, sans-serif; font-size: 0.9em; color: #7f8c8d; font-weight: 500; word-wrap: break-word;"&gt;
    &lt;p&gt;#ArtificialIntelligence #CareerStrategy #NewGrad #SoftwareDevelopment #MachineLearning #LLM #RAG #EdgeAI #TechCareer #SoftwareEngineering #FutureOfWork #MuratKarakayaAkademi&lt;/p&gt;
  &lt;/div&gt;

&lt;/div&gt;</description><link>https://www.muratkarakaya.net/2026/09/the-early-specialization-fallacy-in-ai.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-2793337164590127903</guid><pubDate>Sat, 26 Sep 2026 20:29:41 +0000</pubDate><atom:updated>2026-09-26T23:29:41.005+03:00</atom:updated><category domain="http://www.blogger.com/atom/ns#">Erken Uzmanlaşma</category><category domain="http://www.blogger.com/atom/ns#">Kariyer Stratejisi</category><category domain="http://www.blogger.com/atom/ns#">LLM</category><category domain="http://www.blogger.com/atom/ns#">Meta Öğrenme</category><category domain="http://www.blogger.com/atom/ns#">Murat Karakaya Akademi</category><category domain="http://www.blogger.com/atom/ns#">RAG</category><category domain="http://www.blogger.com/atom/ns#">Teknoloji</category><category domain="http://www.blogger.com/atom/ns#">Yapay Zeka</category><category domain="http://www.blogger.com/atom/ns#">Yazılım</category><category domain="http://www.blogger.com/atom/ns#">Yeni Mezun</category><title>Yapay Zeka Kariyerinde "Erken Uzmanlaşma" Yanılgısı: Yeni Mezunlar Neden Dağınık Kalmalı?</title><description>&lt;div style="font-family: 'Georgia', serif; line-height: 1.8; color: #2c3e50; max-width: 800px; margin: auto;"&gt;

  &lt;p style="font-size: 1.15em; color: #555;"&gt;Geçtiğimiz günlerde, portföyünü incelediğim çok yetenekli bir son sınıf öğrencisinden, bugünün yapay zeka ekosistemine yeni adım atan herkesin hissettiği o devasa anksiyeteyi özetleyen bir e-posta aldım. Şöyle diyordu:&lt;/p&gt;

  &lt;blockquote style="background-color: #f8f9fa; border-left: 6px solid #2c3e50; margin: 2em 0; padding: 1.5em; font-style: italic; border-radius: 4px; box-shadow: 2px 2px 8px rgba(0,0,0,0.05);"&gt;
    &lt;p style="margin: 0; font-size: 1.1em; color: #34495e;"&gt;"Yerel LLM entegrasyonu, RAG (Retrieval-Augmented Generation) denemeleri, Edge AI projeleri... Birçok alana dokundum ama CV’me baktığımda kendimi aşırı dağınık görüyorum. İşe girmeden önce portföyümü daraltıp tek bir alanda uzmanlaşmalı mıyım?"&lt;/p&gt;
  &lt;/blockquote&gt;

  &lt;p&gt;Bu soru, geleneksel yazılım mühendisliği öğretilerinden miras kalan bir yanılgının, &lt;strong&gt;"Yapay Zeka Çağı'nda"&lt;/strong&gt; duvara çarpmasının en net örneğidir. Geleneksel dünyada "T-tipi insan" (bir alanda çok derin, diğerlerinde yüzeysel bilgi sahibi olan) makbuldü. Ancak işler yapay zeka olunca, oyunun kuralları tamamen değişiyor.&lt;/p&gt;

  &lt;h2 style="font-family: 'Helvetica Neue', Helvetica, Arial, sans-serif; color: #1a252f; font-size: 1.6em; margin-top: 40px; margin-bottom: 20px; border-bottom: 2px solid #ecf0f1; padding-bottom: 10px;"&gt;Hızın Yıktığı "Olgunluk" İllüzyonu&lt;/h2&gt;
  
  &lt;p&gt;Bugün uğruna "uzmanlaşsam mı?" diye uykularımızı kaçıran konuların (Transformer mimarileri, Agentic RAG sistemleri, vektör veritabanları, SLM'ler) neredeyse tamamı son 5-6 yılın ürünü. Daha da önemlisi, bu teknolojilerin hiçbiri henüz tam anlamıyla olgunlaşmış değil.&lt;/p&gt;
  
  &lt;p&gt;Şu an sektörde bir tekniğin ömrü aylarla ölçülüyor. Dün &lt;em&gt;Prompt Engineering&lt;/em&gt; bir uzmanlık alanıydı; bugün &lt;em&gt;DSPy&lt;/em&gt; gibi kütüphaneler bu işi algoritmik hale getiriyor. Dün devasa bulut modelleri şarttı; bugün &lt;em&gt;Apple MLX&lt;/em&gt; veya &lt;em&gt;Ollama&lt;/em&gt; ile lokalde koşan 8 milyar parametreli modeller harikalar yaratıyor. &lt;/p&gt;

  &lt;p&gt;Siz bir çerçevede (framework) veya spesifik bir mimaride "uzmanlaştığınızı" ilan ettiğiniz günün ertesi sabahı, o mimariyi tamamen çöpe atan yepyeni bir makale veya açık kaynak kod yayınlanıyor. Böyle hiper-dinamik bir zeminde, henüz sektör tecrübesi olmayan birinin &lt;strong&gt;erken uzmanlaşma&lt;/strong&gt; (premature specialization) kararı alması, aslında bindiği dalı kesmesidir.&lt;/p&gt;

  &lt;h2 style="font-family: 'Helvetica Neue', Helvetica, Arial, sans-serif; color: #1a252f; font-size: 1.6em; margin-top: 40px; margin-bottom: 20px; border-bottom: 2px solid #ecf0f1; padding-bottom: 10px;"&gt;Girişimlerin ve Şirketlerin Asıl Beklentisi&lt;/h2&gt;
  
  &lt;p&gt;Sektörün mutfağına girdiğinizde gerçeğin çok daha pragmatik olduğunu görürsünüz. Birçok şirket ve çevik girişim, laboratuvar mükemmelliği aramıyor. Onlar; hızla değişen bu yeni araçları, modelleri ve kütüphaneleri vakit kaybetmeden mevcut iş süreçlerine &lt;strong&gt;entegre edecek&lt;/strong&gt; profiller arıyor.&lt;/p&gt;
  
  &lt;p&gt;Dolayısıyla işveren; tek bir algoritmada akademik düzeyde derinleşmiş ama sistemin geri kalanına kör kalmış bir "dar uzman" yerine, elini RAG sistemlerine de bulamış, yeri geldiğinde uç cihazda (Edge AI) model koşturmayı da denemiş, oradaki kısıtları bizzat yaşayarak görmüş bir &lt;strong&gt;"teknoloji entegratörünü"&lt;/strong&gt; tercih ediyor. Sizin "dağınıklık" olarak gördüğünüz o geniş yelpaze, aslında en büyük rekabet avantajınız olan &lt;em&gt;"adaptasyon kasınızın"&lt;/em&gt; bir kanıtı.&lt;/p&gt;

  &lt;h2 style="font-family: 'Helvetica Neue', Helvetica, Arial, sans-serif; color: #1a252f; font-size: 1.6em; margin-top: 40px; margin-bottom: 20px; border-bottom: 2px solid #ecf0f1; padding-bottom: 10px;"&gt;Gerçek Derinlik Nerede Başlar?&lt;/h2&gt;
  
  &lt;p&gt;Peki uzmanlaşma hiç mi olmayacak? Elbette olacak. Ancak işe girmeden önce dışarıdan bakarak yapılan "seçimlerle" değil.&lt;/p&gt;
  
  &lt;p&gt;Gerçek bir uzmanlaşma, masaya oturduğunuzda başlar. Şirketinizin 50 milyon satırlık kirli kurumsal verisi üzerinde RAG yapmaya çalıştığınızda ve sisteminiz "halüsinasyon" görmeye başladığında derinleşirsiniz. Hızı artırmak için model kuantizasyonuna (quantization) mecbur kaldığınızda o alanda uzmanlaşırsınız. Projenin ihtiyaçları, sizi doğal ve mecburi bir şekilde spesifik bir alanda derin bir kuyuya çeker. Bazen bu odaklanma, diğer sıcak trendlerden uzak kalma pahasına olur, ancak edindiğiniz derinlik artık &lt;strong&gt;"işe yarar"&lt;/strong&gt; bir derinliktir; varsayımsal değil.&lt;/p&gt;

  &lt;h2 style="font-family: 'Helvetica Neue', Helvetica, Arial, sans-serif; color: #1a252f; font-size: 1.6em; margin-top: 40px; margin-bottom: 20px; border-bottom: 2px solid #ecf0f1; padding-bottom: 10px;"&gt;Yeni Mezunlara Tavsiyem&lt;/h2&gt;
  
  &lt;p&gt;Özetle; portföyünüzdeki çeşitlilikten korkmayın. YZ alanında tek bir konunun müridi olmak yerine, farklı tekniklerle bol bol prototip üretin. Hata yapın, sınırları görün. Ekosistemin genişliğine hakim olun ki, yarın çalıştığınız firmada bir problemle karşılaşıldığında &lt;em&gt;"Bunu LangChain ile değil, şu yeni çıkan ajan mimarisiyle çözersek daha efektif olur"&lt;/em&gt; diyebilecek vizyonunuz olsun.&lt;/p&gt;
  
  &lt;p&gt;Bırakın uzmanlaşma denen o apolet, siz gerçek dünya sorunlarıyla boğuşurken omuzlarınıza kendiliğinden konsun.&lt;/p&gt;

  &lt;div style="background-color: #f0f7fb; border-left: 4px solid #3498db; padding: 20px; margin-top: 45px; font-family: 'Helvetica Neue', Helvetica, Arial, sans-serif;"&gt;
    &lt;p style="margin: 0; font-weight: bold; color: #2c3e50; font-size: 1.1em;"&gt;Siz ne düşünüyorsunuz?&lt;/p&gt;
    &lt;p style="margin: 10px 0 0 0; color: #555;"&gt;Yapay zeka ekosistemindeki bu baş döndürücü hız, kariyer planlamanızı nasıl etkiliyor? Dar bir alanda uzmanlaşmak mı, yoksa geniş bir perspektifte entegratör olmak mı? Fikirlerinizi yorumlarda tartışalım.&lt;/p&gt;
  &lt;/div&gt;

  &lt;div style="margin-top: 40px; font-family: 'Helvetica Neue', Helvetica, Arial, sans-serif; font-size: 0.9em; color: #7f8c8d; font-weight: 500; word-wrap: break-word;"&gt;
    &lt;p&gt;#YapayZeka #KariyerStratejisi #YeniMezun #YazilimGelistirme #MakineOgrenmesi #LLM #RAG #EdgeAI #TeknolojiKariyeri #YazilimMuhendisligi #GeleceginMeslekleri #MuratKarakayaAkademi&lt;/p&gt;
  &lt;/div&gt;

&lt;/div&gt;</description><link>https://www.muratkarakaya.net/2026/09/yapay-zeka-kariyerinde-erken-uzmanlasma.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-1948726412982639855</guid><pubDate>Sat, 26 Sep 2026 20:17:15 +0000</pubDate><atom:updated>2026-09-26T23:17:15.537+03:00</atom:updated><category domain="http://www.blogger.com/atom/ns#">Agentic AI</category><category domain="http://www.blogger.com/atom/ns#">Artificial Intelligence</category><category domain="http://www.blogger.com/atom/ns#">China</category><category domain="http://www.blogger.com/atom/ns#">Claude Fable 5</category><category domain="http://www.blogger.com/atom/ns#">GPT-5.6</category><category domain="http://www.blogger.com/atom/ns#">Inkling</category><category domain="http://www.blogger.com/atom/ns#">Kimi K3</category><category domain="http://www.blogger.com/atom/ns#">LLM</category><category domain="http://www.blogger.com/atom/ns#">Open Source AI</category><category domain="http://www.blogger.com/atom/ns#">Open Weights</category><category domain="http://www.blogger.com/atom/ns#">Qwen3.8</category><category domain="http://www.blogger.com/atom/ns#">USA</category><title>Are Closed LLMs Losing Their Lead? Open-Weight Models Are Moving Closer to the Frontier</title><description>&lt;div style="max-width:100%; box-sizing:border-box; font-family:Arial,Helvetica,sans-serif; line-height:1.75; color:#222; background:#ffffff;"&gt;

  &lt;div style="max-width:920px; margin:0 auto; box-sizing:border-box;"&gt;

```
&lt;div style="padding:28px 24px; margin-bottom:28px; background:#f5f7fa; border-left:5px solid #1f5f99; border-radius:8px; box-sizing:border-box;"&gt;
  &lt;h1 style="margin:0 0 14px 0; font-size:32px; line-height:1.25; color:#17202a;"&gt;
    Are Closed LLMs Losing Their Lead? Open-Weight Models Are Moving Closer to the Frontier
  &lt;/h1&gt;
  &lt;p style="margin:0; font-size:18px; color:#4b5563;"&gt;
    Inkling, Kimi K3, and Qwen3.8 changed the open-weight LLM landscape within just a few weeks, raising important questions about the future of the AI race.
  &lt;/p&gt;
&lt;/div&gt;

&lt;p&gt;
  As we entered the summer of 2026, the strongest players in the large language model race still looked very familiar: OpenAI, Anthropic, and other closed-model providers. Anthropic released Claude Fable 5 on June 9, 2026, while OpenAI made the GPT-5.6 family generally available on July 9 following a limited preview that began on June 26.
&lt;/p&gt;

&lt;p&gt;
  Both models were introduced with strong reasoning, coding, and agentic capabilities. Their advanced capabilities and potential security risks, particularly in cybersecurity, also brought them into broader discussions involving the U.S. government.
&lt;/p&gt;

&lt;p&gt;
  Then, within only a few weeks, three major developments changed the picture on the open-weight side.
&lt;/p&gt;

&lt;div style="margin:30px 0; padding:22px; border:1px solid #dbe3ea; border-radius:10px; background:#fafcfe; box-sizing:border-box;"&gt;
  &lt;div style="font-size:21px; margin-bottom:16px;"&gt;&#128467;️ &lt;strong&gt;A landscape that changed within weeks&lt;/strong&gt;&lt;/div&gt;

  &lt;p style="margin:10px 0;"&gt;
    &#127482;&#127480; &lt;strong&gt;July 15, 2026 — Inkling:&lt;/strong&gt; Thinking Machines Lab released Inkling, an open-weight model with 975 billion total parameters and 41 billion active parameters per token.
  &lt;/p&gt;

  &lt;p style="margin:10px 0;"&gt;
    &#127464;&#127475; &lt;strong&gt;July 16, 2026 — Kimi K3:&lt;/strong&gt; Moonshot AI announced Kimi K3, a 2.8-trillion-parameter model. Its full model weights were released on July 27.
  &lt;/p&gt;

  &lt;p style="margin:10px 0;"&gt;
    &#127464;&#127475; &lt;strong&gt;July 19, 2026 — Qwen3.8 preview:&lt;/strong&gt; Alibaba introduced Qwen3.8-Max-Preview in the 2.4-trillion-parameter class. Qwen3.8-Max was formally introduced on August 3, followed by the open-weight Qwen3.8-2.4T-A95B model on August 12.
  &lt;/p&gt;
&lt;/div&gt;

&lt;h2 style="margin-top:38px; font-size:25px; color:#17324d;"&gt;&#127482;&#127480; Inkling: A major open-weight move from the United States&lt;/h2&gt;

&lt;p&gt;
  The significance of Inkling is not simply that it has 975 billion parameters. The model uses a Mixture-of-Experts architecture with around 41 billion active parameters per token and can reason across text, images, and audio.
&lt;/p&gt;

&lt;p&gt;
  It also supports agentic coding, tool use, and context windows of up to one million tokens.
&lt;/p&gt;

&lt;p&gt;
  What makes Inkling particularly interesting, in my view, is something else. While Chinese companies had become far more visible in frontier open-weight model development, this time a powerful and genuinely open-weight model came from a new U.S.-based AI laboratory.
&lt;/p&gt;

&lt;h2 style="margin-top:38px; font-size:25px; color:#17324d;"&gt;&#127464;&#127475; Kimi K3: Open-weight models approach the 3-trillion-parameter class&lt;/h2&gt;

&lt;p&gt;
  Immediately after Inkling, Moonshot AI introduced Kimi K3, pushing open-weight model scale into a new range.
&lt;/p&gt;

&lt;p&gt;
  The model contains 2.8 trillion parameters. Its MoE architecture includes 896 experts, while only 16 are activated for each token. This allows the total model size to become extremely large without requiring every parameter to participate in every inference step.
&lt;/p&gt;

&lt;p&gt;
  Kimi K3 also attracted attention for more than its size. Around the time of its release, it ranked first on Frontend Code Arena, ahead of Claude Fable 5.
&lt;/p&gt;

&lt;p&gt;
  That does not mean Kimi K3 is now simply “better than GPT-5.6 or Fable 5.” Closed frontier models still hold an advantage in many broad evaluations. But the fact that an open-weight model can outperform the strongest closed models in selected coding and agentic tasks is a substantial change compared with the situation only a few years ago.
&lt;/p&gt;

&lt;h2 style="margin-top:38px; font-size:25px; color:#17324d;"&gt;&#127464;&#127475; Qwen3.8: Alibaba also moves into trillion-scale open models&lt;/h2&gt;

&lt;p&gt;
  Alibaba’s Qwen3.8 move showed that these developments were not isolated events.
&lt;/p&gt;

&lt;p&gt;
  Qwen3.8-Max, previewed in July, reached its formal 2.4-trillion-parameter release in August. The Qwen team then released the weights of Qwen3.8-2.4T-A95B as well.
&lt;/p&gt;

&lt;p&gt;
  As a result, within a short period, the open-weight ecosystem saw new models at 975 billion, 2.4 trillion, and 2.8 trillion parameters.
&lt;/p&gt;

&lt;div style="margin:34px 0; padding:24px; background:#17202a; color:#ffffff; border-radius:10px; box-sizing:border-box;"&gt;
  &lt;div style="font-size:23px; margin-bottom:12px;"&gt;⚖️ The real issue is not parameter count&lt;/div&gt;
  &lt;p style="margin:0; color:#eef2f5;"&gt;
    A 2.8-trillion-parameter model is not automatically better than a 500-billion-parameter model. Training data quality, architecture, post-training, reinforcement learning, tool use, inference infrastructure, and active parameter count can matter just as much as total model size. The real shift is that open-weight models are beginning to operate in the same capability range as frontier systems.
  &lt;/p&gt;
&lt;/div&gt;

&lt;h2 style="margin-top:38px; font-size:25px; color:#17324d;"&gt;&#128275; Are open-weight models still six or seven months behind?&lt;/h2&gt;

&lt;p&gt;
  For several years, there was a fairly reasonable rule of thumb: the strongest capabilities first appeared in closed systems from companies such as OpenAI, Anthropic, or Google, while comparable open models followed months later.
&lt;/p&gt;

&lt;p&gt;
  That assumption now needs to be reconsidered.
&lt;/p&gt;

&lt;p&gt;
  Models such as Kimi K3 can enter the same performance region as GPT-5.6 and Claude Fable 5 on selected tasks. The Qwen family continues to improve rapidly in coding and agentic workloads. Inkling shows that a U.S.-based open-weight frontier model is also possible.
&lt;/p&gt;

&lt;p&gt;
  Open models have not yet surpassed closed models across every dimension. But in many areas, the delay no longer looks like “wait for the next model generation.”
&lt;/p&gt;

&lt;p&gt;
  To me, this is the more important change.
&lt;/p&gt;

&lt;h2 style="margin-top:38px; font-size:25px; color:#17324d;"&gt;&#127464;&#127475; Are Chinese companies catching up with U.S. frontier labs?&lt;/h2&gt;

&lt;p&gt;
  Recent developments make this question increasingly difficult to dismiss.
&lt;/p&gt;

&lt;p&gt;
  Chinese companies are no longer producing only smaller or cheaper alternatives. With Moonshot AI and Alibaba, we are now seeing models at true frontier scale, and some of them are being released with their weights.
&lt;/p&gt;

&lt;p&gt;
  This is particularly significant because Chinese companies are operating under restrictions affecting access to some of the most advanced U.S. AI chips. Despite this, they continue to scale through work on model architecture, sparsity, attention mechanisms, quantization, and distributed inference.
&lt;/p&gt;

&lt;p&gt;
  The competition may therefore increasingly shift from “Which company has the best model?” to “Which national ecosystem can still develop frontier models?”
&lt;/p&gt;

&lt;h2 style="margin-top:38px; font-size:25px; color:#17324d;"&gt;&#128272; Will Chinese companies eventually close their strongest models?&lt;/h2&gt;

&lt;p&gt;
  This may become one of the most interesting questions for the future of the open-weight ecosystem.
&lt;/p&gt;

&lt;p&gt;
  Chinese model developers have benefited significantly from open-weight strategies. Their models have spread quickly across the world, developer communities have formed around them, inference platforms have added support, and the companies themselves have gained global visibility.
&lt;/p&gt;

&lt;p&gt;
  But would a company follow the same strategy if it genuinely believed it had the most capable model in the world?
&lt;/p&gt;

&lt;p&gt;
  Or would it become commercially more attractive to keep the strongest model behind an API while releasing the previous generation as open weights?
&lt;/p&gt;

&lt;p&gt;
  The next few model generations may provide an answer.
&lt;/p&gt;

&lt;h2 style="margin-top:38px; font-size:25px; color:#17324d;"&gt;&#127757; Where are the countries outside the U.S. and China?&lt;/h2&gt;

&lt;p&gt;
  This may be one of the least discussed aspects of the current AI race.
&lt;/p&gt;

&lt;p&gt;
  Two ecosystems now stand out very clearly in frontier model development: the United States and China.
&lt;/p&gt;

&lt;p&gt;
  Europe has strong research institutions and important AI companies. Japan and South Korea have advanced semiconductor industries. India has a very large software and engineering talent base. Yet at the trillion-parameter frontier, the competition is increasingly concentrated in the U.S. and China.
&lt;/p&gt;

&lt;p&gt;
  The reason is that this race cannot be run with good researchers alone. It requires tens of thousands of accelerators, data-center infrastructure, energy, capital, training data, distributed-systems engineering, and a very large research ecosystem.
&lt;/p&gt;

&lt;p&gt;
  One of the strategic questions for the coming years will therefore be whether other countries attempt to build their own frontier-model infrastructure or become increasingly dependent on models developed in the U.S. and China.
&lt;/p&gt;

&lt;div style="margin:36px 0; padding:26px; background:#f4f8fb; border:1px solid #cddce8; border-radius:10px; box-sizing:border-box;"&gt;
  &lt;div style="font-size:23px; margin-bottom:16px;"&gt;❓ Five questions for the next phase&lt;/div&gt;

  &lt;p&gt;&#128313; Can Chinese companies fully catch up with U.S. frontier-model companies technologically?&lt;/p&gt;

  &lt;p&gt;&#128313; Will the performance gap between open-weight and closed models disappear within the next few generations?&lt;/p&gt;

  &lt;p&gt;&#128313; To what extent will independent evaluations confirm the superiority suggested by selected benchmarks from companies such as OpenAI and Anthropic?&lt;/p&gt;

  &lt;p&gt;&#128313; Will Chinese companies continue their open-weight strategies even for their most capable models?&lt;/p&gt;

  &lt;p style="margin-bottom:0;"&gt;&#128313; Can countries outside the U.S. and China re-enter the frontier-model race?&lt;/p&gt;
&lt;/div&gt;

&lt;h2 style="margin-top:38px; font-size:25px; color:#17324d;"&gt;&#128640; Is a new era beginning for open-weight LLMs?&lt;/h2&gt;

&lt;p&gt;
  It is still too early to say that open-weight models have completely overtaken closed systems.
&lt;/p&gt;

&lt;p&gt;
  But something else can now be said with much more confidence: open-weight models are no longer sitting at the edge of the frontier AI race. They are moving closer to its center.
&lt;/p&gt;

&lt;p&gt;
  The rapid appearance of Inkling, Kimi K3, and Qwen3.8 is one of the clearest signs of this shift.
&lt;/p&gt;

&lt;p&gt;
  If the performance gap between open and closed models continues to narrow, the consequences will extend far beyond benchmark tables. It will affect how AI infrastructure is built, which model providers companies depend on, how countries think about AI sovereignty, and even the future of today’s API-centered business models.
&lt;/p&gt;

&lt;p&gt;
  &lt;strong&gt;What do you think? When will open-weight models catch closed models in overall performance? Or will the most capable models always remain closed?&lt;/strong&gt;
&lt;/p&gt;

&lt;hr style="border:0; border-top:1px solid #dde3e8; margin:36px 0 24px 0;"&gt;

&lt;div style="font-size:16px; color:#374151;"&gt;
  &lt;p style="margin-bottom:12px;"&gt;&lt;strong&gt;Murat Karakaya Academy&lt;/strong&gt;&lt;/p&gt;
  &lt;p style="line-height:1.9;"&gt;
    #MuratKarakayaAkademi #ArtificialIntelligence #LLM #OpenWeights #OpenSourceAI #KimiK3 #Qwen38 #Inkling #GPT56 #ClaudeFable5 #AgenticAI #GenerativeAI
  &lt;/p&gt;
&lt;/div&gt;
```

  &lt;/div&gt;
&lt;/div&gt;
</description><link>https://www.muratkarakaya.net/2026/09/are-closed-llms-losing-their-lead-open.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-1583298253394646223</guid><pubDate>Sat, 26 Sep 2026 20:13:40 +0000</pubDate><atom:updated>2026-09-26T23:13:40.966+03:00</atom:updated><category domain="http://www.blogger.com/atom/ns#">ABD</category><category domain="http://www.blogger.com/atom/ns#">Agentic AI</category><category domain="http://www.blogger.com/atom/ns#">Açık Ağırlık</category><category domain="http://www.blogger.com/atom/ns#">Claude Fable 5</category><category domain="http://www.blogger.com/atom/ns#">GPT-5.6</category><category domain="http://www.blogger.com/atom/ns#">Inkling</category><category domain="http://www.blogger.com/atom/ns#">Kimi K3</category><category domain="http://www.blogger.com/atom/ns#">LLM</category><category domain="http://www.blogger.com/atom/ns#">Murat Karakaya Akademi</category><category domain="http://www.blogger.com/atom/ns#">Open Weights</category><category domain="http://www.blogger.com/atom/ns#">Qwen3.8</category><category domain="http://www.blogger.com/atom/ns#">Yapay Zeka</category><category domain="http://www.blogger.com/atom/ns#">Çin</category><title>Kapalı LLM’lerin Saltanatı Sarsılıyor mu?</title><description>&lt;div style="max-width:100%; box-sizing:border-box; font-family:Arial,Helvetica,sans-serif; line-height:1.75; color:#222; background:#ffffff;"&gt;

  &lt;div style="max-width:920px; margin:0 auto; box-sizing:border-box;"&gt;

```
&lt;div style="padding:28px 24px; margin-bottom:28px; background:#f5f7fa; border-left:5px solid #1f5f99; border-radius:8px; box-sizing:border-box;"&gt;
  &lt;h1 style="margin:0 0 14px 0; font-size:32px; line-height:1.25; color:#17202a;"&gt;
    Kapalı LLM’lerin Saltanatı Sarsılıyor mu? Açık Ağırlıklı Modeller Frontier Seviyeye Yaklaşıyor
  &lt;/h1&gt;
  &lt;p style="margin:0; font-size:18px; color:#4b5563;"&gt;
    Inkling, Kimi K3 ve Qwen3.8 ile birkaç hafta içinde değişen açık ağırlıklı LLM dengesi, yapay zekâ yarışının geleceği hakkında önemli sorular ortaya çıkarıyor.
  &lt;/p&gt;
&lt;/div&gt;

&lt;p&gt;
  2026 yazına girerken büyük dil modeli yarışının en güçlü oyuncuları yine oldukça tanıdıktı: OpenAI, Anthropic ve diğer kapalı model sağlayıcıları. Anthropic 9 Haziran 2026'da Claude Fable 5'i, OpenAI ise 26 Haziran'daki sınırlı önizlemenin ardından 9 Temmuz'da GPT-5.6 ailesini genel kullanıma sundu.
&lt;/p&gt;

&lt;p&gt;
  Her iki model de yüksek akıl yürütme, kodlama ve ajan yetenekleriyle duyuruldu. Güvenlik riskleri ve özellikle gelişmiş siber yetenekleri nedeniyle ABD yönetimiyle yürütülen değerlendirme süreçleri de modellerin ne kadar güçlü hâle geldiği tartışmasını beraberinde getirdi.
&lt;/p&gt;

&lt;p&gt;
  Tam bu sırada, açık ağırlıklı LLM dünyasında birkaç hafta içinde oldukça önemli üç gelişme yaşandı.
&lt;/p&gt;

&lt;div style="margin:30px 0; padding:22px; border:1px solid #dbe3ea; border-radius:10px; background:#fafcfe; box-sizing:border-box;"&gt;
  &lt;div style="font-size:21px; margin-bottom:16px;"&gt;&#128467;️ &lt;strong&gt;Birkaç haftada değişen tablo&lt;/strong&gt;&lt;/div&gt;

  &lt;p style="margin:10px 0;"&gt;
    &#127482;&#127480; &lt;strong&gt;15 Temmuz 2026 — Inkling:&lt;/strong&gt; Thinking Machines Lab, 975 milyar toplam ve 41 milyar aktif parametreye sahip açık ağırlıklı Inkling modelini yayımladı.
  &lt;/p&gt;

  &lt;p style="margin:10px 0;"&gt;
    &#127464;&#127475; &lt;strong&gt;16 Temmuz 2026 — Kimi K3:&lt;/strong&gt; Moonshot AI, 2,8 trilyon parametreli Kimi K3'ü duyurdu. Model ağırlıkları 27 Temmuz'da kamuya açıldı.
  &lt;/p&gt;

  &lt;p style="margin:10px 0;"&gt;
    &#127464;&#127475; &lt;strong&gt;19 Temmuz 2026 — Qwen3.8 önizlemesi:&lt;/strong&gt; Alibaba, Qwen3.8-Max-Preview ile 2,4 trilyon parametre sınıfına çıktı. Qwen3.8-Max 3 Ağustos'ta resmen tanıtıldı ve 2,4 trilyon toplam, yaklaşık 95 milyar aktif parametreli açık ağırlıklı Qwen3.8-2.4T-A95B 12 Ağustos'ta yayımlandı.
  &lt;/p&gt;
&lt;/div&gt;

&lt;h2 style="margin-top:38px; font-size:25px; color:#17324d;"&gt;&#127482;&#127480; Inkling: ABD'den uzun süre sonra güçlü bir açık ağırlık hamlesi&lt;/h2&gt;

&lt;p&gt;
  Thinking Machines Lab tarafından yayımlanan Inkling'in önemi yalnızca 975 milyar parametreye sahip olması değil. Model, her token için yaklaşık 41 milyar aktif parametre kullanan Mixture-of-Experts (MoE) mimarisiyle çalışıyor ve metin, görüntü ve ses üzerinde doğal olarak akıl yürütebiliyor.
&lt;/p&gt;

&lt;p&gt;
  Ayrıca ajan tabanlı kodlama, araç kullanımı ve bir milyon token'a kadar bağlam desteği sunuyor.
&lt;/p&gt;

&lt;p&gt;
  Bence Inkling'i asıl ilginç hâle getiren nokta ise başka: Uzun süredir açık ağırlıklı frontier model geliştirme tarafında Çinli şirketler çok daha görünürken, bu kez güçlü ve gerçekten açık ağırlıklı bir model ABD merkezli yeni bir laboratuvardan geldi.
&lt;/p&gt;

&lt;h2 style="margin-top:38px; font-size:25px; color:#17324d;"&gt;&#127464;&#127475; Kimi K3: Açık model ölçeği 3 trilyon parametre sınırına dayandı&lt;/h2&gt;

&lt;p&gt;
  Inkling'in hemen ardından Moonshot AI tarafından açıklanan Kimi K3, açık ağırlıklı modellerin ölçeği açısından yeni bir sınırı temsil etti.
&lt;/p&gt;

&lt;p&gt;
  Model toplam 2,8 trilyon parametreye sahip. MoE yapısında bulunan 896 uzmandan yalnızca 16'sı her token için aktif hâle geliyor. Böylece toplam model büyüklüğü olağanüstü seviyelere çıkarken, bütün parametrelerin her token için çalıştırılması gerekmiyor.
&lt;/p&gt;

&lt;p&gt;
  Kimi K3 yalnızca büyüklüğüyle de dikkat çekmedi. Yayınlandığı dönemde Frontend Code Arena'da Claude Fable 5'i geçerek ilk sıraya yerleşmesi özellikle ilgi çekiciydi.
&lt;/p&gt;

&lt;p&gt;
  Bu sonuçtan “Kimi K3 artık GPT-5.6 veya Fable 5'ten daha iyi” sonucunu çıkarmak doğru olmaz. Genel değerlendirmelerde kapalı frontier modeller hâlâ birçok alanda avantajlarını koruyor. Ancak belirli kodlama ve ajan görevlerinde açık ağırlıklı bir modelin en güçlü kapalı modelleri geçebilmesi, birkaç yıl önceki tabloyla karşılaştırıldığında oldukça önemli.
&lt;/p&gt;

&lt;h2 style="margin-top:38px; font-size:25px; color:#17324d;"&gt;&#127464;&#127475; Qwen3.8: Alibaba da trilyon parametre ölçeğinde açık modele geçti&lt;/h2&gt;

&lt;p&gt;
  Alibaba'nın Qwen3.8 hamlesi bu gelişmelerin tek seferlik olmadığını gösterdi.
&lt;/p&gt;

&lt;p&gt;
  Temmuz ayında önizlenen Qwen3.8-Max, Ağustos ayında 2,4 trilyon parametreli resmî sürümüne ulaştı. Ardından Qwen ekibi Qwen3.8-2.4T-A95B modelinin ağırlıklarını da yayımladı.
&lt;/p&gt;

&lt;p&gt;
  Böylece kısa süre içerisinde açık ağırlıklı ekosistemde 975 milyar, 2,4 trilyon ve 2,8 trilyon parametre ölçeğinde üç yeni model ortaya çıktı.
&lt;/p&gt;

&lt;div style="margin:34px 0; padding:24px; background:#17202a; color:#ffffff; border-radius:10px; box-sizing:border-box;"&gt;
  &lt;div style="font-size:23px; margin-bottom:12px;"&gt;⚖️ Asıl mesele parametre sayısı değil&lt;/div&gt;
  &lt;p style="margin:0; color:#eef2f5;"&gt;
    2,8 trilyon parametreli bir modelin 500 milyar parametreli bir modelden otomatik olarak daha iyi olduğunu söyleyemeyiz. Eğitim verisinin kalitesi, model mimarisi, post-training süreci, reinforcement learning, araç kullanımı, inference altyapısı ve aktif parametre sayısı en az toplam model büyüklüğü kadar önemli. Buradaki asıl değişim, açık ağırlıklı modellerin hem ölçek hem de yetenek açısından frontier modellerin bulunduğu bölgeye girmeye başlaması.
  &lt;/p&gt;
&lt;/div&gt;

&lt;h2 style="margin-top:38px; font-size:25px; color:#17324d;"&gt;&#128275; Açık ağırlıklı modeller hâlâ 6–7 ay geriden mi geliyor?&lt;/h2&gt;

&lt;p&gt;
  Birkaç yıl boyunca oldukça makul bir genelleme vardı: En güçlü yetenek önce OpenAI, Anthropic veya Google gibi kapalı model sağlayıcılarında ortaya çıkıyor; benzer seviyedeki açık modeller ise aylar sonra geliyordu.
&lt;/p&gt;

&lt;p&gt;
  Bugün bu varsayımın yeniden değerlendirilmesi gerekiyor.
&lt;/p&gt;

&lt;p&gt;
  Kimi K3 gibi modeller bazı görevlerde GPT-5.6 ve Claude Fable 5 ile aynı performans bölgesine girebiliyor. Qwen serisi kodlama ve ajan görevlerinde sürekli ilerliyor. Inkling ise ABD'den gelen açık ağırlıklı bir frontier modelin de mümkün olduğunu gösteriyor.
&lt;/p&gt;

&lt;p&gt;
  Açık modeller henüz kapalı modelleri bütün boyutlarda geçmiş değil. Ancak aradaki gecikme birçok alanda artık “bir sonraki nesli beklemek” seviyesinde görünmüyor.
&lt;/p&gt;

&lt;p&gt;
  Bana göre tartışılması gereken asıl değişim de burada.
&lt;/p&gt;

&lt;h2 style="margin-top:38px; font-size:25px; color:#17324d;"&gt;&#127464;&#127475; Çinli şirketler Amerikan şirketlerini yakalıyor mu?&lt;/h2&gt;

&lt;p&gt;
  Son gelişmeler bu soruyu giderek daha ciddi hâle getiriyor.
&lt;/p&gt;

&lt;p&gt;
  Çinli şirketler yalnızca daha ucuz veya daha küçük modeller üretmiyor. Moonshot AI ve Alibaba örneklerinde artık doğrudan frontier ölçeğinde modeller görüyoruz. Üstelik bunların bir bölümü model ağırlıklarıyla birlikte yayımlanıyor.
&lt;/p&gt;

&lt;p&gt;
  Bu durum özellikle önemli çünkü Çinli şirketler aynı zamanda ABD'nin en gelişmiş yapay zekâ çiplerine yönelik ihracat sınırlamaları altında çalışıyor. Buna rağmen model mimarisi, sparsity, attention mekanizmaları, quantization ve dağıtık inference gibi alanlardaki mühendislik çalışmalarıyla ölçek büyütmeye devam ediyorlar.
&lt;/p&gt;

&lt;p&gt;
  Dolayısıyla önümüzdeki dönemde yarış yalnızca “hangi şirketin en iyi modeli var?” sorusundan ibaret olmayabilir. “Hangi ülkenin yapay zekâ ekosistemi frontier model geliştirebiliyor?” sorusu da giderek daha önemli hâle geliyor.
&lt;/p&gt;

&lt;h2 style="margin-top:38px; font-size:25px; color:#17324d;"&gt;&#128272; Çinli şirketler güçlü hâle geldikçe modellerini kapatır mı?&lt;/h2&gt;

&lt;p&gt;
  Bence açık ağırlıklı ekosistemin geleceği açısından en ilginç sorulardan biri bu.
&lt;/p&gt;

&lt;p&gt;
  Çinli model geliştiricileri bugüne kadar açık ağırlık stratejisinden önemli faydalar elde etti. Modeller hızla dünya çapında kullanılmaya başladı, geliştirici toplulukları oluştu, inference sistemleri bu modellere destek ekledi ve şirketlerin küresel görünürlüğü arttı.
&lt;/p&gt;

&lt;p&gt;
  Ancak bir şirket gerçekten dünyanın en güçlü modeline sahip olduğuna inandığında aynı stratejiyi sürdürür mü?
&lt;/p&gt;

&lt;p&gt;
  Yoksa en güçlü modeli API arkasında tutup, bir önceki nesli açık ağırlıklı olarak yayımlamak daha avantajlı hâle mi gelir?
&lt;/p&gt;

&lt;p&gt;
  Önümüzdeki birkaç model nesli bu sorunun cevabı açısından oldukça belirleyici olacak.
&lt;/p&gt;

&lt;h2 style="margin-top:38px; font-size:25px; color:#17324d;"&gt;&#127757; Peki ABD ve Çin dışındaki ülkeler nerede?&lt;/h2&gt;

&lt;p&gt;
  Bu yarışın belki de en az tartışılan tarafı burada.
&lt;/p&gt;

&lt;p&gt;
  Bugün frontier model geliştirme yarışında iki ekosistem çok belirgin biçimde öne çıkıyor: ABD ve Çin.
&lt;/p&gt;

&lt;p&gt;
  Avrupa'nın güçlü araştırma kurumları ve önemli yapay zekâ şirketleri var. Japonya ve Güney Kore ileri yarı iletken teknolojilerine sahip. Hindistan büyük bir yazılım ve mühendislik insan kaynağı barındırıyor. Buna rağmen trilyon parametre ölçeğindeki frontier modellerde yarış giderek ABD ve Çin arasında yoğunlaşıyor.
&lt;/p&gt;

&lt;p&gt;
  Çünkü bu yarış yalnızca iyi araştırmacılarla yürümüyor. On binlerce hızlandırıcı, veri merkezi altyapısı, enerji, sermaye, eğitim verisi, dağıtık sistem mühendisliği ve çok büyük bir araştırma ekosistemi gerekiyor.
&lt;/p&gt;

&lt;p&gt;
  Dolayısıyla önümüzdeki dönemin stratejik sorularından biri de şu olacak: Diğer ülkeler kendi frontier model altyapılarını kurmaya çalışacak mı, yoksa ABD ve Çin tarafından geliştirilen modellere bağımlı mı kalacak?
&lt;/p&gt;

&lt;div style="margin:36px 0; padding:26px; background:#f4f8fb; border:1px solid #cddce8; border-radius:10px; box-sizing:border-box;"&gt;
  &lt;div style="font-size:23px; margin-bottom:16px;"&gt;❓ Önümüzdeki dönem için beş soru&lt;/div&gt;

  &lt;p&gt;&#128313; Çinli şirketler Amerikan frontier model şirketlerini teknolojik olarak tamamen yakalayabilecek mi?&lt;/p&gt;

  &lt;p&gt;&#128313; Açık ağırlıklı ve kapalı modeller arasındaki performans farkı birkaç nesil içinde tamamen ortadan kalkacak mı?&lt;/p&gt;

  &lt;p&gt;&#128313; OpenAI ve Anthropic gibi şirketlerin seçilmiş benchmarklarla oluşturduğu üstünlük algısını bağımsız değerlendirmeler ne ölçüde doğrulayacak?&lt;/p&gt;

  &lt;p&gt;&#128313; Çinli şirketler en güçlü modellerinde de açık ağırlık politikasını sürdürecek mi?&lt;/p&gt;

  &lt;p style="margin-bottom:0;"&gt;&#128313; ABD ve Çin dışındaki ülkeler frontier model yarışına yeniden katılabilecek mi?&lt;/p&gt;
&lt;/div&gt;

&lt;h2 style="margin-top:38px; font-size:25px; color:#17324d;"&gt;&#128640; Açık ağırlıklı LLM dünyasında yeni bir dönem mi başlıyor?&lt;/h2&gt;

&lt;p&gt;
  Açık ağırlıklı modellerin kapalı modelleri tamamen geçtiğini söylemek için henüz erken.
&lt;/p&gt;

&lt;p&gt;
  Fakat artık başka bir şeyi söylemek mümkün: Açık ağırlıklı modeller frontier yapay zekâ yarışının kenarında değil, giderek daha fazla merkezinde yer alıyor.
&lt;/p&gt;

&lt;p&gt;
  Inkling, Kimi K3 ve Qwen3.8'in kısa aralıklarla ortaya çıkması bu değişimin güçlü işaretlerinden biri.
&lt;/p&gt;

&lt;p&gt;
  Eğer açık modeller ile kapalı modeller arasındaki performans farkı gerçekten kapanmaya devam ederse, bunun etkisi yalnızca benchmark tablolarında görülmeyecek. Yapay zekâ altyapılarının nasıl kurulacağı, şirketlerin hangi modellere bağımlı olacağı, ülkelerin yapay zekâ egemenliği ve hatta bugünkü API merkezli iş modellerinin geleceği de bundan etkilenecek.
&lt;/p&gt;

&lt;p&gt;
  &lt;strong&gt;Sizce açık ağırlıklı modeller kapalı modelleri genel performansta ne zaman yakalayacak? Yoksa en güçlü modeller her zaman kapalı sistemlerde mi kalacak?&lt;/strong&gt;
&lt;/p&gt;

&lt;hr style="border:0; border-top:1px solid #dde3e8; margin:36px 0 24px 0;"&gt;

&lt;div style="font-size:16px; color:#374151;"&gt;
  &lt;p style="margin-bottom:12px;"&gt;&lt;strong&gt;Murat Karakaya Akademi&lt;/strong&gt;&lt;/p&gt;
  &lt;p style="line-height:1.9;"&gt;
    #MuratKarakayaAkademi #YapayZeka #LLM #OpenWeights #AcikAgirlik #OpenSourceAI #KimiK3 #Qwen38 #Inkling #GPT56 #ClaudeFable5 #AgenticAI #GenerativeAI #ArtificialIntelligence
  &lt;/p&gt;
&lt;/div&gt;
```

  &lt;/div&gt;
&lt;/div&gt;
</description><link>https://www.muratkarakaya.net/2026/09/kapal-llmlerin-saltanat-sarslyor-mu.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-3569168409338258911</guid><pubDate>Sat, 26 Sep 2026 20:07:53 +0000</pubDate><atom:updated>2026-09-26T23:07:53.185+03:00</atom:updated><category domain="http://www.blogger.com/atom/ns#">Agentic AI</category><category domain="http://www.blogger.com/atom/ns#">AI Agents</category><category domain="http://www.blogger.com/atom/ns#">AI Engineering</category><category domain="http://www.blogger.com/atom/ns#">Artificial Intelligence</category><category domain="http://www.blogger.com/atom/ns#">Deep Learning</category><category domain="http://www.blogger.com/atom/ns#">Future of AI</category><category domain="http://www.blogger.com/atom/ns#">Generative AI</category><category domain="http://www.blogger.com/atom/ns#">LLM</category><category domain="http://www.blogger.com/atom/ns#">Local AI</category><category domain="http://www.blogger.com/atom/ns#">MCP</category><category domain="http://www.blogger.com/atom/ns#">RAG</category><category domain="http://www.blogger.com/atom/ns#">Reasoning Models</category><category domain="http://www.blogger.com/atom/ns#">Transformers</category><title>ARTIFICIAL INTELLIGENCE • LLM • AGENTIC AI • 2027</title><description>&lt;div style="width:100%; box-sizing:border-box; font-family:Arial,Helvetica,sans-serif; color:#222; font-size:17px; line-height:1.75; overflow-wrap:break-word;"&gt;

&lt;!-- INTRO BOX --&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f3f6fb; border:1px solid #dfe5ee; border-radius:14px; padding:28px; margin:0 0 35px 0;"&gt;

  &lt;div style="font-size:13px; color:#687386; letter-spacing:1px; margin-bottom:12px;"&gt;
    ARTIFICIAL INTELLIGENCE • LLM • AGENTIC AI • 2027
  &lt;/div&gt;

  &lt;h2 style="font-size:32px; line-height:1.25; color:#18243e; margin:0 0 18px 0;"&gt;
    Witnessing the AI Revolution: From Deep Learning to Long-Running AI Agents
  &lt;/h2&gt;

  &lt;p style="font-size:19px; color:#46536a; margin:0;"&gt;
    Over the past 15 years, artificial intelligence has not merely produced better models.
    We have moved from systems that recognize images to models that converse with us,
    from conversational models to agents that use tools, and from there to AI systems
    capable of working on a problem continuously for hours or even days.
    So what will be the next major breakthrough in 2027?
  &lt;/p&gt;

&lt;/div&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:35px;"&gt;
A Question That Emerged from a Conversation
&lt;/h2&gt;

&lt;p&gt;
Today, while talking with a friend who is also deeply interested in artificial intelligence,
our conversation naturally turned to the extraordinary pace of development in AI over the past few years.
&lt;/p&gt;

&lt;p&gt;
After a while, we started recalling the technologies we had been discussing over the years:
Deep Learning, RNNs, LSTMs, Transformers, large language models, multimodal models,
reasoning models, RAG, function calling, agents, MCP, computer use, coding agents,
and most recently, AI agents capable of working independently for extended periods of time.
&lt;/p&gt;

&lt;p&gt;
When you put all of these developments side by side, the speed of change becomes much more visible.
&lt;/p&gt;

&lt;p&gt;
In fact, we have not simply developed more capable AI models over the past few years.
The way we use artificial intelligence itself has changed several times.
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f5f7ff; border-left:5px solid #3f5bd9; padding:20px 22px; margin:28px 0;"&gt;
  &lt;span style="font-size:20px; color:#26375f;"&gt;
    Artificial intelligence is evolving from a system that recognizes patterns
    into a system that talks to us, reasons about problems, uses tools,
    and increasingly performs complete tasks on its own.
  &lt;/span&gt;
&lt;/div&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#129504; 1. Deep Learning: The Beginning of the Major Breakthrough
&lt;/h2&gt;

&lt;p&gt;
Artificial neural networks were certainly not new.
However, for many years they could not demonstrate their full potential because of limitations
in data availability, computing power, and training techniques.
&lt;/p&gt;

&lt;p&gt;
In 2012, AlexNet’s success in the ImageNet competition demonstrated how powerful deep neural networks
trained on GPUs could become.
&lt;/p&gt;

&lt;p&gt;
The breakthrough that started in computer vision quickly spread to speech recognition,
natural language processing, and many other machine learning problems.
&lt;/p&gt;

&lt;p&gt;
From today’s perspective, this may seem almost ordinary, but one of the key foundations
of the AI revolution we are experiencing today was established here:
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#fff8e8; border:1px solid #ead9a8; padding:18px 22px; margin:24px 0;"&gt;
  Larger neural networks + more data + more computing power.
&lt;/div&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128257; 2. RNNs, LSTMs, and Sequence-to-Sequence Models
&lt;/h2&gt;

&lt;p&gt;
While CNN architectures achieved remarkable success in image processing,
natural language presented a different challenge: language is sequential.
&lt;/p&gt;

&lt;p&gt;
RNNs, and particularly LSTMs, made it possible to retain information from previous steps
and process sentences, conversations, and time-series data more effectively.
&lt;/p&gt;

&lt;p&gt;
Then came sequence-to-sequence, or seq2seq, architectures.
They enabled a model to receive one sequence and generate another sequence as output.
Machine translation became one of the most important applications of this idea.
&lt;/p&gt;

&lt;p&gt;
Today, we consider it completely natural to give a model a piece of text and receive another piece
of generated text in return. Yet encoder-decoder and seq2seq architectures were important milestones
on the road toward modern generative AI.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128065;️ 3. Attention: Teaching the Model Where to Look
&lt;/h2&gt;

&lt;p&gt;
There is an important chronological detail here:
the attention mechanism did not begin with the Transformer.
&lt;/p&gt;

&lt;p&gt;
Around 2014, particularly in neural machine translation research,
models began to learn how to focus more heavily on relevant parts of the source sentence
while generating each part of the output.
&lt;/p&gt;

&lt;p&gt;
This idea was extremely important because a model no longer had to compress the entire input
into a single fixed representation.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
⚡ 4. 2017: Attention Is All You Need
&lt;/h2&gt;

&lt;p&gt;
In 2017, the publication of &lt;em&gt;Attention Is All You Need&lt;/em&gt; became one of the most important
turning points in the history of artificial intelligence.
&lt;/p&gt;

&lt;p&gt;
The Transformer architecture greatly reduced the dependence on the sequential processing
used by RNNs and placed self-attention at the center of the architecture.
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f4faf8; border-left:5px solid #15956c; padding:20px 22px; margin:25px 0;"&gt;
  With self-attention, relationships between different words in a text could be modeled directly,
  while computations could be parallelized far more efficiently.
&lt;/div&gt;

&lt;p&gt;
Today, the architectural roots of major model families such as GPT, Claude, Gemini,
Llama, Qwen, and Mistral can largely be traced back to this transformation.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128218; 5. Pre-Training and the Foundation Model Paradigm
&lt;/h2&gt;

&lt;p&gt;
After the Transformer, another major paradigm gained momentum.
&lt;/p&gt;

&lt;p&gt;
Instead of training a separate model from scratch for every task,
researchers began training large general-purpose models on massive datasets
and then adapting them to different downstream tasks.
&lt;/p&gt;

&lt;p&gt;
Models such as BERT demonstrated how powerful this approach could be in natural language processing.
&lt;/p&gt;

&lt;p&gt;
The discussion gradually moved beyond the question of
“Which architecture performs better?”
toward a broader question:
“Can we train general-purpose models capable of performing many different tasks?”
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128200; 6. Scaling and GPT-3
&lt;/h2&gt;

&lt;p&gt;
Around 2020, the relationship between model size, training data, computational resources,
and performance became far more visible.
&lt;/p&gt;

&lt;p&gt;
Scaling law research suggested that larger models did not simply perform the same tasks slightly better.
In some cases, new capabilities appeared as scale increased.
&lt;/p&gt;

&lt;p&gt;
GPT-3 made another important phenomenon highly visible:
&lt;/p&gt;

&lt;p&gt;
A model could perform many different tasks without being retrained for each one,
simply by receiving instructions or a few examples in the prompt.
&lt;/p&gt;

&lt;p&gt;
Zero-shot and few-shot prompting became increasingly important.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128196; 7. RAG: Moving Beyond the Model’s Parameters
&lt;/h2&gt;

&lt;p&gt;
One of the major limitations of large language models was their inability to reliably access
current, organization-specific, or previously unseen information.
&lt;/p&gt;

&lt;p&gt;
Retrieval-Augmented Generation, or RAG, approached this problem from a different direction.
&lt;/p&gt;

&lt;p&gt;
Instead of trying to encode every piece of information inside the model’s parameters,
relevant information could be retrieved from an external knowledge source
and added to the model’s context.
&lt;/p&gt;

&lt;p&gt;
Over time, RAG evolved into more sophisticated architectures involving embeddings,
vector databases, semantic search, hybrid search, reranking, query rewriting,
and knowledge bases.
&lt;/p&gt;

&lt;p&gt;
This approach remains extremely important today, particularly in enterprise AI applications.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#127912; 8. Diffusion Models and Generative Visual AI
&lt;/h2&gt;

&lt;p&gt;
Meanwhile, the AI revolution was not limited to text.
&lt;/p&gt;

&lt;p&gt;
With the rise of diffusion models, text-to-image generation rapidly became mainstream.
&lt;/p&gt;

&lt;p&gt;
This was followed by major progress in image editing, style transfer,
image-to-video generation, text-to-video generation, speech synthesis,
and AI-generated music.
&lt;/p&gt;

&lt;p&gt;
The concept of Generative AI was no longer limited to models that generated text.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128064; 9. Multimodal Models
&lt;/h2&gt;

&lt;p&gt;
The next important step was bringing different types of data together within the same model.
&lt;/p&gt;

&lt;p&gt;
The model was no longer limited to understanding the text we typed.
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:8px;"&gt;&#128247; It can understand photographs.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:8px;"&gt;&#128202; It can interpret charts.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:8px;"&gt;&#128209; It can read documents.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:8px;"&gt;&#127897;️ It can understand audio.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:8px;"&gt;&#128421;️ It can see a computer screen.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:20px;"&gt;&#127909; It can analyze video.&lt;/div&gt;

&lt;p&gt;
Because the real world is not composed only of text,
multimodality became an essential step toward more general-purpose artificial intelligence.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128172; 10. 2022: ChatGPT Brings AI to the Masses
&lt;/h2&gt;

&lt;p&gt;
The release of ChatGPT on November 30, 2022 was, in my view,
not only a technological milestone but also a major social turning point.
&lt;/p&gt;

&lt;p&gt;
For the first time, large language models entered the daily lives of ordinary users at massive scale.
&lt;/p&gt;

&lt;p&gt;
Teachers, students, doctors, lawyers, engineers, managers, artists,
software developers, and entrepreneurs all began interacting with the same technology.
&lt;/p&gt;

&lt;p&gt;
Instruction tuning and alignment methods based on human feedback made interaction
with these systems far more natural from the user’s perspective.
&lt;/p&gt;

&lt;p&gt;
At the same time, a new term entered everyday vocabulary:
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#fff8e8; padding:18px 22px; margin:24px 0;"&gt;
  Prompt Engineering.
&lt;/div&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128295; 11. Function Calling and Tool Use
&lt;/h2&gt;

&lt;p&gt;
Large language models could generate highly impressive text,
but their ability to interact with real-world systems was limited.
&lt;/p&gt;

&lt;p&gt;
Function calling and tool use began to change that.
&lt;/p&gt;

&lt;p&gt;
A model could now call an API, access a database,
use a calculator, perform a web search,
or invoke another software component.
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f4faf8; padding:20px 22px; margin:25px 0;"&gt;
  &lt;div style="margin-bottom:10px;"&gt;Old approach: “Answer the question.”&lt;/div&gt;
  &lt;div&gt;New approach: “Use tools when necessary to solve the problem.”&lt;/div&gt;
&lt;/div&gt;

&lt;p&gt;
This seemingly small change later became one of the foundations of agentic systems.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#129504; 12. Reasoning Models
&lt;/h2&gt;

&lt;p&gt;
In 2024, the rise of reasoning models introduced another important shift.
&lt;/p&gt;

&lt;p&gt;
The focus was no longer only on adding more parameters to the model.
Models could also spend more computation on a difficult problem before producing an answer.
&lt;/p&gt;

&lt;p&gt;
The concept of test-time compute became increasingly important.
&lt;/p&gt;

&lt;p&gt;
Especially in mathematics, science, coding, and complex problem solving,
the idea of the “larger model” was joined by another idea:
the “model that spends more time reasoning about the problem.”
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128187; 13. From Chatbot to Coding Agent
&lt;/h2&gt;

&lt;p&gt;
In my view, one of the most important recent transformations has occurred in software development.
&lt;/p&gt;

&lt;p&gt;
At first, we asked models to generate small pieces of code.
&lt;/p&gt;

&lt;p&gt;
Then came copilot-style systems embedded directly inside IDEs.
&lt;/p&gt;

&lt;p&gt;
Later, AI systems began reading entire repositories,
creating files, running terminal commands, debugging errors,
writing tests, and refactoring code.
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:8px;"&gt;&#128172; A system that talks about code&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:8px;"&gt;➡️ A system that suggests code&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:8px;"&gt;➡️ A system that writes code&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:8px;"&gt;➡️ A system that works across a repository&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:20px;"&gt;➡️ An agent that takes responsibility for a software development task&lt;/div&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128268; 14. MCP: Standard Connections Between AI and the Outside World
&lt;/h2&gt;

&lt;p&gt;
As agent systems became more common, another problem became increasingly visible.
&lt;/p&gt;

&lt;p&gt;
Every AI application had to be integrated separately with GitHub,
file systems, databases, enterprise applications,
and many other external services.
&lt;/p&gt;

&lt;p&gt;
The Model Context Protocol, or MCP,
became an important development for establishing more standardized connections
between AI applications, tools, and data sources.
&lt;/p&gt;

&lt;p&gt;
I do not think MCP should be viewed merely as a technical protocol.
For an agent ecosystem to grow, models need standardized and manageable ways
to connect to the outside world.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#129513; 15. Agent Skills
&lt;/h2&gt;

&lt;p&gt;
The next step was to stop explaining the same procedures to an agent again and again,
and instead define reusable skills that describe how particular tasks should be performed.
&lt;/p&gt;

&lt;p&gt;
A skill can, for example, teach an agent:
&lt;/p&gt;

&lt;p&gt;&#128202; How to analyze an Excel file,&lt;/p&gt;
&lt;p&gt;&#128196; How to prepare a professional report,&lt;/p&gt;
&lt;p&gt;&#129514; How to execute a testing procedure,&lt;/p&gt;
&lt;p&gt;&#127970; How to follow an organization-specific workflow,&lt;/p&gt;
&lt;p&gt;&#128187; How to perform a particular software development task.&lt;/p&gt;

&lt;p&gt;
The agent can then use that skill whenever it becomes relevant.
&lt;/p&gt;

&lt;p&gt;
This seems especially important in the transition
from general-purpose AI toward specialized digital workers.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128421;️ 16. Computer Use and Browser Use
&lt;/h2&gt;

&lt;p&gt;
Agents then began doing more than simply calling APIs.
&lt;/p&gt;

&lt;p&gt;
They started seeing screens, moving the mouse,
clicking buttons, navigating websites,
and interacting with the same graphical user interfaces that humans use.
&lt;/p&gt;

&lt;p&gt;
I believe this is a critical development.
&lt;/p&gt;

&lt;p&gt;
Most software in the world does not have an API specifically designed for AI.
But almost every application has a user interface designed for humans.
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f5f7ff; border-left:5px solid #3f5bd9; padding:20px 22px; margin:25px 0;"&gt;
  With computer use, the user interface itself is effectively becoming
  another integration layer that AI systems can operate.
&lt;/div&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#127757; 17. Open Models and Local AI
&lt;/h2&gt;

&lt;p&gt;
While all of these developments were taking place,
another transformation was happening in parallel.
&lt;/p&gt;

&lt;p&gt;
Open-weight models rapidly narrowed the gap with closed models
across many practical use cases.
&lt;/p&gt;

&lt;p&gt;
Model families such as Llama, Mistral, Qwen, and DeepSeek
helped build a powerful open ecosystem.
&lt;/p&gt;

&lt;p&gt;
Quantization, more efficient inference engines,
and advances in hardware made it possible to run models locally
that previously would have required large data centers.
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#fff5f2; border-left:5px solid #d76545; padding:20px 22px; margin:25px 0;"&gt;
  Artificial intelligence no longer has to exist only as a cloud service accessed over the internet.
&lt;/div&gt;

&lt;p&gt;
Local AI systems that keep organizational or personal data on-premise
are becoming increasingly realistic alternatives.
&lt;/p&gt;

&lt;p&gt;
The emergence of personal agent software that can run continuously
on a local computer is another important part of this transformation.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#129302; 18. 2026: The Year of AI Agents
&lt;/h2&gt;

&lt;p&gt;
For some time, many people had predicted that 2026 would become “the year of AI agents.”
&lt;/p&gt;

&lt;p&gt;
I think that prediction has largely come true.
&lt;/p&gt;

&lt;p&gt;
However, the meaning of the word “agent” has also changed.
&lt;/p&gt;

&lt;p&gt;
Early agent architectures looked roughly like this:
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:20px 22px; margin:25px 0; text-align:center;"&gt;
  LLM → call a tool → observe the result → call another tool if necessary → answer
&lt;/div&gt;

&lt;p&gt;
Today, we are talking about far more comprehensive systems.
&lt;/p&gt;

&lt;p&gt;&#129517; They can plan.&lt;/p&gt;
&lt;p&gt;&#129520; They can choose the right tool.&lt;/p&gt;
&lt;p&gt;&#128450;️ They can manage files and context.&lt;/p&gt;
&lt;p&gt;&#129504; They can preserve state across a task.&lt;/p&gt;
&lt;p&gt;&#129309; They can delegate work to other agents.&lt;/p&gt;
&lt;p&gt;&#128187; They can run code and terminal commands.&lt;/p&gt;
&lt;p&gt;&#127760; They can perform actions on the web.&lt;/p&gt;
&lt;p&gt;&#128269; They can evaluate their own outputs.&lt;/p&gt;
&lt;p&gt;&#128737;️ They can operate within permission boundaries.&lt;/p&gt;
&lt;p&gt;&#128100; They can request human approval when necessary.&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#127959;️ 19. The Next Step: Agent Harnesses
&lt;/h2&gt;

&lt;p&gt;
One concept I find particularly important in recent developments is the agent harness.
&lt;/p&gt;

&lt;p&gt;
As we assign agents longer and more complicated tasks,
the intelligence of the underlying model alone is no longer enough.
&lt;/p&gt;

&lt;p&gt;
A model can perform extremely well during the first few steps
and then drift in the wrong direction.
&lt;/p&gt;

&lt;p&gt;
It can incorrectly assume that the task is complete.&lt;/p&gt;

&lt;p&gt;
It can forget previous decisions.&lt;/p&gt;

&lt;p&gt;
It can reach a conclusion without testing its work.&lt;/p&gt;

&lt;p&gt;
Its context can gradually degrade during long-running tasks.&lt;/p&gt;

&lt;p&gt;
This is why we increasingly need an execution environment around the model
that manages and supervises its work.
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f5f7ff; padding:14px 18px; margin-bottom:8px;"&gt;&#128221; It preserves the plan.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f5f7ff; padding:14px 18px; margin-bottom:8px;"&gt;&#128260; It triggers replanning when necessary.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f5f7ff; padding:14px 18px; margin-bottom:8px;"&gt;&#129514; It tests intermediate and final results.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f5f7ff; padding:14px 18px; margin-bottom:8px;"&gt;&#128269; It checks for missing work.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f5f7ff; padding:14px 18px; margin-bottom:8px;"&gt;&#128218; It manages context.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f5f7ff; padding:14px 18px; margin-bottom:8px;"&gt;&#128200; It monitors progress.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f5f7ff; padding:14px 18px; margin-bottom:8px;"&gt;✅ It can independently verify the output.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f5f7ff; padding:14px 18px; margin-bottom:22px;"&gt;&#128721; It helps determine when the agent should stop.&lt;/div&gt;

&lt;p&gt;
This moves us from chatbots that work for a few minutes
toward AI systems that can work toward a goal for hours,
or potentially much longer when supported by the right infrastructure.
&lt;/p&gt;


&lt;div style="width:100%; box-sizing:border-box; background:#18243e; color:#fff; padding:28px; margin:38px 0; border-radius:12px;"&gt;

  &lt;div style="font-size:13px; letter-spacing:1px; color:#bac4d8; margin-bottom:12px;"&gt;
    I THINK THIS IS THE REAL BREAKTHROUGH
  &lt;/div&gt;

  &lt;div style="font-size:25px; line-height:1.4; margin-bottom:14px;"&gt;
    The competition is no longer simply about “who has the best LLM?”
  &lt;/div&gt;

  &lt;div style="font-size:17px; line-height:1.7; color:#e1e6ef;"&gt;
    The real competition is about how effectively a model can operate
    with the right data, the right tools, the right memory,
    the right skills, the right security mechanisms,
    and the right verification environment.
  &lt;/div&gt;

&lt;/div&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
How Has the Short History of Modern AI Actually Changed?
&lt;/h2&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:15px 18px; margin-bottom:8px;"&gt;
&#129504; 2010s → The model recognizes patterns.
&lt;/div&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:15px 18px; margin-bottom:8px;"&gt;
&#128172; 2022 → The model talks to us.
&lt;/div&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:15px 18px; margin-bottom:8px;"&gt;
&#128064; 2023–2024 → The model sees and hears.
&lt;/div&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:15px 18px; margin-bottom:8px;"&gt;
&#129504; 2024 → The model spends more time reasoning about the problem.
&lt;/div&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:15px 18px; margin-bottom:8px;"&gt;
&#128295; 2023–2025 → The model uses tools.
&lt;/div&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:15px 18px; margin-bottom:8px;"&gt;
&#128421;️ 2024–2025 → The model uses computers and browsers.
&lt;/div&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:15px 18px; margin-bottom:8px;"&gt;
&#129302; 2025–2026 → The model becomes an agent.
&lt;/div&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:15px 18px; margin-bottom:24px;"&gt;
&#127959;️ 2026 → The agent is placed inside an environment designed for long-running work.
&lt;/div&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:45px;"&gt;
&#128302; So What Will Define AI in 2027?
&lt;/h2&gt;

&lt;p&gt;
I do not think the next major leap will necessarily be just another,
slightly more powerful conversational model.
&lt;/p&gt;

&lt;p&gt;
One of my strongest candidates is the &lt;em&gt;persistent agent&lt;/em&gt;:
a personal or enterprise AI system that can work continuously
while preserving its state and context over time.
&lt;/p&gt;

&lt;p&gt;
In today’s chatbot paradigm, we go to the AI.
&lt;/p&gt;

&lt;p&gt;
We open a window, ask a question, receive an answer,
and then continue working ourselves.
&lt;/p&gt;

&lt;p&gt;
The next stage may be very different:
AI could become a persistent part of our working environment.
&lt;/p&gt;

&lt;p&gt;&#129504; It could recall our previous work when relevant.&lt;/p&gt;
&lt;p&gt;&#128231; It could work with our email and messages.&lt;/p&gt;
&lt;p&gt;&#128197; It could follow calendars and tasks.&lt;/p&gt;
&lt;p&gt;&#128193; It could understand our files and projects.&lt;/p&gt;
&lt;p&gt;&#127760; It could conduct research when necessary.&lt;/p&gt;
&lt;p&gt;&#128187; It could perform actions on our computers.&lt;/p&gt;
&lt;p&gt;&#129513; It could use the skills it needs.&lt;/p&gt;
&lt;p&gt;&#128268; It could connect to enterprise systems.&lt;/p&gt;
&lt;p&gt;&#129309; It could collaborate with specialized agents.&lt;/p&gt;
&lt;p&gt;&#128737;️ It could understand its permission boundaries.&lt;/p&gt;
&lt;p&gt;&#128100; It could request human approval for critical actions.&lt;/p&gt;
&lt;p&gt;✅ It could verify its own work before reporting completion.&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
☁️ Cloud or Local?
&lt;/h2&gt;

&lt;p&gt;
I also do not think this future will be entirely cloud-based.
&lt;/p&gt;

&lt;p&gt;
As personal and enterprise data becomes more important,
hybrid architectures will probably become increasingly attractive.
&lt;/p&gt;

&lt;p&gt;
Small and fast models may run continuously on local devices.
&lt;/p&gt;

&lt;p&gt;
When more complex reasoning is required,
the system could call more powerful cloud models.
&lt;/p&gt;

&lt;p&gt;
Organization-specific knowledge could remain on-premise.
&lt;/p&gt;

&lt;p&gt;
Highly sensitive tasks could be handled entirely by local models.
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f4faf8; border-left:5px solid #15956c; padding:20px 22px; margin:25px 0; font-size:20px;"&gt;
  In the future, instead of asking “Which model are you using?”
  we may increasingly ask: “What kind of AI architecture is your system running on?”
&lt;/div&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#129302; Another Strong Candidate: Physical AI
&lt;/h2&gt;

&lt;p&gt;
Another major field to watch is artificial intelligence moving beyond the computer screen
and into the physical world.
&lt;/p&gt;

&lt;p&gt;
When vision, reasoning, planning, and agent technologies are combined with robotics,
we may see major advances in embodied AI or physical AI.
&lt;/p&gt;

&lt;p&gt;
An agent that learns how to click a button on a computer screen today
and an AI system that controls a physical robot tomorrow
may be more closely related problems than they first appear.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:45px;"&gt;
&#128640; Perhaps the Most Important Part of the Revolution Is Just Beginning
&lt;/h2&gt;

&lt;p&gt;
When I look back at the developments of the past 15 years,
the most interesting change for me is how our questions have changed.
&lt;/p&gt;

&lt;p&gt;
During the Deep Learning era, we asked:
&lt;/p&gt;

&lt;p&gt;
“Can the machine recognize this?”
&lt;/p&gt;

&lt;p&gt;
During the LLM era, the question became:
&lt;/p&gt;

&lt;p&gt;
“Can the machine understand and explain this?”
&lt;/p&gt;

&lt;p&gt;
With reasoning models, we started asking:
&lt;/p&gt;

&lt;p&gt;
“Can the machine solve this problem?”
&lt;/p&gt;

&lt;p&gt;
In the age of agents, the question is very different:
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#18243e; color:white; text-align:center; padding:28px 20px; margin:30px 0; border-radius:12px; font-size:25px; line-height:1.4;"&gt;
  “Can I give this entire task to AI from beginning to end?”
&lt;/div&gt;

&lt;p&gt;
I do not think this is a small change.
&lt;/p&gt;

&lt;p&gt;
For most of the history of computing, humans learned how to use computers.
&lt;/p&gt;

&lt;p&gt;
Now computers are beginning to learn how to use the same tools that we use.
&lt;/p&gt;

&lt;p&gt;
Perhaps the most important transformation of the next few years will happen precisely here.
&lt;/p&gt;


&lt;div style="width:100%; box-sizing:border-box; background:#f3f6fb; border:1px solid #dfe5ee; border-radius:12px; padding:25px; margin-top:38px;"&gt;

  &lt;div style="font-size:24px; color:#18243e; margin-bottom:15px;"&gt;
    &#129300; What Is Your Prediction?
  &lt;/div&gt;

  &lt;p&gt;
    Which technology or approach do you think will become the defining AI trend of 2027?
  &lt;/p&gt;

  &lt;p&gt;&#129302; Persistent personal AI agents?&lt;/p&gt;
  &lt;p&gt;&#127970; Enterprise agent systems?&lt;/p&gt;
  &lt;p&gt;&#127959;️ Agent harnesses and long-running autonomous work?&lt;/p&gt;
  &lt;p&gt;&#129504; More powerful reasoning models?&lt;/p&gt;
  &lt;p&gt;&#128421;️ Computer use and browser use?&lt;/p&gt;
  &lt;p&gt;&#127757; Open and locally running models?&lt;/p&gt;
  &lt;p&gt;&#129302; Physical AI and robotics?&lt;/p&gt;

  &lt;p style="margin-bottom:0;"&gt;
    Or will the next major breakthrough come from something
    that we are not yet talking about enough?
  &lt;/p&gt;

&lt;/div&gt;


&lt;h3 style="font-size:21px; color:#374151; margin-top:45px;"&gt;
Selected Milestones
&lt;/h3&gt;

&lt;div style="font-size:14px; color:#626b77; line-height:1.65;"&gt;

&lt;p&gt;Krizhevsky, Sutskever &amp;amp; Hinton — ImageNet Classification with Deep Convolutional Neural Networks, 2012&lt;/p&gt;

&lt;p&gt;Sutskever, Vinyals &amp;amp; Le — Sequence to Sequence Learning with Neural Networks, 2014&lt;/p&gt;

&lt;p&gt;Bahdanau, Cho &amp;amp; Bengio — Neural Machine Translation by Jointly Learning to Align and Translate, 2014&lt;/p&gt;

&lt;p&gt;Vaswani et al. — Attention Is All You Need, 2017&lt;/p&gt;

&lt;p&gt;Devlin et al. — BERT, 2018&lt;/p&gt;

&lt;p&gt;Kaplan et al. — Scaling Laws for Neural Language Models, 2020&lt;/p&gt;

&lt;p&gt;Brown et al. — Language Models are Few-Shot Learners, 2020&lt;/p&gt;

&lt;p&gt;Lewis et al. — Retrieval-Augmented Generation, 2020&lt;/p&gt;

&lt;p&gt;Ho, Jain &amp;amp; Abbeel — Denoising Diffusion Probabilistic Models, 2020&lt;/p&gt;

&lt;/div&gt;


&lt;div style="width:100%; box-sizing:border-box; border-top:1px solid #e1e5ea; margin-top:35px; padding-top:22px; color:#315caa; font-size:15px; line-height:1.8;"&gt;
 #MuratKarakayaAkademi #ArtificialIntelligence &amp;nbsp; #GenerativeAI &amp;nbsp; #LLM &amp;nbsp; #AIAgents &amp;nbsp; #AgenticAI &amp;nbsp; #DeepLearning &amp;nbsp; #Transformer &amp;nbsp; #ReasoningModels &amp;nbsp; #RAG &amp;nbsp; #MCP &amp;nbsp; #AgentSkills &amp;nbsp; #ComputerUse &amp;nbsp; #OpenSourceAI &amp;nbsp; #LocalAI &amp;nbsp; #AIEngineering &amp;nbsp; #FutureOfAI
&lt;/div&gt;

&lt;/div&gt;</description><link>https://www.muratkarakaya.net/2026/09/artificial-intelligence-llm-agentic-ai.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-5003745461074221164</guid><pubDate>Sat, 26 Sep 2026 20:05:40 +0000</pubDate><atom:updated>2026-09-26T23:05:40.233+03:00</atom:updated><category domain="http://www.blogger.com/atom/ns#">Agentic AI</category><category domain="http://www.blogger.com/atom/ns#">AI</category><category domain="http://www.blogger.com/atom/ns#">AI Agents</category><category domain="http://www.blogger.com/atom/ns#">Artificial Intelligence</category><category domain="http://www.blogger.com/atom/ns#">Engineering</category><category domain="http://www.blogger.com/atom/ns#">Generative AI</category><category domain="http://www.blogger.com/atom/ns#">Large Language Models</category><category domain="http://www.blogger.com/atom/ns#">LLM</category><category domain="http://www.blogger.com/atom/ns#">Project Management</category><category domain="http://www.blogger.com/atom/ns#">R&amp;D</category><category domain="http://www.blogger.com/atom/ns#">RAG</category><category domain="http://www.blogger.com/atom/ns#">Research and Development</category><category domain="http://www.blogger.com/atom/ns#">Systems Engineering</category><title>Artificial Intelligence • R&amp;amp;D • Project Management</title><description>&lt;div style="max-width: 920px; margin: 0 auto; font-family: Arial, Helvetica, sans-serif; color: #222; line-height: 1.75; font-size: 17px;"&gt;

  &lt;!-- HEADER --&gt;

  &lt;div style="padding: 36px 32px; margin-bottom: 32px; background: linear-gradient(135deg, #0f172a, #1e3a5f); color: white; border-radius: 16px;"&gt;
    &lt;div style="font-size: 14px; letter-spacing: 1.5px; text-transform: uppercase; opacity: 0.8; margin-bottom: 12px;"&gt;
      Artificial Intelligence • R&amp;amp;D • Project Management
    &lt;/div&gt;

```
&lt;h1 style="font-size: 38px; line-height: 1.2; margin: 0 0 18px 0; color: white;"&gt;
  &#128680; Why Do AI Projects Fail?
&lt;/h1&gt;

&lt;p style="font-size: 21px; line-height: 1.55; margin: 0; color: #e2e8f0;"&gt;
  Project proposals are becoming increasingly impressive. But can we say the same about the engineering capacity of the teams expected to turn those proposals into working systems?
&lt;/p&gt;
```

  &lt;/div&gt;

  &lt;!-- INTRO --&gt;

  &lt;p&gt;
    Recently, I have been noticing the same problem more and more often during project evaluations.
  &lt;/p&gt;

  &lt;p&gt;
    When you start reading an AI-related project proposal, you often see almost every recent method, technique, and technology somehow included in the project:
  &lt;/p&gt;

  &lt;div style="background-color: #f8fafc; border-left: 5px solid #2563eb; padding: 20px 24px; margin: 28px 0; border-radius: 6px;"&gt;
    Large language models, RAG, AI agents, vector databases, personalization, fine-tuning, multi-agent architectures, real-time analytics, prediction systems, decision support, explainable AI...
  &lt;/div&gt;

  &lt;p&gt;
    On paper, the project looks highly impressive.
  &lt;/p&gt;

  &lt;p&gt;
    However, once you meet the team face to face and begin asking more detailed questions, a very different picture often emerges.
  &lt;/p&gt;

  &lt;!-- SECTION --&gt;

  &lt;h2 style="font-size: 28px; color: #0f172a; margin-top: 48px; padding-bottom: 10px; border-bottom: 2px solid #e2e8f0;"&gt;
    &#128101; The technology is in the proposal — but is the experience in the team?
  &lt;/h2&gt;

  &lt;p&gt;
    Sometimes, the engineers on the project team have never developed even a small-scale application using the technologies described in the proposal.
  &lt;/p&gt;

  &lt;p&gt;
    The company itself may also have no prior experience building, deploying, operating, or maintaining products based on those methods.
  &lt;/p&gt;

  &lt;p&gt;
    After a few questions, the following issues begin to surface:
  &lt;/p&gt;

  &lt;ul style="padding-left: 26px;"&gt;
    &lt;li style="margin-bottom: 10px;"&gt;Why was this technology selected?&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;What real problem does it solve in the project?&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;Why is it being used together with the other components?&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;Were alternative methods evaluated?&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;How will success be measured?&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;If the system fails, how will the team determine which component failed and why?&lt;/li&gt;
  &lt;/ul&gt;

  &lt;p&gt;
    Unfortunately, in some projects, there are no clear answers.
  &lt;/p&gt;

  &lt;p&gt;
    The technology choices appear to be less the result of an engineering analysis and more the result of connecting fashionable concepts to one another.
  &lt;/p&gt;

  &lt;!-- HIGHLIGHT --&gt;

  &lt;div style="background-color: #fff7ed; border: 1px solid #fed7aa; padding: 26px; margin: 34px 0; border-radius: 12px;"&gt;
    &lt;div style="font-size: 21px; line-height: 1.6; color: #9a3412;"&gt;
      &#128161; The problem is not that too many technologies are being used. The problem is that the team cannot explain, from an engineering perspective, why each technology is actually necessary.
    &lt;/div&gt;
  &lt;/div&gt;

  &lt;!-- SECTION --&gt;

  &lt;h2 style="font-size: 28px; color: #0f172a; margin-top: 48px; padding-bottom: 10px; border-bottom: 2px solid #e2e8f0;"&gt;
    &#129513; A list of technologies is not a system design
  &lt;/h2&gt;

  &lt;p&gt;
    Today, it is relatively easy to produce impressive project documents with generative AI.
  &lt;/p&gt;

  &lt;p&gt;
    You can list the advantages of every technology. You can summarize their benefits from the literature. You can draw an architecture with a few boxes and arrows.
  &lt;/p&gt;

  &lt;p&gt;
    But none of this, by itself, constitutes systems engineering.
  &lt;/p&gt;

  &lt;p&gt;
    In a real system design, every component has a reason to exist.
  &lt;/p&gt;

  &lt;p&gt;
    A component should not be added because it is fashionable. It should be added because it satisfies a specific requirement.
  &lt;/p&gt;

  &lt;p&gt;
    For every selected method, the team should be able to explain:
  &lt;/p&gt;

  &lt;ul style="padding-left: 26px;"&gt;
    &lt;li style="margin-bottom: 10px;"&gt;which requirement it addresses,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;which alternatives it was compared against,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;which metric it is expected to improve,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;which costs it introduces,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;and which new risks it adds to the system.&lt;/li&gt;
  &lt;/ul&gt;

  &lt;p&gt;
    Otherwise, what we have is not a system architecture. It is merely a list of current technologies.
  &lt;/p&gt;

  &lt;!-- SECTION --&gt;

  &lt;h2 style="font-size: 28px; color: #0f172a; margin-top: 48px; padding-bottom: 10px; border-bottom: 2px solid #e2e8f0;"&gt;
    &#128269; Saying “we will use RAG” is easy. Building a working RAG system is not.
  &lt;/h2&gt;

  &lt;p&gt;
    For example, the following sentence has become very common in AI project proposals:
  &lt;/p&gt;

  &lt;div style="background-color: #f1f5f9; padding: 22px 26px; margin: 26px 0; border-radius: 10px; font-style: italic;"&gt;
    “A RAG-based artificial intelligence system will be developed to generate reliable answers from corporate documents.”
  &lt;/div&gt;

  &lt;p&gt;
    Fine.
  &lt;/p&gt;

  &lt;p&gt;
    But the real engineering problem begins exactly where that sentence ends.
  &lt;/p&gt;

  &lt;div style="background-color: #f8fafc; padding: 26px; border-radius: 12px; margin: 28px 0;"&gt;
    &lt;p style="margin-top: 0;"&gt;&#128204; How will system performance be measured?&lt;/p&gt;
    &lt;p&gt;&#128204; How will the benchmark dataset be created?&lt;/p&gt;
    &lt;p&gt;&#128204; How will real user questions be represented in the evaluation set?&lt;/p&gt;
    &lt;p&gt;&#128204; How will retrieval quality be separated from final answer quality?&lt;/p&gt;
    &lt;p&gt;&#128204; How will chunk size and overlap be selected?&lt;/p&gt;
    &lt;p&gt;&#128204; Which experiment will determine the embedding model?&lt;/p&gt;
    &lt;p&gt;&#128204; Which metadata fields will distinguish text chunks from tens of thousands of documents?&lt;/p&gt;
    &lt;p&gt;&#128204; If multiple versions of the same document exist, which one will be used?&lt;/p&gt;
    &lt;p&gt;&#128204; How will obsolete or withdrawn information be removed from the system?&lt;/p&gt;
    &lt;p&gt;&#128204; How will unauthorized users be prevented from retrieving restricted content?&lt;/p&gt;
    &lt;p&gt;&#128204; How will hallucination be measured?&lt;/p&gt;
    &lt;p&gt;&#128204; How should the system behave when a reliable answer cannot be generated?&lt;/p&gt;
    &lt;p&gt;&#128204; How will latency be controlled under thousands of concurrent users?&lt;/p&gt;
    &lt;p style="margin-bottom: 0;"&gt;&#128204; How will the trade-off between cost, accuracy, and response time be managed?&lt;/p&gt;
  &lt;/div&gt;

  &lt;p&gt;
    Every one of these is a real-world engineering problem.
  &lt;/p&gt;

  &lt;p&gt;
    Yet in many project documents, instead of answering these questions, the proposal simply explains the generic benefits of RAG.
  &lt;/p&gt;

  &lt;p&gt;
    That tells us that the project may be conceptually defined, but its operational reality has not been sufficiently considered.
  &lt;/p&gt;

  &lt;!-- SECTION --&gt;

  &lt;h2 style="font-size: 28px; color: #0f172a; margin-top: 48px; padding-bottom: 10px; border-bottom: 2px solid #e2e8f0;"&gt;
    ⚠️ Using AI to prepare a project is not the same as developing an AI project
  &lt;/h2&gt;

  &lt;p&gt;
    I believe a new risk has emerged in recent years.
  &lt;/p&gt;

  &lt;p&gt;
    Generative AI has made project preparation dramatically easier.
  &lt;/p&gt;

  &lt;p&gt;
    Literature reviews, technology descriptions, methodology sections, and technical documentation that once took weeks can now be produced much faster.
  &lt;/p&gt;

  &lt;p&gt;
    This is not inherently a bad development.
  &lt;/p&gt;

  &lt;p&gt;
    On the contrary, when used properly, it can create enormous productivity gains.
  &lt;/p&gt;

  &lt;p&gt;
    The problem begins when the process turns into something like this:
  &lt;/p&gt;

  &lt;div style="background: linear-gradient(135deg, #fff1f2, #fff7ed); border: 1px solid #fecdd3; padding: 28px; margin: 30px 0; border-radius: 12px;"&gt;
    &lt;p style="font-size: 19px; margin: 0 0 14px 0;"&gt;&#129302; Ask AI to write the project proposal.&lt;/p&gt;
    &lt;p style="font-size: 19px; margin: 0 0 14px 0;"&gt;&#128176; Apply for funding.&lt;/p&gt;
    &lt;p style="font-size: 19px; margin: 0;"&gt;&#129489;‍&#128187; If the funding is approved, ask AI to write the code too.&lt;/p&gt;
  &lt;/div&gt;

  &lt;p&gt;
    This mindset can easily turn into a dangerous form of shortcut-driven project development.
  &lt;/p&gt;

  &lt;p&gt;
    The engineering gap between a project proposal and a real system cannot automatically be closed simply by using more AI.
  &lt;/p&gt;

  &lt;!-- SECTION --&gt;

  &lt;h2 style="font-size: 28px; color: #0f172a; margin-top: 48px; padding-bottom: 10px; border-bottom: 2px solid #e2e8f0;"&gt;
    &#129504; AI does not automatically create experience that the team does not have
  &lt;/h2&gt;

  &lt;p&gt;
    Today’s generative AI systems can write code, suggest architectures, create APIs, generate test scenarios, and produce documentation.
  &lt;/p&gt;

  &lt;p&gt;
    But there is still a critical distinction:
  &lt;/p&gt;

  &lt;div style="background-color: #ecfeff; border-left: 5px solid #0891b2; padding: 22px 26px; margin: 28px 0; border-radius: 6px;"&gt;
    AI can generate a solution for you. But determining whether that solution is actually the right solution for your problem still requires knowledge and experience.
  &lt;/div&gt;

  &lt;p&gt;
    Because real engineering is not simply about generating code.
  &lt;/p&gt;

  &lt;p&gt;
    Real engineering means:
  &lt;/p&gt;

  &lt;ul style="padding-left: 26px;"&gt;
    &lt;li style="margin-bottom: 10px;"&gt;defining the right problem,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;identifying the requirements,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;comparing alternative approaches,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;defining measurable success criteria,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;anticipating bottlenecks,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;handling unexpected cases,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;testing the system under real user behavior,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;and considering scalability and sustainability.&lt;/li&gt;
  &lt;/ul&gt;

  &lt;!-- SECTION --&gt;

  &lt;h2 style="font-size: 28px; color: #0f172a; margin-top: 48px; padding-bottom: 10px; border-bottom: 2px solid #e2e8f0;"&gt;
    &#128201; So what happens in the end?
  &lt;/h2&gt;

  &lt;p&gt;
    A system that looked extremely strong in the project proposal fails to deliver the expected performance during implementation.
  &lt;/p&gt;

  &lt;p&gt;
    At first, a few demonstrations are prepared.
  &lt;/p&gt;

  &lt;p&gt;
    The system performs well on carefully selected examples.
  &lt;/p&gt;

  &lt;p&gt;
    But once real data, real users, and real operational conditions enter the picture, problems begin to appear.
  &lt;/p&gt;

  &lt;p&gt;
    Costs increase.
  &lt;/p&gt;

  &lt;p&gt;
    Latency rises.
  &lt;/p&gt;

  &lt;p&gt;
    Data quality problems emerge.
  &lt;/p&gt;

  &lt;p&gt;
    Authorization requirements become important.
  &lt;/p&gt;

  &lt;p&gt;
    The model begins producing unexpected answers.
  &lt;/p&gt;

  &lt;p&gt;
    Real user questions do not resemble the clean examples used in demonstrations.
  &lt;/p&gt;

  &lt;p&gt;
    And because there is no properly designed benchmark, the team may not even know which change actually improves the system.
  &lt;/p&gt;

  &lt;p&gt;
    As time passes, the project schedule slips.
  &lt;/p&gt;

  &lt;p&gt;
    The team changes.
  &lt;/p&gt;

  &lt;p&gt;
    Employees leave.
  &lt;/p&gt;

  &lt;p&gt;
    New team members do not understand why previous architectural decisions were made.
  &lt;/p&gt;

  &lt;p&gt;
    Eventually, the project is either scaled down, diverted from its original objective, or left behind as an unused prototype.
  &lt;/p&gt;

  &lt;!-- SECTION --&gt;

  &lt;h2 style="font-size: 28px; color: #0f172a; margin-top: 48px; padding-bottom: 10px; border-bottom: 2px solid #e2e8f0;"&gt;
    &#127919; Perhaps we are asking the wrong question
  &lt;/h2&gt;

  &lt;p&gt;
    When evaluating AI projects, we often focus on this question:
  &lt;/p&gt;

  &lt;div style="text-align: center; font-size: 23px; font-style: italic; color: #1e3a8a; margin: 32px 0;"&gt;
    “Which technologies will be used in the project?”
  &lt;/div&gt;

  &lt;p&gt;
    I believe there is a more important question:
  &lt;/p&gt;

  &lt;div style="text-align: center; font-size: 24px; line-height: 1.5; color: #9a3412; margin: 32px 0;"&gt;
    “Does this team actually have the engineering capacity to build the system described in the proposal?”
  &lt;/div&gt;

  &lt;p&gt;
    This should be followed by several additional questions:
  &lt;/p&gt;

  &lt;ul style="padding-left: 26px;"&gt;
    &lt;li style="margin-bottom: 10px;"&gt;Has the team solved a similar problem before?&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;Has it at least built a small-scale prototype?&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;Does the team understand the technical risks?&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;Does it know which assumptions may turn out to be wrong?&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;Can it design experiments to measure success?&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;Does it know how to validate the system under real user load?&lt;/li&gt;
  &lt;/ul&gt;

  &lt;!-- SECTION --&gt;

  &lt;h2 style="font-size: 28px; color: #0f172a; margin-top: 48px; padding-bottom: 10px; border-bottom: 2px solid #e2e8f0;"&gt;
    &#127959;️ How should a good AI project begin?
  &lt;/h2&gt;

  &lt;p&gt;
    In my view, the process should begin with the problem, not with technology names.
  &lt;/p&gt;

  &lt;div style="overflow-x: auto; margin: 30px 0;"&gt;
    &lt;table style="width: 100%; border-collapse: collapse; font-size: 16px;"&gt;
      &lt;thead&gt;
        &lt;tr style="background-color: #0f172a; color: white;"&gt;
          &lt;th style="padding: 14px; border: 1px solid #334155; text-align: left;"&gt;Wrong starting point&lt;/th&gt;
          &lt;th style="padding: 14px; border: 1px solid #334155; text-align: left;"&gt;A better starting point&lt;/th&gt;
        &lt;/tr&gt;
      &lt;/thead&gt;
      &lt;tbody&gt;
        &lt;tr&gt;
          &lt;td style="padding: 14px; border: 1px solid #e2e8f0;"&gt;“Let’s use RAG.”&lt;/td&gt;
          &lt;td style="padding: 14px; border: 1px solid #e2e8f0;"&gt;“Let’s understand which information users cannot access and why.”&lt;/td&gt;
        &lt;/tr&gt;
        &lt;tr style="background-color: #f8fafc;"&gt;
          &lt;td style="padding: 14px; border: 1px solid #e2e8f0;"&gt;“Let’s add agents.”&lt;/td&gt;
          &lt;td style="padding: 14px; border: 1px solid #e2e8f0;"&gt;“Let’s identify which tasks genuinely require multi-step reasoning and decision-making.”&lt;/td&gt;
        &lt;/tr&gt;
        &lt;tr&gt;
          &lt;td style="padding: 14px; border: 1px solid #e2e8f0;"&gt;“Let’s personalize everything with an LLM.”&lt;/td&gt;
          &lt;td style="padding: 14px; border: 1px solid #e2e8f0;"&gt;“Let’s measure which user information actually improves response quality.”&lt;/td&gt;
        &lt;/tr&gt;
        &lt;tr style="background-color: #f8fafc;"&gt;
          &lt;td style="padding: 14px; border: 1px solid #e2e8f0;"&gt;“Let’s use the most powerful model.”&lt;/td&gt;
          &lt;td style="padding: 14px; border: 1px solid #e2e8f0;"&gt;“Let’s experimentally determine the right trade-off between quality, cost, and latency.”&lt;/td&gt;
        &lt;/tr&gt;
      &lt;/tbody&gt;
    &lt;/table&gt;
  &lt;/div&gt;

  &lt;!-- CONCLUSION --&gt;

  &lt;h2 style="font-size: 28px; color: #0f172a; margin-top: 48px; padding-bottom: 10px; border-bottom: 2px solid #e2e8f0;"&gt;
    &#128640; A strong project proposal does not necessarily mean a strong project
  &lt;/h2&gt;

  &lt;p&gt;
    Today, it is possible to prepare an excellent project proposal with the help of AI.
  &lt;/p&gt;

  &lt;p&gt;
    In fact, compared with just a few years ago, proposals can now be more professional, more technical, and far more persuasive.
  &lt;/p&gt;

  &lt;p&gt;
    But the quality of the proposal and the feasibility of the project are not the same thing.
  &lt;/p&gt;

  &lt;p&gt;
    AI can accelerate project preparation.
  &lt;/p&gt;

  &lt;p&gt;
    It can accelerate software development.
  &lt;/p&gt;

  &lt;p&gt;
    It can simplify documentation.
  &lt;/p&gt;

  &lt;p&gt;
    It can generate test scenarios.
  &lt;/p&gt;

  &lt;p&gt;
    It can suggest architectural alternatives.
  &lt;/p&gt;

  &lt;p&gt;
    But it does not magically create engineering experience, domain knowledge, systems thinking, or product development culture that the team does not already possess.
  &lt;/p&gt;

  &lt;div style="background: linear-gradient(135deg, #0f172a, #1e293b); color: white; padding: 30px; margin: 36px 0; border-radius: 14px;"&gt;
    &lt;p style="font-size: 21px; line-height: 1.65; margin: 0; color: white;"&gt;
      &#127919; One of the biggest risks in AI projects may no longer be lack of access to technology. It may be the opposite: because access has become so easy, teams may begin highly ambitious projects without sufficient preparation or experience.
    &lt;/p&gt;
  &lt;/div&gt;

  &lt;p&gt;
    At the end of the day, we still need people who understand the problem, ask the right questions, design meaningful measurements, anticipate failure points, and architect the system from end to end.
  &lt;/p&gt;

  &lt;p style="font-size: 20px; margin-top: 34px;"&gt;
    &#128172; Have you also encountered AI projects that look extremely strong in project documentation but fail to translate into real-world systems?
  &lt;/p&gt;

  &lt;!-- HASHTAGS --&gt;

  &lt;div style="margin-top: 48px; padding-top: 24px; border-top: 1px solid #e2e8f0; color: #475569; font-size: 15px; line-height: 1.8;"&gt;
    #MuratKarakayaAkademi #ArtificialIntelligence #AI #GenerativeAI #LLM #RAG #AgenticAI #Engineering #SystemsEngineering #ProjectManagement #ResearchAndDevelopment #ProductDevelopment #AIProjects #Innovation #Technology
  &lt;/div&gt;

&lt;/div&gt;
</description><link>https://www.muratkarakaya.net/2026/09/artificial-intelligence-r-project.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-1939872800708057056</guid><pubDate>Sat, 26 Sep 2026 19:56:23 +0000</pubDate><atom:updated>2026-09-26T23:03:21.246+03:00</atom:updated><category domain="http://www.blogger.com/atom/ns#">Agentic</category><category domain="http://www.blogger.com/atom/ns#">Büyük Dil Modelleri</category><category domain="http://www.blogger.com/atom/ns#">LLM</category><category domain="http://www.blogger.com/atom/ns#">Yapay Zekâ</category><category domain="http://www.blogger.com/atom/ns#">Yapay Zekâ Ajanları</category><category domain="http://www.blogger.com/atom/ns#">Üretken Yapay Zekâ</category><title>Yapay Zekanın Yolculuğu</title><description>&lt;div style="width:100%; box-sizing:border-box; font-family:Arial,Helvetica,sans-serif; color:#222; font-size:17px; line-height:1.75; overflow-wrap:break-word;"&gt;

&lt;!-- GİRİŞ KUTUSU --&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f3f6fb; border:1px solid #dfe5ee; border-radius:14px; padding:28px; margin:0 0 35px 0;"&gt;

  &lt;div style="font-size:13px; color:#687386; letter-spacing:1px; margin-bottom:12px;"&gt;
    YAPAY ZEKÂ • LLM • AGENTIC AI • 2027
  &lt;/div&gt;

  &lt;h2 style="font-size:32px; line-height:1.25; color:#18243e; margin:0 0 18px 0;"&gt;
    Yapay Zekâdaki Devrimi Yaşarken Fark Etmek: Deep Learning’den Uzun Süre Çalışan Agent’lara
  &lt;/h2&gt;

  &lt;p style="font-size:19px; color:#46536a; margin:0;"&gt;
    Son 15 yılda yapay zekâ yalnızca daha başarılı modeller üretmedi.
    Görüntüyü tanıyan sistemlerden konuşan modellere, konuşan modellerden araç kullanan agent’lara,
    oradan da saatlerce hatta günlerce bir problem üzerinde çalışabilen yapay zekâ sistemlerine geçtik.
    Peki 2027’nin kırılma noktası ne olacak?
  &lt;/p&gt;

&lt;/div&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:35px;"&gt;
Bir sohbetten geriye kalan soru
&lt;/h2&gt;

&lt;p&gt;
Bugün yapay zekâya meraklı bir dostumla sohbet ederken konu doğal olarak son yıllarda bu alanda yaşanan baş döndürücü gelişmelere geldi.
&lt;/p&gt;

&lt;p&gt;
Bir süre sonra geçmişte hangi teknolojileri konuştuğumuzu hatırlamaya başladık. Deep Learning, RNN, LSTM, Transformer, büyük dil modelleri, multimodal modeller, reasoning modeller, RAG, function calling, agent’lar, MCP, computer use, coding agent’ları ve son olarak uzun süre çalışabilen agent sistemleri...
&lt;/p&gt;

&lt;p&gt;
Hepsini arka arkaya sıraladığımızda değişimin hızını daha net fark ediyorsunuz.
&lt;/p&gt;

&lt;p&gt;
Aslında son birkaç yılda yalnızca daha başarılı yapay zekâ modelleri geliştirmedik. Yapay zekâyı kullanma biçimimiz de birkaç kez tamamen değişti.
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f5f7ff; border-left:5px solid #3f5bd9; padding:20px 22px; margin:28px 0;"&gt;
  &lt;span style="font-size:20px; color:#26375f;"&gt;
    Yapay zekâ, örüntüleri tanıyan bir sistemden bizimle konuşan bir sisteme,
    oradan düşünen, araç kullanan ve giderek kendi başına iş yapabilen bir sisteme dönüşüyor.
  &lt;/span&gt;
&lt;/div&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#129504; 1. Deep Learning: Büyük kırılmanın başlangıcı
&lt;/h2&gt;

&lt;p&gt;
Yapay sinir ağları elbette yeni değildi. Ancak uzun yıllar boyunca yeterli veri, hesaplama gücü ve uygun eğitim tekniklerinin bulunmaması nedeniyle potansiyellerinin önemli bir kısmını gösterememişlerdi.
&lt;/p&gt;

&lt;p&gt;
2012’de AlexNet’in ImageNet yarışmasında elde ettiği başarı, GPU üzerinde eğitilen çok katmanlı yapay sinir ağlarının ne kadar güçlü olabileceğini gösterdi.
&lt;/p&gt;

&lt;p&gt;
Computer vision alanında başlayan bu sıçrama kısa sürede konuşma tanıma, doğal dil işleme ve diğer makine öğrenmesi problemlerine yayıldı.
&lt;/p&gt;

&lt;p&gt;
Bugünden baktığımızda basit görünebilir ama bugün yaşadığımız yapay zekâ devriminin önemli başlangıç noktalarından biri burada oluştu:
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#fff8e8; border:1px solid #ead9a8; padding:18px 22px; margin:24px 0;"&gt;
  Daha büyük ağlar + daha fazla veri + daha fazla hesaplama gücü.
&lt;/div&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128257; 2. RNN, LSTM ve Sequence-to-Sequence
&lt;/h2&gt;

&lt;p&gt;
Görüntülerde CNN mimarileri büyük başarı gösterirken doğal dil tarafında farklı bir sorun vardı. Dil sıralı bir yapıydı.
&lt;/p&gt;

&lt;p&gt;
RNN ve özellikle LSTM gibi mimariler geçmişten gelen bilgiyi belirli ölçüde taşıyarak cümleler, konuşmalar ve zaman serileri üzerinde çalışmayı mümkün hale getirdi.
&lt;/p&gt;

&lt;p&gt;
Ardından sequence-to-sequence, yani seq2seq mimarileri geldi. Bir diziyi alıp başka bir dizi üretmek mümkün hale geldi. Makine çevirisi bunun en önemli uygulamalarından biri oldu.
&lt;/p&gt;

&lt;p&gt;
Bugün bir modele verdiğimiz metnin karşılığında yeni bir metin üretmesini çok doğal karşılıyoruz. Ancak bugünkü üretken modellerin geçmişindeki önemli basamaklardan biri encoder-decoder ve seq2seq çalışmalarıydı.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128065;️ 3. Attention: Model nereye bakması gerektiğini öğreniyor
&lt;/h2&gt;

&lt;p&gt;
Burada kronolojik olarak önemli bir ayrıntı var. Attention mekanizması Transformer ile başlamadı.
&lt;/p&gt;

&lt;p&gt;
2014 civarında özellikle neural machine translation çalışmalarında modelin kaynak cümlenin her bölümüne aynı şekilde yaklaşmak yerine, çıktı üretirken ilgili bölümlere daha fazla dikkat verebilmesi gösterildi.
&lt;/p&gt;

&lt;p&gt;
Bu fikir son derece önemliydi. Çünkü model artık bütün girdiyi tek bir sabit temsil içerisine sıkıştırmak zorunda değildi.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
⚡ 4. 2017: Attention Is All You Need
&lt;/h2&gt;

&lt;p&gt;
2017’de yayımlanan &lt;em&gt;Attention Is All You Need&lt;/em&gt; makalesi yapay zekâ tarihinin en önemli dönüm noktalarından biri oldu.
&lt;/p&gt;

&lt;p&gt;
Transformer mimarisi, RNN’lerdeki sıralı işleme bağımlılığını büyük ölçüde ortadan kaldırdı ve self-attention mekanizmasını merkeze aldı.
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f4faf8; border-left:5px solid #15956c; padding:20px 22px; margin:25px 0;"&gt;
  Self-attention sayesinde bir metindeki farklı kelimeler arasındaki ilişkiler doğrudan modellenebiliyor ve hesaplamalar çok daha verimli biçimde paralelleştirilebiliyordu.
&lt;/div&gt;

&lt;p&gt;
Bugün kullandığımız GPT, Claude, Gemini, Llama, Qwen, Mistral ve benzeri büyük model ailelerinin mimari köklerinde büyük ölçüde bu dönüşüm bulunuyor.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128218; 5. Pre-training ve Foundation Model yaklaşımı
&lt;/h2&gt;

&lt;p&gt;
Transformer’dan sonra başka bir paradigma güç kazandı.
&lt;/p&gt;

&lt;p&gt;
Her görev için ayrı bir model eğitmek yerine, çok büyük veri kümeleri üzerinde genel amaçlı bir model eğitip daha sonra bunu farklı görevlere uyarlamak.
&lt;/p&gt;

&lt;p&gt;
BERT gibi modeller bu yaklaşımın doğal dil işlemede ne kadar güçlü olabileceğini gösterdi.
&lt;/p&gt;

&lt;p&gt;
Böylece tartışma yalnızca “hangi mimari daha başarılı?” sorusundan uzaklaşıp, çok sayıda görevi gerçekleştirebilecek genel amaçlı modeller geliştirmeye doğru kaydı.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128200; 6. Scaling ve GPT-3
&lt;/h2&gt;

&lt;p&gt;
2020 civarında model büyüklüğü, eğitim verisi ve hesaplama gücünün artırılmasıyla performans arasındaki ilişki çok daha görünür hale geldi.
&lt;/p&gt;

&lt;p&gt;
Scaling laws çalışmaları, daha büyük modellerin yalnızca aynı görevleri biraz daha iyi yapmadığını, bazı yeni yeteneklerin de model ölçeği arttıkça ortaya çıkabildiğini gösteriyordu.
&lt;/p&gt;

&lt;p&gt;
GPT-3 ise başka bir önemli özelliği görünür hale getirdi:
&lt;/p&gt;

&lt;p&gt;
Bir modele her görev için yeniden eğitim vermeden, yalnızca talimat ve birkaç örnek aracılığıyla çok farklı görevler yaptırılabiliyordu.
&lt;/p&gt;

&lt;p&gt;
Zero-shot ve few-shot kullanım giderek daha önemli hale geldi.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128196; 7. RAG: Modelin parametrelerinin dışına çıkmak
&lt;/h2&gt;

&lt;p&gt;
Büyük dil modellerinin önemli problemlerinden biri güncel, kuruma özel veya modelin eğitiminde bulunmayan bilgiye erişememeleriydi.
&lt;/p&gt;

&lt;p&gt;
Retrieval-Augmented Generation, yani RAG yaklaşımı bu problemi farklı bir şekilde ele aldı.
&lt;/p&gt;

&lt;p&gt;
Her bilgiyi modelin parametrelerine koymak yerine, soruyla ilgili bilgi dışarıdaki bir bilgi kaynağından bulunuyor ve modelin context’ine ekleniyordu.
&lt;/p&gt;

&lt;p&gt;
Daha sonra embedding, vector database, semantic search, hybrid search, reranking, query rewriting ve knowledge base mimarileriyle RAG çok daha gelişmiş bir yapıya dönüştü.
&lt;/p&gt;

&lt;p&gt;
Özellikle kurumsal yapay zekâ uygulamalarında bu yaklaşım bugün hâlâ son derece önemli.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#127912; 8. Diffusion modelleri ve üretken görsel yapay zekâ
&lt;/h2&gt;

&lt;p&gt;
Bu sırada yapay zekâ devrimi yalnızca metin üzerinde ilerlemiyordu.
&lt;/p&gt;

&lt;p&gt;
Diffusion modellerinin gelişmesiyle metinden görüntü üretmek çok hızlı biçimde yaygınlaştı.
&lt;/p&gt;

&lt;p&gt;
Ardından görüntü düzenleme, stil değiştirme, görselden video üretme, metinden video üretme, ses sentezi ve müzik üretimi gibi alanlarda hızlı ilerlemeler yaşandı.
&lt;/p&gt;

&lt;p&gt;
Generative AI kavramı artık yalnızca metin üreten modelleri ifade etmiyordu.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128064; 9. Multimodal modeller
&lt;/h2&gt;

&lt;p&gt;
Bir sonraki önemli adım farklı veri türlerinin aynı model içerisinde bir araya gelmesi oldu.
&lt;/p&gt;

&lt;p&gt;
Model artık yalnızca yazdığımız metni anlamıyordu.
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:8px;"&gt;&#128247; Fotoğrafları anlayabiliyor.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:8px;"&gt;&#128202; Grafikleri yorumlayabiliyor.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:8px;"&gt;&#128209; Belgeleri okuyabiliyor.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:8px;"&gt;&#127897;️ Sesi anlayabiliyor.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:8px;"&gt;&#128421;️ Bilgisayar ekranını görebiliyor.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:20px;"&gt;&#127909; Videoları analiz edebiliyor.&lt;/div&gt;

&lt;p&gt;
Gerçek dünya yalnızca metinden oluşmadığı için multimodalite yapay zekânın genel amaçlı hale gelmesi açısından çok önemli bir adımdı.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128172; 10. 2022: ChatGPT ile yapay zekâ geniş kitlelerle buluştu
&lt;/h2&gt;

&lt;p&gt;
30 Kasım 2022’de ChatGPT’nin kullanıma açılması bence teknolojik olduğu kadar toplumsal bir kırılma noktasıydı.
&lt;/p&gt;

&lt;p&gt;
Büyük dil modelleri ilk kez bu ölçekte sıradan kullanıcıların günlük hayatına girdi.
&lt;/p&gt;

&lt;p&gt;
Öğretmenler, öğrenciler, doktorlar, avukatlar, mühendisler, yöneticiler, sanatçılar ve yazılımcılar aynı teknolojiyle konuşmaya başladı.
&lt;/p&gt;

&lt;p&gt;
Instruction tuning ve insan geri bildirimiyle hizalama gibi çalışmalar sayesinde kullanıcı açısından çok daha doğal bir etkileşim ortaya çıktı.
&lt;/p&gt;

&lt;p&gt;
Aynı dönemde yeni bir kavram da günlük dilimize girdi:
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#fff8e8; padding:18px 22px; margin:24px 0;"&gt;
  Prompt Engineering.
&lt;/div&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128295; 11. Function Calling ve Tool Use
&lt;/h2&gt;

&lt;p&gt;
Büyük dil modelleri çok etkileyici metinler üretebiliyordu. Ancak gerçek dünyadaki sistemlerle etkileşimleri sınırlıydı.
&lt;/p&gt;

&lt;p&gt;
Function calling ve tool use yaklaşımı bunu değiştirmeye başladı.
&lt;/p&gt;

&lt;p&gt;
Model artık bir API çağırabiliyor, veritabanına erişebiliyor, hesaplama aracı kullanabiliyor, arama yapabiliyor veya başka bir yazılım bileşenini çalıştırabiliyordu.
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f4faf8; padding:20px 22px; margin:25px 0;"&gt;
  &lt;div style="margin-bottom:10px;"&gt;Eski yaklaşım: “Soruyu cevapla.”&lt;/div&gt;
  &lt;div&gt;Yeni yaklaşım: “Problemi çözmek için gerekiyorsa araç kullan.”&lt;/div&gt;
&lt;/div&gt;

&lt;p&gt;
Bu küçük görünen değişim daha sonra agent sistemlerinin temelini oluşturdu.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#129504; 12. Reasoning modeller
&lt;/h2&gt;

&lt;p&gt;
2024’te reasoning modellerin yaygınlaşmasıyla başka bir paradigma daha ortaya çıktı.
&lt;/p&gt;

&lt;p&gt;
Artık yalnızca modele daha fazla parametre eklemekten değil, modelin cevap vermeden önce problem üzerinde daha fazla hesaplama yapmasından söz ediyorduk.
&lt;/p&gt;

&lt;p&gt;
Test-time compute kavramı önem kazandı.
&lt;/p&gt;

&lt;p&gt;
Özellikle matematik, bilim, kodlama ve karmaşık problem çözme alanlarında “daha büyük model” yaklaşımının yanına “problem üzerinde daha uzun çalışan model” yaklaşımı eklendi.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128187; 13. Chatbot’tan Coding Agent’a
&lt;/h2&gt;

&lt;p&gt;
Bence yakın dönemin en önemli dönüşümlerinden biri yazılım geliştirme alanında gerçekleşti.
&lt;/p&gt;

&lt;p&gt;
İlk aşamada modellerden kod parçaları istemeye başladık.
&lt;/p&gt;

&lt;p&gt;
Sonra IDE içerisine yerleşen copilot sistemleri geldi.
&lt;/p&gt;

&lt;p&gt;
Daha sonra yapay zekâ bütün repository’yi okumaya, dosya oluşturmaya, terminal çalıştırmaya, hata ayıklamaya, test yazmaya ve kodu yeniden düzenlemeye başladı.
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:8px;"&gt;&#128172; Kod hakkında konuşan sistem&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:8px;"&gt;➡️ Kod öneren sistem&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:8px;"&gt;➡️ Kod yazan sistem&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:8px;"&gt;➡️ Repository üzerinde çalışan sistem&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:14px 18px; margin-bottom:20px;"&gt;➡️ Yazılım geliştirme görevini üstlenen agent&lt;/div&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128268; 14. MCP: Yapay zekâ ile dış dünya arasında standart bağlantılar
&lt;/h2&gt;

&lt;p&gt;
Agent sistemleri yaygınlaştıkça başka bir problem ortaya çıktı.
&lt;/p&gt;

&lt;p&gt;
Her yapay zekâ uygulamasını GitHub, dosya sistemi, veritabanı, kurumsal yazılımlar ve farklı servislerle ayrı ayrı entegre etmek gerekiyordu.
&lt;/p&gt;

&lt;p&gt;
Model Context Protocol, yani MCP, yapay zekâ uygulamaları ile araçlar ve veri kaynakları arasında daha standart bağlantılar kurulabilmesi açısından önemli bir gelişme oldu.
&lt;/p&gt;

&lt;p&gt;
Bence MCP’yi yalnızca teknik bir protokol olarak görmemek gerekiyor. Agent ekosisteminin büyümesi için modellerin dış dünyaya standart ve yönetilebilir biçimde bağlanması çok önemli.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#129513; 15. Agent Skills
&lt;/h2&gt;

&lt;p&gt;
Sonraki aşamada agent’lara belirli işleri nasıl yapacaklarını tekrar tekrar açıklamak yerine yeniden kullanılabilir beceriler tanımlama yaklaşımı güç kazandı.
&lt;/p&gt;

&lt;p&gt;
Bir skill örneğin:
&lt;/p&gt;

&lt;p&gt;&#128202; Bir Excel dosyasının nasıl analiz edileceğini,&lt;/p&gt;
&lt;p&gt;&#128196; Profesyonel bir raporun nasıl hazırlanacağını,&lt;/p&gt;
&lt;p&gt;&#129514; Bir test sürecinin nasıl yürütüleceğini,&lt;/p&gt;
&lt;p&gt;&#127970; Kuruma özel bir prosedürün nasıl uygulanacağını,&lt;/p&gt;
&lt;p&gt;&#128187; Belirli bir yazılım geliştirme işinin nasıl yapılacağını&lt;/p&gt;

&lt;p&gt;
agent’a gerektiği anda öğretebiliyor.
&lt;/p&gt;

&lt;p&gt;
Bu yaklaşım genel amaçlı yapay zekâdan uzmanlaşmış dijital çalışanlara geçiş açısından oldukça önemli görünüyor.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#128421;️ 16. Computer Use ve Browser Use
&lt;/h2&gt;

&lt;p&gt;
Daha sonra agent’lar yalnızca API kullanmakla kalmadı.
&lt;/p&gt;

&lt;p&gt;
Ekranı görmeye, fareyi hareket ettirmeye, butonlara basmaya, web sayfalarında gezinmeye ve bizim kullandığımız grafik kullanıcı arayüzlerini kullanmaya başladı.
&lt;/p&gt;

&lt;p&gt;
Bu bence çok kritik bir gelişme.
&lt;/p&gt;

&lt;p&gt;
Çünkü dünyadaki yazılımların büyük kısmının yapay zekâ için hazırlanmış API’leri yok. Ancak neredeyse hepsinin insanlar için geliştirilmiş bir kullanıcı arayüzü var.
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f5f7ff; border-left:5px solid #3f5bd9; padding:20px 22px; margin:25px 0;"&gt;
  Computer use ile birlikte kullanıcı arayüzünün kendisi bir anlamda yapay zekânın kullanabileceği yeni bir entegrasyon katmanına dönüşüyor.
&lt;/div&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#127757; 17. Açık modeller ve lokal yapay zekâ
&lt;/h2&gt;

&lt;p&gt;
Bütün bu gelişmeler devam ederken paralel bir dönüşüm daha yaşandı.
&lt;/p&gt;

&lt;p&gt;
Açık ağırlıklı modeller, kapalı modellerle aralarındaki farkı birçok kullanım alanında hızla azalttı.
&lt;/p&gt;

&lt;p&gt;
Llama, Mistral, Qwen, DeepSeek ve benzeri model aileleri güçlü bir açık ekosistem oluşturdu.
&lt;/p&gt;

&lt;p&gt;
Quantization, daha verimli inference motorları ve donanım gelişmeleri sayesinde daha önce yalnızca büyük veri merkezlerinde çalıştırılabilecek modeller kişisel bilgisayarlarda ve kurum içi sunucularda çalışmaya başladı.
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#fff5f2; border-left:5px solid #d76545; padding:20px 22px; margin:25px 0;"&gt;
  Yapay zekâ artık yalnızca internet üzerinden eriştiğimiz bir bulut servisi olmak zorunda değil.
&lt;/div&gt;

&lt;p&gt;
Verinin kurum dışına çıkmadan işlendiği lokal yapay zekâ sistemleri giderek daha gerçekçi bir seçenek haline geliyor.
&lt;/p&gt;

&lt;p&gt;
Kişisel bilgisayarda sürekli çalışan agent yazılımlarının ortaya çıkması da bu dönüşümün önemli parçalarından biri.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#129302; 18. 2026: Agent yılı
&lt;/h2&gt;

&lt;p&gt;
2026 için uzun zamandır “agent’ların yılı olacak” deniyordu.
&lt;/p&gt;

&lt;p&gt;
Bence bu tahmin önemli ölçüde gerçekleşti.
&lt;/p&gt;

&lt;p&gt;
Ancak agent kavramının kendisi de değişti.
&lt;/p&gt;

&lt;p&gt;
İlk agent mimarilerinde yapı kabaca şöyleydi:
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:20px 22px; margin:25px 0; text-align:center;"&gt;
  LLM → araç çağır → sonucu gör → gerekirse başka araç çağır → cevap ver
&lt;/div&gt;

&lt;p&gt;
Bugün ise çok daha kapsamlı sistemlerden söz ediyoruz.
&lt;/p&gt;

&lt;p&gt;&#129517; Planlama yapıyor.&lt;/p&gt;
&lt;p&gt;&#129520; Doğru aracı seçiyor.&lt;/p&gt;
&lt;p&gt;&#128450;️ Dosyaları ve context’i yönetiyor.&lt;/p&gt;
&lt;p&gt;&#129504; Durum bilgisini koruyor.&lt;/p&gt;
&lt;p&gt;&#129309; Başka agent’lara görev verebiliyor.&lt;/p&gt;
&lt;p&gt;&#128187; Kod ve terminal çalıştırabiliyor.&lt;/p&gt;
&lt;p&gt;&#127760; Web üzerinde işlem yapabiliyor.&lt;/p&gt;
&lt;p&gt;&#128269; Kendi ürettiği sonucu kontrol edebiliyor.&lt;/p&gt;
&lt;p&gt;&#128737;️ Yetki sınırlarını dikkate alabiliyor.&lt;/p&gt;
&lt;p&gt;&#128100; Gerektiğinde insan onayı isteyebiliyor.&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#127959;️ 19. Sonraki aşama: Agent Harness
&lt;/h2&gt;

&lt;p&gt;
Son dönemde benim özellikle önemli bulduğum kavramlardan biri agent harness.
&lt;/p&gt;

&lt;p&gt;
Çünkü agent’lara daha karmaşık ve uzun görevler vermeye başladığımızda yalnızca modelin ne kadar akıllı olduğu yeterli olmuyor.
&lt;/p&gt;

&lt;p&gt;
Bir model ilk birkaç adımı çok başarılı tamamladıktan sonra yanlış bir yöne gidebilir.
&lt;/p&gt;

&lt;p&gt;
Görevin bittiğini erken düşünebilir.&lt;/p&gt;

&lt;p&gt;
Önceki kararlarını unutabilir.&lt;/p&gt;

&lt;p&gt;
Test etmeden sonuca ulaşabilir.&lt;/p&gt;

&lt;p&gt;
Uzun süreli görevlerde context giderek bozulabilir.&lt;/p&gt;

&lt;p&gt;
Bu nedenle modelin etrafında onu yöneten bir çalışma ortamına ihtiyaç duyuluyor.
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f5f7ff; padding:14px 18px; margin-bottom:8px;"&gt;&#128221; Planı korur.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f5f7ff; padding:14px 18px; margin-bottom:8px;"&gt;&#128260; Gerektiğinde yeniden planlama yaptırır.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f5f7ff; padding:14px 18px; margin-bottom:8px;"&gt;&#129514; Sonuçları test eder.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f5f7ff; padding:14px 18px; margin-bottom:8px;"&gt;&#128269; Eksikleri tekrar kontrol ettirir.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f5f7ff; padding:14px 18px; margin-bottom:8px;"&gt;&#128218; Context yönetimini gerçekleştirir.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f5f7ff; padding:14px 18px; margin-bottom:8px;"&gt;&#128200; Çalışmanın ilerlemesini izler.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f5f7ff; padding:14px 18px; margin-bottom:8px;"&gt;✅ Sonucu bağımsız olarak doğrulatabilir.&lt;/div&gt;
&lt;div style="width:100%; box-sizing:border-box; background:#f5f7ff; padding:14px 18px; margin-bottom:22px;"&gt;&#128721; Agent’ın ne zaman duracağına karar verilmesine yardımcı olur.&lt;/div&gt;

&lt;p&gt;
Böylece birkaç dakika boyunca çalışan chatbot yerine saatlerce hatta uygun altyapılarda çok daha uzun süre bir hedef üzerinde çalışabilen sistemlere doğru ilerliyoruz.
&lt;/p&gt;


&lt;div style="width:100%; box-sizing:border-box; background:#18243e; color:#fff; padding:28px; margin:38px 0; border-radius:12px;"&gt;

  &lt;div style="font-size:13px; letter-spacing:1px; color:#bac4d8; margin-bottom:12px;"&gt;
    BENCE ASIL KIRILMA BURADA
  &lt;/div&gt;

  &lt;div style="font-size:25px; line-height:1.4; margin-bottom:14px;"&gt;
    Yarış artık yalnızca “en iyi LLM kimde?” yarışı değil.
  &lt;/div&gt;

  &lt;div style="font-size:17px; line-height:1.7; color:#e1e6ef;"&gt;
    Yarış; modeli doğru veri, doğru araçlar, doğru bellek, doğru beceriler,
    doğru güvenlik mekanizmaları ve doğru doğrulama sistemi içerisinde ne kadar başarılı çalıştırabildiğinizle ilgili.
  &lt;/div&gt;

&lt;/div&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
Yapay zekânın kısa tarihi aslında nasıl değişti?
&lt;/h2&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:15px 18px; margin-bottom:8px;"&gt;
&#129504; 2010’lar → Model örüntüleri tanıyor.
&lt;/div&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:15px 18px; margin-bottom:8px;"&gt;
&#128172; 2022 → Model bizimle konuşuyor.
&lt;/div&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:15px 18px; margin-bottom:8px;"&gt;
&#128064; 2023–2024 → Model görüyor ve duyuyor.
&lt;/div&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:15px 18px; margin-bottom:8px;"&gt;
&#129504; 2024 → Model problem üzerinde daha uzun düşünüyor.
&lt;/div&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:15px 18px; margin-bottom:8px;"&gt;
&#128295; 2023–2025 → Model araç kullanıyor.
&lt;/div&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:15px 18px; margin-bottom:8px;"&gt;
&#128421;️ 2024–2025 → Model bilgisayar ve tarayıcı kullanıyor.
&lt;/div&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:15px 18px; margin-bottom:8px;"&gt;
&#129302; 2025–2026 → Model agent’a dönüşüyor.
&lt;/div&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f7f8fa; padding:15px 18px; margin-bottom:24px;"&gt;
&#127959;️ 2026 → Agent uzun süre çalışabileceği bir sistem içerisine yerleşiyor.
&lt;/div&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:45px;"&gt;
&#128302; Peki 2027 neyin yılı olacak?
&lt;/h2&gt;

&lt;p&gt;
Bence bir sonraki büyük sıçrama yalnızca yeni ve daha güçlü bir sohbet modeli olmayabilir.
&lt;/p&gt;

&lt;p&gt;
Benim güçlü adaylarımdan biri &lt;em&gt;persistent agent&lt;/em&gt;, yani uzun süre çalışan, durumunu ve bağlamını koruyabilen kişisel ve kurumsal yapay zekâ sistemleri.
&lt;/p&gt;

&lt;p&gt;
Bugünkü chatbot yaklaşımında biz yapay zekâya gideriz.
&lt;/p&gt;

&lt;p&gt;
Bir pencere açarız, sorumuzu sorarız ve cevabı aldıktan sonra çalışmaya devam ederiz.
&lt;/p&gt;

&lt;p&gt;
Bir sonraki aşamada ise yapay zekâ çalışma ortamımızın sürekli bir parçası olabilir.
&lt;/p&gt;

&lt;p&gt;&#129504; Geçmiş çalışmalarımızı gerektiğinde hatırlayabilir.&lt;/p&gt;
&lt;p&gt;&#128231; E-posta ve mesajlarla çalışabilir.&lt;/p&gt;
&lt;p&gt;&#128197; Takvim ve görevleri takip edebilir.&lt;/p&gt;
&lt;p&gt;&#128193; Dosya ve projeleri anlayabilir.&lt;/p&gt;
&lt;p&gt;&#127760; Gerektiğinde araştırma yapabilir.&lt;/p&gt;
&lt;p&gt;&#128187; Bilgisayarda işlem gerçekleştirebilir.&lt;/p&gt;
&lt;p&gt;&#129513; İhtiyacı olan becerileri kullanabilir.&lt;/p&gt;
&lt;p&gt;&#128268; Kurumsal sistemlere bağlanabilir.&lt;/p&gt;
&lt;p&gt;&#129309; Başka uzman agent’larla çalışabilir.&lt;/p&gt;
&lt;p&gt;&#128737;️ Yetki sınırlarını dikkate alabilir.&lt;/p&gt;
&lt;p&gt;&#128100; Kritik işlemlerde insan onayı isteyebilir.&lt;/p&gt;
&lt;p&gt;✅ Kendi yaptığı işi doğrulayabilir.&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
☁️ Bulut mu, lokal mi?
&lt;/h2&gt;

&lt;p&gt;
Bunun tamamen bulutta çalışan bir yapı olacağını da düşünmüyorum.
&lt;/p&gt;

&lt;p&gt;
Kişisel ve kurumsal verinin önemi arttıkça hibrit mimarilerin daha anlamlı hale gelmesi muhtemel.
&lt;/p&gt;

&lt;p&gt;
Küçük ve hızlı modeller lokal olarak sürekli çalışabilir.
&lt;/p&gt;

&lt;p&gt;
Daha karmaşık reasoning gerektiğinde güçlü bulut modellerinden destek alınabilir.
&lt;/p&gt;

&lt;p&gt;
Kuruma özel bilgi kurum içerisinde tutulabilir.
&lt;/p&gt;

&lt;p&gt;
Son derece hassas görevler tamamen lokal modellerle yürütülebilir.
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#f4faf8; border-left:5px solid #15956c; padding:20px 22px; margin:25px 0; font-size:20px;"&gt;
  Gelecekte “hangi modeli kullanıyorsunuz?” sorusu yerine “yapay zekâ sisteminiz nasıl bir mimari üzerinde çalışıyor?” sorusunu daha fazla sorabiliriz.
&lt;/div&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:42px;"&gt;
&#129302; Bir başka güçlü aday: Physical AI
&lt;/h2&gt;

&lt;p&gt;
Bir başka önemli gelişme alanı ise yapay zekânın bilgisayar ekranından fiziksel dünyaya çıkması.
&lt;/p&gt;

&lt;p&gt;
Vision, reasoning, planning ve agent teknolojileri robotik sistemlerle birleştiğinde embodied AI veya physical AI alanında büyük gelişmeler görebiliriz.
&lt;/p&gt;

&lt;p&gt;
Bugün bilgisayar ekranındaki bir butona basmayı öğrenen agent’ın yarın fiziksel dünyadaki bir robotu yönlendirmesi aslında birbirinden tamamen bağımsız iki problem değil.
&lt;/p&gt;


&lt;h2 style="font-size:27px; color:#18243e; margin-top:45px;"&gt;
&#128640; Belki de devrimin en önemli kısmı daha yeni başlıyor
&lt;/h2&gt;

&lt;p&gt;
Son 15 yıldaki gelişime baktığımda benim için en ilginç değişim sorularımızın değişmesi.
&lt;/p&gt;

&lt;p&gt;
Deep Learning döneminde:
&lt;/p&gt;

&lt;p&gt;
“Makine bunu tanıyabilir mi?”
&lt;/p&gt;

&lt;p&gt;
diye soruyorduk.
&lt;/p&gt;

&lt;p&gt;
LLM döneminde:
&lt;/p&gt;

&lt;p&gt;
“Makine bunu anlayıp anlatabilir mi?”
&lt;/p&gt;

&lt;p&gt;
sorusuna geçtik.
&lt;/p&gt;

&lt;p&gt;
Reasoning modellerle:
&lt;/p&gt;

&lt;p&gt;
“Makine bu problemi çözebilir mi?”
&lt;/p&gt;

&lt;p&gt;
diye sormaya başladık.
&lt;/p&gt;

&lt;p&gt;
Agent döneminde ise soru çok daha farklı:
&lt;/p&gt;

&lt;div style="width:100%; box-sizing:border-box; background:#18243e; color:white; text-align:center; padding:28px 20px; margin:30px 0; border-radius:12px; font-size:25px; line-height:1.4;"&gt;
  “Bu işi baştan sona yapay zekâya verebilir miyim?”
&lt;/div&gt;

&lt;p&gt;
Bence bu küçük bir değişim değil.
&lt;/p&gt;

&lt;p&gt;
Bilgisayar tarihinin büyük bölümünde insanlar bilgisayarları kullanmayı öğrendi.
&lt;/p&gt;

&lt;p&gt;
Şimdi bilgisayarlar bizim kullandığımız araçları kullanmayı öğrenmeye başlıyor.
&lt;/p&gt;

&lt;p&gt;
Belki de önümüzdeki dönemin asıl dönüşümü burada yaşanacak.
&lt;/p&gt;


&lt;div style="width:100%; box-sizing:border-box; background:#f3f6fb; border:1px solid #dfe5ee; border-radius:12px; padding:25px; margin-top:38px;"&gt;

  &lt;div style="font-size:24px; color:#18243e; margin-bottom:15px;"&gt;
    &#129300; Sizin tahmininiz nedir?
  &lt;/div&gt;

  &lt;p&gt;
    2027’de yapay zekâda hangi teknoloji veya yaklaşımın öne çıkacağını düşünüyorsunuz?
  &lt;/p&gt;

  &lt;p&gt;&#129302; Kişisel ve sürekli çalışan AI agent’lar mı?&lt;/p&gt;
  &lt;p&gt;&#127970; Kurumsal agent sistemleri mi?&lt;/p&gt;
  &lt;p&gt;&#127959;️ Agent harness ve uzun süreli otonom çalışma mı?&lt;/p&gt;
  &lt;p&gt;&#129504; Daha güçlü reasoning modeller mi?&lt;/p&gt;
  &lt;p&gt;&#128421;️ Computer use ve browser use mı?&lt;/p&gt;
  &lt;p&gt;&#127757; Açık ve lokal çalışan modeller mi?&lt;/p&gt;
  &lt;p&gt;&#129302; Physical AI ve robotik mi?&lt;/p&gt;

  &lt;p style="margin-bottom:0;"&gt;
    Yoksa bugün henüz yeterince konuşmadığımız tamamen farklı bir kırılma mı geliyor?
  &lt;/p&gt;

&lt;/div&gt;


&lt;h3 style="font-size:21px; color:#374151; margin-top:45px;"&gt;
Seçilmiş kilometre taşları
&lt;/h3&gt;

&lt;div style="font-size:14px; color:#626b77; line-height:1.65;"&gt;

&lt;p&gt;Krizhevsky, Sutskever &amp;amp; Hinton — ImageNet Classification with Deep Convolutional Neural Networks, 2012&lt;/p&gt;

&lt;p&gt;Sutskever, Vinyals &amp;amp; Le — Sequence to Sequence Learning with Neural Networks, 2014&lt;/p&gt;

&lt;p&gt;Bahdanau, Cho &amp;amp; Bengio — Neural Machine Translation by Jointly Learning to Align and Translate, 2014&lt;/p&gt;

&lt;p&gt;Vaswani et al. — Attention Is All You Need, 2017&lt;/p&gt;

&lt;p&gt;Devlin et al. — BERT, 2018&lt;/p&gt;

&lt;p&gt;Kaplan et al. — Scaling Laws for Neural Language Models, 2020&lt;/p&gt;

&lt;p&gt;Brown et al. — Language Models are Few-Shot Learners, 2020&lt;/p&gt;

&lt;p&gt;Lewis et al. — Retrieval-Augmented Generation, 2020&lt;/p&gt;

&lt;p&gt;Ho, Jain &amp;amp; Abbeel — Denoising Diffusion Probabilistic Models, 2020&lt;/p&gt;

&lt;/div&gt;


&lt;div style="width:100%; box-sizing:border-box; border-top:1px solid #e1e5ea; margin-top:35px; padding-top:22px; color:#315caa; font-size:15px; line-height:1.8;"&gt;
#YapayZeka &amp;nbsp; #ArtificialIntelligence &amp;nbsp; #GenerativeAI &amp;nbsp; #LLM &amp;nbsp; #AIAgents &amp;nbsp; #AgenticAI &amp;nbsp; #DeepLearning &amp;nbsp; #Transformer &amp;nbsp; #ReasoningModels &amp;nbsp; #RAG &amp;nbsp; #MCP &amp;nbsp; #AgentSkills &amp;nbsp; #ComputerUse &amp;nbsp; #OpenSourceAI &amp;nbsp; #LocalAI &amp;nbsp; #AIEngineering &amp;nbsp; #FutureOfAI
&lt;/div&gt;

&lt;/div&gt;</description><link>https://www.muratkarakaya.net/2026/09/yapay-zekann-yolculugu.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-3064609125188112076</guid><pubDate>Sat, 26 Sep 2026 19:46:18 +0000</pubDate><atom:updated>2026-09-26T22:46:18.340+03:00</atom:updated><category domain="http://www.blogger.com/atom/ns#">Ar-Ge</category><category domain="http://www.blogger.com/atom/ns#">Proje Yönetimi</category><category domain="http://www.blogger.com/atom/ns#">Yapay Zeka</category><title> Yapay Zekâ • Ar-Ge • Proje Yönetimi</title><description>&lt;div style="max-width: 920px; margin: 0 auto; font-family: Arial, Helvetica, sans-serif; color: #222; line-height: 1.75; font-size: 17px;"&gt;

  &lt;!-- HEADER --&gt;

  &lt;div style="padding: 36px 32px; margin-bottom: 32px; background: linear-gradient(135deg, #0f172a, #1e3a5f); color: white; border-radius: 16px;"&gt;
    &lt;div style="font-size: 14px; letter-spacing: 1.5px; text-transform: uppercase; opacity: 0.8; margin-bottom: 12px;"&gt;
      Yapay Zekâ • Ar-Ge • Proje Yönetimi
    &lt;/div&gt;

```
&lt;h1 style="font-size: 38px; line-height: 1.2; margin: 0 0 18px 0; color: white;"&gt;
  &#128680; Yapay Zekâ Projeleri Neden Batıyor?
&lt;/h1&gt;

&lt;p style="font-size: 21px; line-height: 1.55; margin: 0; color: #e2e8f0;"&gt;
  Proje dosyaları giderek daha etkileyici hale geliyor. Peki aynı şeyi bu projeleri hayata geçirecek ekiplerin mühendislik kapasitesi için de söyleyebilir miyiz?
&lt;/p&gt;
```

  &lt;/div&gt;

  &lt;!-- INTRO --&gt;

  &lt;p&gt;
    Son zamanlarda proje değerlendirmelerinde giderek daha fazla dikkatimi çeken bir durum var.
  &lt;/p&gt;

  &lt;p&gt;
    Bir proje dosyasını okumaya başladığınızda, özellikle yapay zekâ alanındaki neredeyse bütün güncel kavramların proje içerisine yerleştirildiğini görebiliyorsunuz:
  &lt;/p&gt;

  &lt;div style="background-color: #f8fafc; border-left: 5px solid #2563eb; padding: 20px 24px; margin: 28px 0; border-radius: 6px;"&gt;
    Büyük dil modelleri, RAG, yapay zekâ ajanları, vektör veri tabanları, kişiselleştirme, ince ayar, çoklu ajan mimarileri, gerçek zamanlı analiz, tahminleme, karar destek sistemleri, açıklanabilir yapay zekâ...
  &lt;/div&gt;

  &lt;p&gt;
    Kâğıt üzerinde proje gerçekten etkileyici görünüyor.
  &lt;/p&gt;

  &lt;p&gt;
    Fakat ekip ile yüz yüze görüşmeye geçip biraz daha ayrıntılı sorular sormaya başladığınızda, çoğu zaman bambaşka bir tablo ortaya çıkıyor.
  &lt;/p&gt;

  &lt;!-- SECTION --&gt;

  &lt;h2 style="font-size: 28px; color: #0f172a; margin-top: 48px; padding-bottom: 10px; border-bottom: 2px solid #e2e8f0;"&gt;
    &#128101; Projede teknoloji var, ama tecrübe var mı?
  &lt;/h2&gt;

  &lt;p&gt;
    Bazen projede kullanılacağı belirtilen teknolojilerle ekipteki mühendislerin daha önce küçük ölçekte bile bir uygulama geliştirmediğini görüyorsunuz.
  &lt;/p&gt;

  &lt;p&gt;
    Firmanın da benzer yöntemlerle gerçek bir ürün geliştirme, canlıya alma, kullanıcıya sunma ve operasyon altında çalıştırma konusunda herhangi bir deneyimi bulunmayabiliyor.
  &lt;/p&gt;

  &lt;p&gt;
    Bir süre sonra şu sorular ortaya çıkıyor:
  &lt;/p&gt;

  &lt;ul style="padding-left: 26px;"&gt;
    &lt;li style="margin-bottom: 10px;"&gt;Bu teknoloji neden seçildi?&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;Projede hangi gerçek problemi çözüyor?&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;Diğer bileşenlerle neden birlikte kullanılıyor?&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;Alternatifleri değerlendirildi mi?&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;Başarılı olup olmadığı nasıl ölçülecek?&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;Çalışmadığında hangi bileşenin neden çalışmadığı nasıl anlaşılacak?&lt;/li&gt;
  &lt;/ul&gt;

  &lt;p&gt;
    Ne yazık ki bazen bu soruların net cevapları bulunmuyor.
  &lt;/p&gt;

  &lt;p&gt;
    Çünkü proje dokümanında yer alan teknoloji seçimi, gerçek bir mühendislik analizinin sonucu olmaktan çok, güncel kavramların birbirine bağlanmasıyla oluşturulmuş gibi duruyor.
  &lt;/p&gt;

  &lt;!-- HIGHLIGHT --&gt;

  &lt;div style="background-color: #fff7ed; border: 1px solid #fed7aa; padding: 26px; margin: 34px 0; border-radius: 12px;"&gt;
    &lt;div style="font-size: 21px; line-height: 1.6; color: #9a3412;"&gt;
      &#128161; Sorun, projede çok fazla teknoloji kullanılması değil. Sorun, kullanılan her teknolojinin neden gerekli olduğunun mühendislik açısından açıklanamaması.
    &lt;/div&gt;
  &lt;/div&gt;

  &lt;!-- SECTION --&gt;

  &lt;h2 style="font-size: 28px; color: #0f172a; margin-top: 48px; padding-bottom: 10px; border-bottom: 2px solid #e2e8f0;"&gt;
    &#129513; Teknoloji listesi, sistem tasarımı değildir
  &lt;/h2&gt;

  &lt;p&gt;
    Günümüzde yapay zekâ araçlarıyla oldukça etkileyici proje metinleri üretmek mümkün.
  &lt;/p&gt;

  &lt;p&gt;
    Her teknolojinin avantajları sıralanabilir. Kullanılan yöntemlerin literatürdeki faydaları anlatılabilir. Mimari birkaç kutu ve okla gösterilebilir.
  &lt;/p&gt;

  &lt;p&gt;
    Fakat bunların hiçbiri tek başına sistem mühendisliği anlamına gelmiyor.
  &lt;/p&gt;

  &lt;p&gt;
    Gerçek bir sistem tasarımında her bileşenin bir gerekçesi vardır.
  &lt;/p&gt;

  &lt;p&gt;
    Bir bileşen, moda olduğu için değil, belirli bir ihtiyacı karşıladığı için sisteme girer.
  &lt;/p&gt;

  &lt;p&gt;
    Seçilen yöntemin:
  &lt;/p&gt;

  &lt;ul style="padding-left: 26px;"&gt;
    &lt;li style="margin-bottom: 10px;"&gt;hangi gereksinimi karşıladığı,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;hangi alternatife göre tercih edildiği,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;hangi metriği iyileştirmesi beklendiği,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;hangi maliyeti oluşturduğu,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;hangi yeni riski sisteme eklediği&lt;/li&gt;
  &lt;/ul&gt;

  &lt;p&gt;
    açıklanabilmelidir.
  &lt;/p&gt;

  &lt;p&gt;
    Aksi halde elimizde bir sistem mimarisi değil, yalnızca güncel teknolojilerin yer aldığı bir liste vardır.
  &lt;/p&gt;

  &lt;!-- SECTION --&gt;

  &lt;h2 style="font-size: 28px; color: #0f172a; margin-top: 48px; padding-bottom: 10px; border-bottom: 2px solid #e2e8f0;"&gt;
    &#128269; RAG demek kolay; çalışan bir RAG sistemi geliştirmek zor
  &lt;/h2&gt;

  &lt;p&gt;
    Örneğin proje dokümanında şu cümleyi görmek artık oldukça sıradan:
  &lt;/p&gt;

  &lt;div style="background-color: #f1f5f9; padding: 22px 26px; margin: 26px 0; border-radius: 10px; font-style: italic;"&gt;
    “Kurumsal dokümanlardan güvenilir cevaplar üreten RAG tabanlı yapay zekâ sistemi geliştirilecektir.”
  &lt;/div&gt;

  &lt;p&gt;
    Güzel.
  &lt;/p&gt;

  &lt;p&gt;
    Fakat gerçek mühendislik problemi tam da bu cümlenin bittiği yerde başlıyor.
  &lt;/p&gt;

  &lt;div style="background-color: #f8fafc; padding: 26px; border-radius: 12px; margin: 28px 0;"&gt;
    &lt;p style="margin-top: 0;"&gt;&#128204; Sistemin başarısı nasıl ölçülecek?&lt;/p&gt;
    &lt;p&gt;&#128204; Benchmark veri seti nasıl hazırlanacak?&lt;/p&gt;
    &lt;p&gt;&#128204; Kullanıcıların gerçekten soracağı sorular nasıl temsil edilecek?&lt;/p&gt;
    &lt;p&gt;&#128204; Retrieval başarısı ile nihai cevap başarısı birbirinden nasıl ayrılacak?&lt;/p&gt;
    &lt;p&gt;&#128204; Chunk büyüklüğü ve overlap değerleri nasıl seçilecek?&lt;/p&gt;
    &lt;p&gt;&#128204; Embedding modeli hangi deney sonucuna göre belirlenecek?&lt;/p&gt;
    &lt;p&gt;&#128204; On binlerce dokümandaki parçalar hangi metadata alanlarıyla birbirinden ayrılacak?&lt;/p&gt;
    &lt;p&gt;&#128204; Aynı belgenin farklı sürümleri varsa hangisi kullanılacak?&lt;/p&gt;
    &lt;p&gt;&#128204; Yürürlükten kalkmış bilgi nasıl sistem dışına çıkarılacak?&lt;/p&gt;
    &lt;p&gt;&#128204; Yetkisiz kullanıcının erişmemesi gereken içerik retrieval aşamasında nasıl engellenecek?&lt;/p&gt;
    &lt;p&gt;&#128204; Hallucination nasıl ölçülecek?&lt;/p&gt;
    &lt;p&gt;&#128204; Cevap üretilemediğinde sistem nasıl davranacak?&lt;/p&gt;
    &lt;p&gt;&#128204; Binlerce eşzamanlı kullanıcı olduğunda gecikme süresi nasıl kontrol altında tutulacak?&lt;/p&gt;
    &lt;p style="margin-bottom: 0;"&gt;&#128204; Maliyet, doğruluk ve cevap süresi arasındaki denge nasıl kurulacak?&lt;/p&gt;
  &lt;/div&gt;

  &lt;p&gt;
    Bunların her biri gerçek hayatta karşınıza çıkacak mühendislik problemidir.
  &lt;/p&gt;

  &lt;p&gt;
    Fakat bazı proje dosyalarında bu sorulara cevap vermek yerine, RAG teknolojisinin genel faydalarının anlatıldığını görüyoruz.
  &lt;/p&gt;

  &lt;p&gt;
    Bu da bize projenin teorik olarak tanımlandığını, ancak operasyonel gerçekliğinin yeterince düşünülmediğini gösteriyor.
  &lt;/p&gt;

  &lt;!-- SECTION --&gt;

  &lt;h2 style="font-size: 28px; color: #0f172a; margin-top: 48px; padding-bottom: 10px; border-bottom: 2px solid #e2e8f0;"&gt;
    ⚠️ Yapay zekâ ile proje hazırlamak başka, yapay zekâ projesi geliştirmek başka
  &lt;/h2&gt;

  &lt;p&gt;
    Burada son yıllarda ortaya çıkan yeni bir risk olduğunu düşünüyorum.
  &lt;/p&gt;

  &lt;p&gt;
    Yapay zekâ araçları proje hazırlama sürecini olağanüstü kolaylaştırdı.
  &lt;/p&gt;

  &lt;p&gt;
    Daha önce haftalarca sürebilecek literatür taramaları, teknoloji açıklamaları, yöntem bölümleri ve teknik metinler artık çok daha kısa sürede hazırlanabiliyor.
  &lt;/p&gt;

  &lt;p&gt;
    Bu kendi başına kötü bir gelişme değil.
  &lt;/p&gt;

  &lt;p&gt;
    Tam tersine, doğru kullanıldığında ciddi bir verimlilik sağlıyor.
  &lt;/p&gt;

  &lt;p&gt;
    Problem şu noktada başlıyor:
  &lt;/p&gt;

  &lt;div style="background: linear-gradient(135deg, #fff1f2, #fff7ed); border: 1px solid #fecdd3; padding: 28px; margin: 30px 0; border-radius: 12px;"&gt;
    &lt;p style="font-size: 19px; margin: 0 0 14px 0;"&gt;&#129302; Yapay zekâya proje dosyasını yazdır.&lt;/p&gt;
    &lt;p style="font-size: 19px; margin: 0 0 14px 0;"&gt;&#128176; Destek başvurusunu yap.&lt;/p&gt;
    &lt;p style="font-size: 19px; margin: 0;"&gt;&#129489;‍&#128187; Destek çıkarsa kodu da yapay zekâya yazdır.&lt;/p&gt;
  &lt;/div&gt;

  &lt;p&gt;
    Bu yaklaşım giderek tehlikeli bir kolaycılığa dönüşebiliyor.
  &lt;/p&gt;

  &lt;p&gt;
    Çünkü proje dokümanı ile gerçek uygulama arasındaki mühendislik boşluğu, yalnızca daha fazla yapay zekâ kullanılarak otomatik olarak kapatılamıyor.
  &lt;/p&gt;

  &lt;!-- SECTION --&gt;

  &lt;h2 style="font-size: 28px; color: #0f172a; margin-top: 48px; padding-bottom: 10px; border-bottom: 2px solid #e2e8f0;"&gt;
    &#129504; Yapay zekâ, olmayan tecrübeyi otomatik olarak üretmiyor
  &lt;/h2&gt;

  &lt;p&gt;
    Bugünkü üretken yapay zekâ sistemleri kod yazabiliyor, mimari önerebiliyor, API oluşturabiliyor, test senaryosu hazırlayabiliyor ve dokümantasyon üretebiliyor.
  &lt;/p&gt;

  &lt;p&gt;
    Ancak hâlâ çok kritik bir fark var:
  &lt;/p&gt;

  &lt;div style="background-color: #ecfeff; border-left: 5px solid #0891b2; padding: 22px 26px; margin: 28px 0; border-radius: 6px;"&gt;
    Yapay zekâ size bir çözüm üretebilir; fakat o çözümün sizin probleminiz için gerçekten doğru çözüm olup olmadığını değerlendirebilmek için yine bilgiye ve tecrübeye ihtiyacınız vardır.
  &lt;/div&gt;

  &lt;p&gt;
    Çünkü gerçek mühendislik yalnızca kod üretmek değildir.
  &lt;/p&gt;

  &lt;p&gt;
    Gerçek mühendislik:
  &lt;/p&gt;

  &lt;ul style="padding-left: 26px;"&gt;
    &lt;li style="margin-bottom: 10px;"&gt;doğru problemi tanımlamak,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;gereksinimleri ortaya çıkarmak,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;alternatifleri karşılaştırmak,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;ölçülebilir başarı kriterleri belirlemek,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;darboğazları önceden görmek,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;beklenmeyen durumları yönetmek,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;sistemi gerçek kullanıcı davranışı altında test etmek,&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;ölçeklenebilirlik ve sürdürülebilirliği düşünmek&lt;/li&gt;
  &lt;/ul&gt;

  &lt;p&gt;
    demektir.
  &lt;/p&gt;

  &lt;!-- SECTION --&gt;

  &lt;h2 style="font-size: 28px; color: #0f172a; margin-top: 48px; padding-bottom: 10px; border-bottom: 2px solid #e2e8f0;"&gt;
    &#128201; Sonuçta ne oluyor?
  &lt;/h2&gt;

  &lt;p&gt;
    Proje dosyasında oldukça güçlü görünen sistem, uygulama aşamasına geldiğinde beklenen performansı vermiyor.
  &lt;/p&gt;

  &lt;p&gt;
    Başlangıçta birkaç demo hazırlanıyor.
  &lt;/p&gt;

  &lt;p&gt;
    Kontrollü örneklerde sistem güzel çalışıyor.
  &lt;/p&gt;

  &lt;p&gt;
    Fakat gerçek veri, gerçek kullanıcı ve gerçek operasyon devreye girdiğinde problemler ortaya çıkmaya başlıyor.
  &lt;/p&gt;

  &lt;p&gt;
    Maliyet yükseliyor.
  &lt;/p&gt;

  &lt;p&gt;
    Gecikme süreleri artıyor.
  &lt;/p&gt;

  &lt;p&gt;
    Veri kalitesi problemleri görülüyor.
  &lt;/p&gt;

  &lt;p&gt;
    Yetkilendirme konusu ortaya çıkıyor.
  &lt;/p&gt;

  &lt;p&gt;
    Model beklenmeyen cevaplar üretmeye başlıyor.
  &lt;/p&gt;

  &lt;p&gt;
    Kullanıcıların sorduğu sorular, demo sırasında hazırlanmış örneklere benzemiyor.
  &lt;/p&gt;

  &lt;p&gt;
    Sistemin başarısını gösterecek düzgün bir benchmark olmadığı için hangi değişikliğin sistemi iyileştirdiği bile anlaşılamıyor.
  &lt;/p&gt;

  &lt;p&gt;
    Zaman geçtikçe proje takvimi kayıyor.
  &lt;/p&gt;

  &lt;p&gt;
    Ekip değişiyor.
  &lt;/p&gt;

  &lt;p&gt;
    Çalışanlar ayrılıyor.
  &lt;/p&gt;

  &lt;p&gt;
    Yeni ekip eski kararların neden alındığını anlamıyor.
  &lt;/p&gt;

  &lt;p&gt;
    Sonunda proje ya küçültülüyor, ya amacından uzaklaşıyor ya da kullanılmayan bir prototip olarak kalıyor.
  &lt;/p&gt;

  &lt;!-- SECTION --&gt;

  &lt;h2 style="font-size: 28px; color: #0f172a; margin-top: 48px; padding-bottom: 10px; border-bottom: 2px solid #e2e8f0;"&gt;
    &#127919; Belki de yanlış soruyu soruyoruz
  &lt;/h2&gt;

  &lt;p&gt;
    Yapay zekâ projelerini değerlendirirken çoğu zaman şu soruya odaklanıyoruz:
  &lt;/p&gt;

  &lt;div style="text-align: center; font-size: 23px; font-style: italic; color: #1e3a8a; margin: 32px 0;"&gt;
    “Projede hangi teknolojiler kullanılacak?”
  &lt;/div&gt;

  &lt;p&gt;
    Bence bundan daha önemli bir soru var:
  &lt;/p&gt;

  &lt;div style="text-align: center; font-size: 24px; line-height: 1.5; color: #9a3412; margin: 32px 0;"&gt;
    “Bu ekip, yazdığı sistemi gerçekten geliştirebilecek mühendislik kapasitesine sahip mi?”
  &lt;/div&gt;

  &lt;p&gt;
    Hatta bunun devamında şu sorular da sorulmalı:
  &lt;/p&gt;

  &lt;ul style="padding-left: 26px;"&gt;
    &lt;li style="margin-bottom: 10px;"&gt;Ekip benzer bir problemi daha önce çözmüş mü?&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;En azından küçük ölçekli bir prototip geliştirmiş mi?&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;Teknik risklerin farkında mı?&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;Hangi varsayımların yanlış çıkabileceğini biliyor mu?&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;Başarıyı ölçmek için deney tasarlayabiliyor mu?&lt;/li&gt;
    &lt;li style="margin-bottom: 10px;"&gt;Sistemi gerçek kullanıcı yükü altında nasıl doğrulayacağını biliyor mu?&lt;/li&gt;
  &lt;/ul&gt;

  &lt;!-- SECTION --&gt;

  &lt;h2 style="font-size: 28px; color: #0f172a; margin-top: 48px; padding-bottom: 10px; border-bottom: 2px solid #e2e8f0;"&gt;
    &#127959;️ İyi bir yapay zekâ projesi nasıl başlamalı?
  &lt;/h2&gt;

  &lt;p&gt;
    Bana göre süreç teknoloji isimleriyle değil, problemle başlamalı.
  &lt;/p&gt;

  &lt;div style="overflow-x: auto; margin: 30px 0;"&gt;
    &lt;table style="width: 100%; border-collapse: collapse; font-size: 16px;"&gt;
      &lt;thead&gt;
        &lt;tr style="background-color: #0f172a; color: white;"&gt;
          &lt;th style="padding: 14px; border: 1px solid #334155; text-align: left;"&gt;Yanlış başlangıç&lt;/th&gt;
          &lt;th style="padding: 14px; border: 1px solid #334155; text-align: left;"&gt;Daha sağlıklı başlangıç&lt;/th&gt;
        &lt;/tr&gt;
      &lt;/thead&gt;
      &lt;tbody&gt;
        &lt;tr&gt;
          &lt;td style="padding: 14px; border: 1px solid #e2e8f0;"&gt;“Projede RAG kullanalım.”&lt;/td&gt;
          &lt;td style="padding: 14px; border: 1px solid #e2e8f0;"&gt;“Kullanıcının hangi bilgiye neden erişemediğini anlayalım.”&lt;/td&gt;
        &lt;/tr&gt;
        &lt;tr style="background-color: #f8fafc;"&gt;
          &lt;td style="padding: 14px; border: 1px solid #e2e8f0;"&gt;“Agent ekleyelim.”&lt;/td&gt;
          &lt;td style="padding: 14px; border: 1px solid #e2e8f0;"&gt;“Hangi görevlerin gerçekten çok adımlı karar gerektirdiğini belirleyelim.”&lt;/td&gt;
        &lt;/tr&gt;
        &lt;tr&gt;
          &lt;td style="padding: 14px; border: 1px solid #e2e8f0;"&gt;“LLM ile kişiselleştirelim.”&lt;/td&gt;
          &lt;td style="padding: 14px; border: 1px solid #e2e8f0;"&gt;“Hangi kullanıcı bilgisinin gerçekten cevap kalitesini artırdığını ölçelim.”&lt;/td&gt;
        &lt;/tr&gt;
        &lt;tr style="background-color: #f8fafc;"&gt;
          &lt;td style="padding: 14px; border: 1px solid #e2e8f0;"&gt;“En güçlü modeli kullanalım.”&lt;/td&gt;
          &lt;td style="padding: 14px; border: 1px solid #e2e8f0;"&gt;“Kalite, maliyet ve gecikme arasındaki optimum noktayı deneyle bulalım.”&lt;/td&gt;
        &lt;/tr&gt;
      &lt;/tbody&gt;
    &lt;/table&gt;
  &lt;/div&gt;

  &lt;!-- CONCLUSION --&gt;

  &lt;h2 style="font-size: 28px; color: #0f172a; margin-top: 48px; padding-bottom: 10px; border-bottom: 2px solid #e2e8f0;"&gt;
    &#128640; Sonuç: Güçlü proje dosyası, güçlü proje demek değildir
  &lt;/h2&gt;

  &lt;p&gt;
    Yapay zekâ ile bugün çok iyi bir proje dosyası hazırlamak mümkün.
  &lt;/p&gt;

  &lt;p&gt;
    Hatta belki birkaç yıl öncesine kıyasla çok daha profesyonel, çok daha teknik ve çok daha ikna edici dokümanlar hazırlayabiliyoruz.
  &lt;/p&gt;

  &lt;p&gt;
    Fakat proje dosyasının kalitesi ile projenin uygulanabilirliği aynı şey değil.
  &lt;/p&gt;

  &lt;p&gt;
    Yapay zekâ, proje hazırlama sürecini hızlandırabilir.
  &lt;/p&gt;

  &lt;p&gt;
    Kod geliştirmeyi hızlandırabilir.
  &lt;/p&gt;

  &lt;p&gt;
    Dokümantasyonu kolaylaştırabilir.
  &lt;/p&gt;

  &lt;p&gt;
    Test senaryoları üretebilir.
  &lt;/p&gt;

  &lt;p&gt;
    Mimari alternatifler önerebilir.
  &lt;/p&gt;

  &lt;p&gt;
    Fakat ekibin sahip olmadığı mühendislik tecrübesini, alan bilgisini, sistem düşüncesini ve ürün geliştirme kültürünü sihirli bir şekilde oluşturmaz.
  &lt;/p&gt;

  &lt;div style="background: linear-gradient(135deg, #0f172a, #1e293b); color: white; padding: 30px; margin: 36px 0; border-radius: 14px;"&gt;
    &lt;p style="font-size: 21px; line-height: 1.65; margin: 0; color: white;"&gt;
      &#127919; Yapay zekâ projelerindeki en büyük risklerden biri artık teknolojiye erişememek değil; teknolojiye erişmenin çok kolaylaşması nedeniyle, yeterli hazırlık ve tecrübe olmadan çok iddialı projelere başlanması olabilir.
    &lt;/p&gt;
  &lt;/div&gt;

  &lt;p&gt;
    Günün sonunda hâlâ problemi anlayan, doğru soruları soran, ölçüm yapabilen, hata noktalarını öngörebilen ve sistemi uçtan uca tasarlayabilen insanlara ihtiyacımız var.
  &lt;/p&gt;

  &lt;p style="font-size: 20px; margin-top: 34px;"&gt;
    &#128172; Siz de proje dokümanında çok güçlü görünen ancak uygulama aşamasında karşılığı olmayan yapay zekâ projeleriyle karşılaşıyor musunuz?
  &lt;/p&gt;

  &lt;!-- HASHTAGS --&gt;

  &lt;div style="margin-top: 48px; padding-top: 24px; border-top: 1px solid #e2e8f0; color: #475569; font-size: 15px; line-height: 1.8;"&gt;
  #MuratKarakayaAkademi  #YapayZekâ #ArtificialIntelligence #LLM #RAG #AgenticAI #ArGe #Mühendislik #SistemMühendisliği #ProjeYönetimi #ÜrünGeliştirme #YapayZekâProjeleri #GenerativeAI #Teknoloji #İnovasyon #TÜBİTAK #KOSGEB
  &lt;/div&gt;

&lt;/div&gt;
</description><link>https://www.muratkarakaya.net/2026/09/yapay-zeka-ar-ge-proje-yonetimi.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-5517825777233752643</guid><pubDate>Sat, 26 Sep 2026 19:43:09 +0000</pubDate><atom:updated>2026-09-26T22:43:09.941+03:00</atom:updated><category domain="http://www.blogger.com/atom/ns#">Ar-Ge</category><category domain="http://www.blogger.com/atom/ns#">Evren</category><category domain="http://www.blogger.com/atom/ns#">LLM</category><category domain="http://www.blogger.com/atom/ns#">MLOps</category><category domain="http://www.blogger.com/atom/ns#">Model Eğitimi</category><category domain="http://www.blogger.com/atom/ns#">Savunma Sanayi</category><category domain="http://www.blogger.com/atom/ns#">Türkçe</category><category domain="http://www.blogger.com/atom/ns#">Yapay Zekâ</category><category domain="http://www.blogger.com/atom/ns#">ücretsiz</category><title>EVREN: Türkiye'de Yapay Zekâ Geliştirmek İçin Yeni Bir İmkân</title><description>&lt;!-- Blogger yazı başlığı: EVREN: Türkiye'de Yapay Zekâ Geliştirmek İçin Yeni Bir İmkân --&gt;
&lt;!-- Arama açıklaması: EVREN ile model kullanımı, API erişimi ve model eğitimi. 1.000 krediyle başlangıç, kullanım örnekleri ve Murat Karakaya'dan uygulamalı video. --&gt;
&lt;!-- Blogger etiket önerileri: EVREN, Yapay Zekâ, LLM, Model Eğitimi, MLOps, Ar-Ge --&gt;


&lt;div lang="tr" style="font-family: inherit; font-size: 17px; line-height: 1.8; overflow-wrap: break-word;"&gt;


  &lt;p&gt;Bir yapay zekâ fikriniz var. Hazır bir modeli denemek, kendi verilerinizle bir model eğitmek veya geliştirdiğiniz uygulamaya dil modeli bağlamak istiyorsunuz. Ancak GPU erişimi, sunucu kurulumu ve bunların maliyeti daha başlangıçta önünüze çıkıyor. EVREN, bu ihtiyaçları ortak bir ortamda karşılamayı amaçlayan ve Türkiye&amp;#39;deki araştırmacıların, öğrencilerin ve geliştiricilerin yakından tanıması gerektiğini düşündüğüm bir platform.&lt;/p&gt;

  &lt;p&gt;Son canlı yayınımda EVREN&amp;#39;i tanıttım ve demosunu yaptım. Kullanım deneyimim sonrasında kişisel değerlendirmem şu: Oldukça olgunlaşmış bir platformla karşı karşıyayız. Bu yazıda EVREN&amp;#39;in sunduğu imkânları, kendi çalışmalarımızda nasıl değerlendirebileceğimizi ve Türkiye açısından neden stratejik önem taşıdığını paylaşmak istiyorum.&lt;/p&gt;

  &lt;p&gt;Platformu benimle birlikte adım adım incelemek isterseniz &lt;a href="https://youtube.com/live/Ojl5C6xvlVU" target="_blank" rel="noopener noreferrer"&gt;canlı yayın kaydını buradan izleyebilirsiniz&lt;/a&gt;. Videoyu yazının sonunda da bulabilirsiniz.&lt;/p&gt;

  &lt;/div&gt;&lt;a href="https://www.muratkarakaya.net/2026/09/evren-turkiyede-yapay-zeka-gelistirmek.html#more"&gt;&lt;/a&gt;</description><link>https://www.muratkarakaya.net/2026/09/evren-turkiyede-yapay-zeka-gelistirmek.html</link><author>noreply@blogger.com (Murat Karakaya)</author><media:thumbnail xmlns:media="http://search.yahoo.com/mrss/" height="72" url="https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEgm_qW6oSZz1a9D6G8T0skGqKauvEI9l8dKcWlp0nlhaSPpJ9M1qoemIm_Vv75qPuLnMuQtQV9g1YOfMx_uDYQv3Oz3BCBMQaRW8eb7zBFwZa6Y743rp1GPl83-w9USts6A48BNkFbltliEAD3alPENSqEfHmvNtjxFnLY3ITC7G3-F9II4raHdzeQYu5ep/s72-c/chrome_bLXrVrkA8v.png" width="72"/><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-7046448474143332191</guid><pubDate>Fri, 19 Jun 2026 20:12:26 +0000</pubDate><atom:updated>2026-06-19T23:12:26.352+03:00</atom:updated><category domain="http://www.blogger.com/atom/ns#">Agents</category><category domain="http://www.blogger.com/atom/ns#">Prompt Engineering</category><title>Yapay Zeka Ajanınız Neden Başarısız? Seviye 0'dan Seviye 7'ye Ajan Mimarisi</title><description>&lt;!DOCTYPE html&gt;
&lt;html lang="tr"&gt;
&lt;head&gt;
&lt;meta charset="UTF-8"&gt;
&lt;meta name="viewport" content="width=device-width, initial-scale=1.0"&gt;
&lt;title&gt;Gerçek Yapay Zeka Ajanları Nasıl Tasarlanır? — Seviye 0'dan Seviye 7'ye Adım Adım Kılavuz&lt;/title&gt;
&lt;style&gt;
/* Blogger uyumlu stil — sayfa seviyesinde, post içerik sınıfları */
body {
  font-family: 'Segoe UI', 'Helvetica Neue', Arial, sans-serif;
  line-height: 1.75;
  color: #1e293b;
  max-width: 900px;
  margin: 0 auto;
  padding: 24px;
  background: #f8fafc;
}

.post-header {
  background: linear-gradient(135deg, #1e3a5f 0%, #0f766e 100%);
  color: #ffffff;
  padding: 40px 32px;
  border-radius: 16px;
  margin-bottom: 36px;
  box-shadow: 0 4px 24px rgba(15,118,110,0.18);
}
.post-header h1 {
  font-size: 1.8rem;
  margin: 0 0 12px 0;
  line-height: 1.3;
}
.post-header .subtitle {
  font-size: 1.05rem;
  opacity: 0.92;
  margin: 0;
}
.post-header .meta {
  font-size: 0.85rem;
  opacity: 0.75;
  margin-top: 16px;
}

h2.section-title {
  font-size: 1.45rem;
  color: #1e3a5f;
  border-left: 5px solid #0f766e;
  padding-left: 16px;
  margin-top: 48px;
  margin-bottom: 20px;
}

h3.level-title {
  font-size: 1.2rem;
  color: #0f766e;
  margin-top: 32px;
  margin-bottom: 12px;
}

.level-card {
  background: #ffffff;
  border: 1px solid #e2e8f0;
  border-radius: 12px;
  padding: 24px 28px;
  margin-bottom: 28px;
  box-shadow: 0 1px 6px rgba(0,0,0,0.05);
}
.level-card .level-badge {
  display: inline-block;
  background: #0f766e;
  color: #fff;
  font-weight: 700;
  font-size: 0.8rem;
  padding: 4px 14px;
  border-radius: 999px;
  margin-bottom: 12px;
  letter-spacing: 0.03em;
  text-transform: uppercase;
}
.level-card .level-badge.green { background: #059669; }
.level-card .level-badge.blue  { background: #2563eb; }
.level-card .level-badge.purple{ background: #7c3aed; }
.level-card .level-badge.amber { background: #d97706; }
.level-card .level-badge.red   { background: #dc2626; }
.level-card .level-badge.slate { background: #475569; }

.level-card .scenario {
  background: #f0fdf4;
  border-left: 4px solid #22c55e;
  padding: 12px 16px;
  border-radius: 0 8px 8px 0;
  margin: 16px 0;
  font-size: 0.95rem;
}
.level-card .prompt-box {
  background: #0f172a;
  color: #e2e8f0;
  padding: 16px 20px;
  border-radius: 10px;
  font-family: 'Fira Code', 'Consolas', monospace;
  font-size: 0.85rem;
  overflow-x: auto;
  margin: 16px 0;
  white-space: pre-wrap;
  line-height: 1.6;
}
.level-card .why-better {
  background: #eff6ff;
  border-left: 4px solid #3b82f6;
  padding: 12px 16px;
  border-radius: 0 8px 8px 0;
  margin: 16px 0;
  font-size: 0.95rem;
}
.level-card .why-better ul,
.level-card .why-better ol {
  margin: 8px 0 0 20px;
  padding: 0;
}
.level-card .why-better li {
  margin-bottom: 6px;
}
.level-card .missing {
  background: #fffbeb;
  border-left: 4px solid #f59e0b;
  padding: 12px 16px;
  border-radius: 0 8px 8px 0;
  margin: 16px 0;
  font-size: 0.95rem;
}

table.comparison {
  width: 100%;
  border-collapse: collapse;
  margin: 28px 0;
  font-size: 0.88rem;
}
table.comparison th {
  background: #1e3a5f;
  color: #fff;
  padding: 12px 14px;
  text-align: left;
}
table.comparison td {
  border: 1px solid #e2e8f0;
  padding: 10px 14px;
  vertical-align: top;
}
table.comparison tr:nth-child(even) td {
  background: #f8fafc;
}

.text-center { text-align: center; }
.mt-2 { margin-top: 16px; }
.mb-1 { margin-bottom: 8px; }

.code-block {
  background: #0f172a;
  color: #e2e8f0;
  padding: 16px 20px;
  border-radius: 10px;
  font-family: 'Fira Code', 'Consolas', monospace;
  font-size: 0.83rem;
  overflow-x: auto;
  white-space: pre;
  line-height: 1.55;
  margin: 16px 0;
}

.callout {
  background: linear-gradient(135deg, #eff6ff 0%, #f0fdf4 100%);
  border: 1px solid #bfdbfe;
  border-radius: 12px;
  padding: 24px;
  margin: 28px 0;
}
.callout h4 {
  margin: 0 0 10px 0;
  color: #1e3a5f;
}
.callout p {
  margin: 0;
  font-size: 0.95rem;
}

.final-cta {
  background: linear-gradient(135deg, #1e3a5f 0%, #0f766e 100%);
  color: #fff;
  border-radius: 14px;
  padding: 32px;
  margin: 40px 0 24px 0;
  text-align: center;
}
.final-cta h3 {
  color: #fff;
  margin: 0 0 12px 0;
}
.final-cta p {
  opacity: 0.92;
  font-size: 1rem;
  margin: 0;
}

.rubric-list {
  list-style: none;
  padding: 0;
  margin: 12px 0;
}
.rubric-list li {
  background: #f8fafc;
  border: 1px solid #e2e8f0;
  border-radius: 8px;
  padding: 10px 16px;
  margin-bottom: 8px;
  font-size: 0.95rem;
}
.rubric-list li strong {
  color: #0f766e;
}

.success-message {
  background: #ecfdf5;
  border: 1px solid #a7f3d0;
  border-radius: 10px;
  padding: 18px 22px;
  margin: 20px 0;
  text-align: center;
  font-weight: 600;
  color: #065f46;
}

@media (max-width: 600px) {
  body { padding: 12px; }
  .post-header { padding: 24px 16px; }
  .post-header h1 { font-size: 1.35rem; }
  .level-card { padding: 16px 18px; }
  table.comparison { font-size: 0.78rem; }
}
&lt;/style&gt;
&lt;/head&gt;
&lt;body&gt;

&lt;!-- ======================== HEADER ======================== --&gt;
&lt;div class="post-header"&gt;
  &lt;h1&gt;Gerçek Yapay Zeka Ajanları (Agents) Nasıl Tasarlanır?&lt;/h1&gt;
  &lt;p class="subtitle"&gt;&lt;strong&gt;"Persona"lardan Araç Kullanan (Tool-Using) Karar Mekanizmalarına Geçiş&lt;/strong&gt;&lt;br&gt;
  Seviye 0'dan Seviye 7'ye adım adım senior düzeyde AI Engineer rehberi.&lt;/p&gt;
  &lt;p class="meta"&gt;Yayınlanma: Haziran 2026 &amp;nbsp;|&amp;nbsp; Okuma süresi: ~25 dk &amp;nbsp;|&amp;nbsp; Kategori: Yapay Zeka Mühendisliği&lt;/p&gt;
&lt;/div&gt;

&lt;!-- ======================== GİRİŞ ======================== --&gt;
&lt;h2 class="section-title"&gt;&#127919; Giriş: Neden "Sen Dünyanın En İyi YZ Mühendisisin" Yetersizdir?&lt;/h2&gt;

&lt;div class="level-card"&gt;
  &lt;p&gt;İnternetteki eğitimlerin &lt;strong&gt;%90'ı&lt;/strong&gt; size bir ajan yaratmak için şunu yazmanızı söyler:&lt;/p&gt;
  &lt;div class="prompt-box"&gt;"Sen dünyanın en iyi yazılım mühendisisin..."&lt;/div&gt;
  &lt;p&gt;Eğer production'a, yani canlıya bir uygulama çıkacaksanız, sadece persona vererek yazdığınız promptlar sisteminizi bir &lt;strong&gt;halüsinasyon bombasına&lt;/strong&gt; çevirir.&lt;/p&gt;
  &lt;p&gt;Agent'ların asıl gücü onlara verdiğiniz rollerdan değil; &lt;strong&gt;Tool Kullanımı (Araç Kullanımı)&lt;/strong&gt;, &lt;strong&gt;Grounding (Temellendirme)&lt;/strong&gt; ve &lt;strong&gt;Açıklanabilirlik (Explainability)&lt;/strong&gt;'dir.&lt;/p&gt;

  &lt;div class="callout"&gt;
    &lt;h4&gt;&#128161; Bu Yazıda Ne Öğreneceksiniz?&lt;/h4&gt;
    &lt;p&gt;Bu yazıda basit bir problem olan &lt;strong&gt;"iki metnin benzerliğini ölçme"&lt;/strong&gt; görevini alacağız ve acemi bir yaklaşımdan, senior bir YZ mühendisinin tasarlayacağı otonom seviyeye kadar adım adım nasıl geliştireceğimizi göreceğiz. Her seviyede prompt, mimari ve maliyet karşılaştırması yapacağız.&lt;/p&gt;
  &lt;/div&gt;

  &lt;div class="callout" style="background: linear-gradient(135deg, #fef2f2 0%, #fff7ed 100%); border-color: #fca5a5;"&gt;
    &lt;h4&gt;&#127916; Video ile Öğrenin: Murat Karakaya Akademi&lt;/h4&gt;
    &lt;p&gt;Bu eğitimi &lt;strong&gt;Murat Karakaya Akademi&lt;/strong&gt; YouTube kanalında da izleyebilirsiniz. Adım adım canlı demo, kod açıklaması ve mimari analiz ile Seviye 0'dan Seviye 7'ye aynı yolculuğu video olarak da takip edebilirsiniz.&lt;/p&gt;
    &lt;p class="mt-2"&gt;&lt;a href="https://youtube.com/live/cM4Bkon5q4U" target="_blank" rel="noopener" style="display: inline-block; background: #dc2626; color: #fff; padding: 10px 24px; border-radius: 8px; text-decoration: none; font-weight: 700; font-size: 0.95rem;"&gt;▶ YouTube'da İzle — Murat Karakaya Akademi&lt;/a&gt;&lt;/p&gt;
  &lt;/div&gt;
&lt;/div&gt;

&lt;!-- ======================== SEVİYE 0 ======================== --&gt;
&lt;h2 class="section-title"&gt;&#128204; Seviye 0 — Acemi Yaklaşım: Kara Kutu &amp;amp; Halüsinasyon&lt;/h2&gt;

&lt;div class="level-card"&gt;
  &lt;span class="level-badge red"&gt;Başlangıç&lt;/span&gt;

  &lt;p&gt;&lt;strong&gt;Senaryo:&lt;/strong&gt; Uygulamamızda üretilen Metin 1 ile Metin 2'nin ne kadar benzer olduğunu &lt;strong&gt;0–100&lt;/strong&gt; arasında bulmamız gerekiyor. (Örneğin bir RAG sisteminde, üretilen cevabın referans metne ne kadar sadık olduğunu ölçüyoruz).&lt;/p&gt;

  &lt;div class="scenario"&gt;
    &lt;strong&gt;Acemi Prompt:&lt;/strong&gt;&lt;br&gt;
    "Sen bir metin analistisin. Sana verilen Metin 1 ve Metin 2'yi karşılaştırıp ne kadar benzer olduklarına karar ver. Benzerliği 0–100 arasında bir sayısal değer ile döndür. 0: Hiç benzemiyorlar, 100: Tamamen kelime kelime aynılar."
  &lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Neden Kötü?&lt;/strong&gt;&lt;/p&gt;
  &lt;div class="why-better"&gt;
    &lt;ol&gt;
      &lt;li&gt;&lt;strong&gt;Kara Kutu (Black Box):&lt;/strong&gt; Model sadece "75" döndürdü. Neden 75? 74 veya 80 değil? Bilmiyoruz.&lt;/li&gt;
      &lt;li&gt;&lt;strong&gt;Subjektiflik:&lt;/strong&gt; Aynı metinlere bir gün 60, ertesi gün 85 verebilir.&lt;/li&gt;
      &lt;li&gt;&lt;strong&gt;Halüsinasyon Riski:&lt;/strong&gt; LLM'ler matematiksel ölçüm yapamaz; kelime tahmin eder. "75" uydurulmuş bir değerdir.&lt;/li&gt;
    &lt;/ol&gt;
  &lt;/div&gt;
&lt;/div&gt;

&lt;!-- ======================== SEVİYE 1 ======================== --&gt;
&lt;h2 class="section-title"&gt;&#128204; Seviye 1 — Yapılandırılmış Çıktı &amp;amp; Düşünce Zinciri&lt;/h2&gt;

&lt;div class="level-card"&gt;
  &lt;span class="level-badge blue"&gt;İyileştirme&lt;/span&gt;

  &lt;p&gt;İstediğimiz ilk şey &lt;strong&gt;Açıklanabilirlik (Explainability)&lt;/strong&gt;. Modele "bana sadece sonuç verme, düşünme sürecini de ver" diyoruz.&lt;/p&gt;

  &lt;div class="prompt-box"&gt;"Sen bir metin analistisin. Amacın iki metnin anlamsal benzerliğini ölçmektir.

Çıktını KESİNLİKLE geçerli bir JSON object olarak vermelisin. Markdown veya ek metin kullanma.

Zorunlu JSON alanları:
- reasoning: Bu iki metne özel benzerlik ve farklılık gerekçesi.
- similarity_score: 0-100 arası bir tam sayı.

Beklenen JSON şeması:
{
  &amp;quot;type&amp;quot;: &amp;quot;object&amp;quot;,
  &amp;quot;required&amp;quot;: [&amp;quot;reasoning&amp;quot;, &amp;quot;similarity_score&amp;quot;],
  &amp;quot;properties&amp;quot;: {
    &amp;quot;reasoning&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
    &amp;quot;similarity_score&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;integer&amp;quot;, &amp;quot;minimum&amp;quot;: 0, &amp;quot;maximum&amp;quot;: 100}
  }
}"&lt;/div&gt;

  &lt;div class="why-better"&gt;
    &lt;strong&gt;Neden Daha İyi?&lt;/strong&gt;
    &lt;ul&gt;
      &lt;li&gt;&lt;strong&gt;Chain of Thought (Düşünce Zinciri):&lt;/strong&gt; Modelden önce skoru istemedik. Önce reasoning alanını doldurmasını istedik. Model açıklamayı yazarken kendi çıkarımını temellendirir.&lt;/li&gt;
      &lt;li&gt;&lt;strong&gt;Hata Ayıklama (Debugging):&lt;/strong&gt; Skor 40 geldiğinde, "Model şu detayı kaçırmış, o yüzden düşük vermiş" diyebiliriz.&lt;/li&gt;
      &lt;li&gt;&lt;strong&gt;Entegrasyon:&lt;/strong&gt; JSON çıktısı uygulama kodu tarafından parse edilebilir.&lt;/li&gt;
    &lt;/ul&gt;
  &lt;/div&gt;

  &lt;div class="missing"&gt;
    &lt;strong&gt;⚠️ Eksik Kalan:&lt;/strong&gt; Skor hâlâ subjektif. 0–100 arası skala bir LLM için çok geniştir ve kuralsızdır.
  &lt;/div&gt;
&lt;/div&gt;

&lt;!-- ======================== SEVİYE 2 ======================== --&gt;
&lt;h2 class="section-title"&gt;&#128204; Seviye 2 — Rubrik ve Objektivite (Divide &amp;amp; Conquer)&lt;/h2&gt;

&lt;div class="level-card"&gt;
  &lt;span class="level-badge blue"&gt;Özerklik Artışı&lt;/span&gt;

  &lt;p&gt;Agent'ın "benzerlik" gibi soyut bir kavramı kendi başına yorumlamasına izin vermiyoruz. Ona &lt;strong&gt;kurallar (rubrik)&lt;/strong&gt; veriyoruz. Problemi parçalara bölüyoruz.&lt;/p&gt;

  &lt;p&gt;&lt;strong&gt;Rubrik:&lt;/strong&gt;&lt;/p&gt;
  &lt;ul class="rubric-list"&gt;
    &lt;li&gt;&lt;strong&gt;1. Ana Fikir (0-20):&lt;/strong&gt; İki metin aynı ana mesajı mı veriyor?&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;2. Ton ve Üslup (0-20):&lt;/strong&gt; İki metnin resmiyeti ve duygusu aynı mı?&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;3. Varlıklar / Entities (0-20):&lt;/strong&gt; Metinlerde geçen isimler, tarihler, sayılar eşleşiyor mu?&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;4. Eksik Bilgi (0-20):&lt;/strong&gt; Metin 2, Metin 1'deki önemli bir detayı atlamış mı?&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;5. Akıcılık (0-20):&lt;/strong&gt; Metin 2 yapısal olarak ne kadar tutarlı?&lt;/li&gt;
  &lt;/ul&gt;

  &lt;div class="prompt-box"&gt;"Sen uzman bir değerlendiricisin. İki metni yukarıdaki 5 kritere göre değerlendir. Her kriter 0-20 puan arasındadır.

Çıktını KESİNLİKLE geçerli bir JSON object olarak üret.

Zorunlu JSON alanları:
- evaluations: 5 öğeli liste [criterion, score, explanation]
- total_score: Beş kriter skorunun toplamı.

Beklenen JSON şeması:
{
  &amp;quot;type&amp;quot;: &amp;quot;object&amp;quot;,
  &amp;quot;required&amp;quot;: [&amp;quot;evaluations&amp;quot;, &amp;quot;total_score&amp;quot;],
  &amp;quot;properties&amp;quot;: {
    &amp;quot;evaluations&amp;quot;: {
      &amp;quot;type&amp;quot;: &amp;quot;array&amp;quot;,
      &amp;quot;minItems&amp;quot;: 5,
      &amp;quot;maxItems&amp;quot;: 5,
      &amp;quot;items&amp;quot;: {
        &amp;quot;type&amp;quot;: &amp;quot;object&amp;quot;,
        &amp;quot;required&amp;quot;: [&amp;quot;criterion&amp;quot;, &amp;quot;score&amp;quot;, &amp;quot;explanation&amp;quot;],
        &amp;quot;properties&amp;quot;: {
          &amp;quot;criterion&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
          &amp;quot;score&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;integer&amp;quot;, &amp;quot;minimum&amp;quot;: 0, &amp;quot;maximum&amp;quot;: 20},
          &amp;quot;explanation&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;}
        }
      }
    },
    &amp;quot;total_score&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;integer&amp;quot;, &amp;quot;minimum&amp;quot;: 0, &amp;quot;maximum&amp;quot;: 100}
  }
}"&lt;/div&gt;

  &lt;div class="why-better"&gt;
    &lt;strong&gt;Neden Daha İyi?&lt;/strong&gt;
    &lt;ul&gt;
      &lt;li&gt;&lt;strong&gt;Grounding (Temellendirme):&lt;/strong&gt; Modeli soyut 100 üzerinden değerlendirmeden kurtardık.&lt;/li&gt;
      &lt;li&gt;&lt;strong&gt;Objektiflik:&lt;/strong&gt; Farklı zamanlarda çalıştırsanız bile skorlar çok daha tutarlı olur (varyans azalır).&lt;/li&gt;
      &lt;li&gt;&lt;strong&gt;Karşılaştırılabilirlik:&lt;/strong&gt; Farklı promptları aynı rubrik üzerinden değerlendirebiliriz.&lt;/li&gt;
    &lt;/ul&gt;
  &lt;/div&gt;
&lt;/div&gt;

&lt;!-- ======================== SEVİYE 3 ======================== --&gt;
&lt;h2 class="section-title"&gt;&#128204; Seviye 3 — Örnekli Rubrik Kalibrasyonu (One-Shot / Few-Shot)&lt;/h2&gt;

&lt;div class="level-card"&gt;
  &lt;span class="level-badge amber"&gt;Kalibrasyon&lt;/span&gt;

  &lt;p&gt;Seviye 2'de rubrik verdik; fakat bu hâlâ &lt;strong&gt;sıfır-örnekli prompting (zero-shot prompting)&lt;/strong&gt; idi. Model kriterleri okudu ama "20 puan ne zaman verilir?", "10 puan hangi sınır durumu?" gibi karar sınırlarını örneklerden görmedi.&lt;/p&gt;

  &lt;p&gt;Bu seviyede her kriter için küçük ve temsil gücü yüksek örnekler veriyoruz.&lt;/p&gt;

  &lt;p&gt;&lt;strong&gt;Puanlama Kalibrasyonu Örnekleri:&lt;/strong&gt;&lt;/p&gt;

  &lt;ul class="rubric-list"&gt;
    &lt;li&gt;&lt;strong&gt;Ana Fikir — 20 puan:&lt;/strong&gt; "Veri temizliği model başarısı için kritiktir" ve "Model kalitesi büyük ölçüde temiz veriye bağlıdır" aynı ana fikri verir.&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;Ana Fikir — 10 puan:&lt;/strong&gt; İki metin de veri kalitesinden söz eder ama biri güvenlik risklerine, diğeri model performansına odaklanır.&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;Ana Fikir — 0 puan:&lt;/strong&gt; Biri veri temizliğini, diğeri spor müsabakası sonucunu anlatır.&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;Ton ve Üslup — 20 puan:&lt;/strong&gt; İki metin de akademik ve resmi bir dille yazılmıştır.&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;Ton ve Üslup — 10 puan:&lt;/strong&gt; Biri resmi, diğeri daha konuşma dilindedir ama anlam korunmuştur.&lt;/li&gt;
  &lt;/ul&gt;

  &lt;div class="why-better"&gt;
    &lt;strong&gt;Neden Daha İyi?&lt;/strong&gt;
    &lt;ul&gt;
      &lt;li&gt;&lt;strong&gt;Tutarlılık:&lt;/strong&gt; Model aynı puan aralıklarını daha kararlı kullanır.&lt;/li&gt;
      &lt;li&gt;&lt;strong&gt;Öğretilebilirlik:&lt;/strong&gt; Rubrik artık yalnızca soyut bir liste değil, davranış örnekleriyle desteklenmiştir.&lt;/li&gt;
      &lt;li&gt;&lt;strong&gt;Maliyet:&lt;/strong&gt; Few-shot örnekler giriş tokenını artırır — bu yüzden örnekler kısa ve net seçilir.&lt;/li&gt;
    &lt;/ul&gt;
  &lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Beklenen JSON şeması:&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="code-block"&gt;{
  &amp;quot;type&amp;quot;: &amp;quot;object&amp;quot;,
  &amp;quot;required&amp;quot;: [&amp;quot;evaluations&amp;quot;, &amp;quot;calibration_note&amp;quot;, &amp;quot;total_score&amp;quot;],
  &amp;quot;properties&amp;quot;: {
    &amp;quot;evaluations&amp;quot;: {
      &amp;quot;type&amp;quot;: &amp;quot;array&amp;quot;,
      &amp;quot;minItems&amp;quot;: 5,
      &amp;quot;maxItems&amp;quot;: 5,
      &amp;quot;items&amp;quot;: {
        &amp;quot;type&amp;quot;: &amp;quot;object&amp;quot;,
        &amp;quot;required&amp;quot;: [&amp;quot;criterion&amp;quot;, &amp;quot;score&amp;quot;, &amp;quot;explanation&amp;quot;],
        &amp;quot;properties&amp;quot;: {
          &amp;quot;criterion&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
          &amp;quot;score&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;integer&amp;quot;, &amp;quot;minimum&amp;quot;: 0, &amp;quot;maximum&amp;quot;: 20},
          &amp;quot;explanation&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;}
        }
      }
    },
    &amp;quot;calibration_note&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
    &amp;quot;total_score&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;integer&amp;quot;, &amp;quot;minimum&amp;quot;: 0, &amp;quot;maximum&amp;quot;: 100}
  }
}&lt;/div&gt;
&lt;/div&gt;

&lt;!-- ======================== SEVİYE 4 ======================== --&gt;
&lt;h2 class="section-title"&gt;&#128204; Seviye 4 — Araç Kullanımı (Tool Calling) ve İş Akışı (Workflow)&lt;/h2&gt;

&lt;div class="level-card"&gt;
  &lt;span class="level-badge purple"&gt;Temellendirme Başlıyor&lt;/span&gt;

  &lt;p&gt;Seviye 3'te rubriği örneklerle kalibre ettik; fakat değerlendirme hâlâ yalnızca LLM yorumuna dayanıyor. Şimdi &lt;strong&gt;dış sistemlerden gelen deterministik metrikleri&lt;/strong&gt; kanıt olarak ekliyoruz.&lt;/p&gt;

  &lt;p&gt;&lt;strong&gt;Kullanılan Metrikler:&lt;/strong&gt;&lt;/p&gt;
  &lt;ul class="rubric-list"&gt;
    &lt;li&gt;&lt;strong&gt;ROUGE-L F1:&lt;/strong&gt; Kelime dizisi örtüşmesini ölçer.&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;Hafif Benzerlik Skoru:&lt;/strong&gt; Token cosine + Token Jaccard + Karakter 3-gram cosine + Sequence ratio birleşimi.&lt;/li&gt;
  &lt;/ul&gt;

  &lt;div class="scenario"&gt;
    &lt;strong&gt;Metrik Yorumlama Kuralları:&lt;/strong&gt;
    &lt;ul&gt;
      &lt;li&gt;ROUGE-L F1 düşükse kelime örtüşmesinin düşük olduğunu gösterir; tek başına düşük anlamsal benzerlik kanıtı değildir.&lt;/li&gt;
      &lt;li&gt;Hafif Benzerlik Skoru ROUGE-L'ye göre daha yüksekse, metinler farklı kelimelerle benzer mesaj veriyor olabilir.&lt;/li&gt;
      &lt;li&gt;Hafif Benzerlik Skoru gerçek semantik embedding değildir; karar destek sinyali olarak kullanılmalı, tek karar verici yapılmamalıdır.&lt;/li&gt;
    &lt;/ul&gt;
  &lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;&#127757; Gerçek Dünyada Neden Bu Kullanılır?&lt;/strong&gt;&lt;/p&gt;
  &lt;div class="why-better"&gt;
    &lt;ul&gt;
      &lt;li&gt;&lt;strong&gt;Güvenilirlik:&lt;/strong&gt; ROUGE ve hafif benzerlik skorları deterministiktir — aynı metin çiftine her zaman aynı skorları verir.&lt;/li&gt;
      &lt;li&gt;&lt;strong&gt;İzlenebilirlik:&lt;/strong&gt; LLM yorumu dış kanıta dayandığı için değerlendirme daha denetlenebilir olur.&lt;/li&gt;
      &lt;li&gt;&lt;strong&gt;Maliyet Kontrolü:&lt;/strong&gt; Hafif metrikler hızlıdır ve ağır model bağımlılığı gerektirmez.&lt;/li&gt;
    &lt;/ul&gt;
  &lt;/div&gt;

  &lt;div class="callout"&gt;
    &lt;h4&gt;&#128300; Deney Hijyeni Notu&lt;/h4&gt;
    &lt;p&gt;Seviye 4, Seviye 3 ile aynı rubrik metnini kullanır. Bu bilinçli bir karar: iki seviye arasındaki fark rubrik değişikliği değil, yalnızca &lt;strong&gt;dış metrik bağlamıdır&lt;/strong&gt;.&lt;/p&gt;
  &lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Seviye 4'te eklenen Zorunlu JSON alanları:&lt;/strong&gt;&lt;/p&gt;
  &lt;ul&gt;
    &lt;li&gt;&lt;code&gt;metric_interpretation&lt;/code&gt;: ROUGE-L F1 ve Hafif Benzerlik Skoru değerlerini nasıl yorumladığını açıkla.&lt;/li&gt;
    &lt;li&gt;&lt;code&gt;calibration_note&lt;/code&gt;: Rubrik kalibrasyon örneklerinin puanlamanı nasıl etkilediğini açıkla.&lt;/li&gt;
  &lt;/ul&gt;

  &lt;p&gt;&lt;strong&gt;Beklenen JSON şeması (Seviye 4):&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="code-block"&gt;{
  &amp;quot;type&amp;quot;: &amp;quot;object&amp;quot;,
  &amp;quot;required&amp;quot;: [&amp;quot;evaluations&amp;quot;, &amp;quot;metric_interpretation&amp;quot;, &amp;quot;calibration_note&amp;quot;, &amp;quot;total_score&amp;quot;],
  &amp;quot;properties&amp;quot;: {
    &amp;quot;evaluations&amp;quot;: { /* Seviye 3 ile aynı 5 öğeli liste */ },
    &amp;quot;metric_interpretation&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
    &amp;quot;calibration_note&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
    &amp;quot;total_score&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;integer&amp;quot;, &amp;quot;minimum&amp;quot;: 0, &amp;quot;maximum&amp;quot;: 100}
  }
}&lt;/div&gt;
&lt;/div&gt;

&lt;!-- ======================== SEVİYE 5 ======================== --&gt;
&lt;h2 class="section-title"&gt;&#128204; Seviye 5 — ReAct ve Ollama Tool Calling ile Gerçek Ajan Döngüsü&lt;/h2&gt;

&lt;div class="level-card"&gt;
  &lt;span class="level-badge purple"&gt;Gerçek Agentic Loop&lt;/span&gt;

  &lt;p&gt;Seviye 4'te güçlü bir iş akışı kurduk ama kurduğumuz sistem gerçek bir ajan mıydı? Tam olarak &lt;strong&gt;hayır&lt;/strong&gt;. Çünkü metrikleri Python ile biz hesapladık. Gerçek ajan davranışında &lt;strong&gt;model hangi kanıta ihtiyaç duyduğunu belirler&lt;/strong&gt;, yazılım katmanı aracı çalıştırır ve sonuç tekrar modele döner.&lt;/p&gt;

  &lt;h3 class="mb-1"&gt;ReAct Döngüsü:&lt;/h3&gt;
  &lt;ul&gt;
    &lt;li&gt;&lt;strong&gt;Reasoning (Muhakeme):&lt;/strong&gt; Model hangi dış kanıta ihtiyaç duyduğunu belirler.&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;Action (Eylem):&lt;/strong&gt; Model düz metin yazmak yerine Ollama'nın native &lt;code&gt;tool_calls&lt;/code&gt; alanını üretir.&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;Observation (Gözlem):&lt;/strong&gt; Python ilgili fonksiyonu çalıştırır, sonucu &lt;code&gt;role="tool"&lt;/code&gt; olarak modele geri verir.&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;Final Answer (Nihai Cevap):&lt;/strong&gt; Model tool sonuçlarını kullanarak rubrik temelli JSON değerlendirmesini üretir.&lt;/li&gt;
  &lt;/ul&gt;

  &lt;p&gt;&lt;strong&gt;Ollama Native Tool Calling Akışı:&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="code-block"&gt;1. Fonksiyonlar `tools=[...]` listesiyle modele tanıtılır.
2. Model gerekiyorsa `response.message.tool_calls` üretir.
3. Python bu tool call'ları çalıştırır.
4. Sonuçlar `role="tool"` mesajları olarak konuşmaya eklenir.
5. Model tool sonuçlarına dayanarak nihai cevabı üretir.&lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Tool Tanımları (Python):&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="code-block"&gt;def calculate_rouge_tool(reference_text: str = "", candidate_text: str = "") -&amp;gt; str:
    """ROUGE skorlarını hesapla."""
    return json.dumps(calculate_rouge(METIN_1, METIN_2), ensure_ascii=False)

def calculate_lightweight_similarity_tool(reference_text: str = "", candidate_text: str = "") -&amp;gt; str:
    """Notebook'taki sabit Metin 1 ve Metin 2 için hafif benzerlik skorunu hesapla.
    
    Args:
        reference_text: Eğitim demosunda dikkate alınmaz; tool sabit METIN_1 kullanır.
        candidate_text: Eğitim demosunda dikkate alınmaz; tool sabit METIN_2 kullanır.
    
    Returns:
        Token cosine, token Jaccard, karakter 3-gram cosine, sequence ratio 
        ve birleşik skoru içeren JSON string.
    """
    return json.dumps(calculate_lightweight_similarity(METIN_1, METIN_2), ensure_ascii=False)&lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Tool Calling Loop (Pseudo-Python):&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="code-block"&gt;messages = [
    {"role": "system", "content": TOOL_CALLING_SYSTEM_PROMPT},
    {"role": "user", "content": "Metin 1: ..., Metin 2: ..."},
]

while True:
    response = client.chat(
        model=MODEL,
        messages=messages,
        tools=[calculate_rouge_tool, calculate_lightweight_similarity_tool],
    )
    messages.append(response.message)

    if not response.message.tool_calls:
        print(response.message.content)  # Final cevap
        break

    for tool_call in response.message.tool_calls:
        tool_name = tool_call.function.name
        result = available_tools[tool_name](**tool_call.function.arguments)
        messages.append({
            "role": "tool",
            "tool_name": tool_name,
            "content": str(result),
        })&lt;/div&gt;

  &lt;div class="why-better"&gt;
    &lt;strong&gt;Neden Bu Ustalık Sınıfıdır?&lt;/strong&gt;
    &lt;ul&gt;
      &lt;li&gt;ReAct yalnızca &lt;code&gt;Thought / Action / Observation&lt;/code&gt; yazdırmak değildir; düşünceyi gerçek araç yürütmeye bağlamaktır.&lt;/li&gt;
      &lt;li&gt;Ajan; &lt;strong&gt;prompt&lt;/strong&gt;, &lt;strong&gt;tool calling&lt;/strong&gt;, &lt;strong&gt;execution loop&lt;/strong&gt;, &lt;strong&gt;grounding&lt;/strong&gt; ve &lt;strong&gt;hata kontrolü&lt;/strong&gt; katmanlarının birlikte tasarlanmasıdır.&lt;/li&gt;
      &lt;li&gt;Model araç ihtiyacını belirler, Python aracı gerçekten çalıştırır, nihai değerlendirme dış kanıtla desteklenir.&lt;/li&gt;
    &lt;/ul&gt;
  &lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;&#128202; Seviye 5 Token Maliyeti:&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="scenario"&gt;
    &lt;p&gt;&lt;em&gt;"Seviye 5 çok turlu (multi-turn) bir ajan loop'u olduğu için giriş tokenı (input token), yalnızca ilk kullanıcı promptunun uzunluğu değildir. Her &lt;code&gt;client.chat&lt;/code&gt; çağrısında sistem mesajı, kullanıcı mesajı, önceki asistan mesajları ve tool sonuçları yeniden bağlama girer. Bu nedenle Seviye 5'teki input token toplamı, tekil token sayısı değil, &lt;strong&gt;kümülatif işlenen token / maliyet göstergesidir&lt;/strong&gt;."&lt;/em&gt;&lt;/p&gt;
  &lt;/div&gt;

  &lt;div class="callout"&gt;
    &lt;h4&gt;&#128737;️ Gerçek Dünyada: Üretim Ortamı Koruma Katmanları&lt;/h4&gt;
    &lt;p&gt;Gerçek üretimde şu korumalar eklenir: şema doğrulama (schema validation), maksimum adım sayısı (max step limit), tool izin listesi (tool allowlist), yeniden deneme (retry) ve izleme/günlükleme (tracing/logging).&lt;/p&gt;
  &lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Seviye 5 Tool Calling JSON şeması:&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="code-block"&gt;{
  &amp;quot;type&amp;quot;: &amp;quot;object&amp;quot;,
  &amp;quot;required&amp;quot;: [&amp;quot;evaluations&amp;quot;, &amp;quot;metric_interpretation&amp;quot;, &amp;quot;calibration_note&amp;quot;, &amp;quot;total_score&amp;quot;],
  &amp;quot;properties&amp;quot;: {
    &amp;quot;evaluations&amp;quot;: {
      &amp;quot;type&amp;quot;: &amp;quot;array&amp;quot;,
      &amp;quot;minItems&amp;quot;: 5,
      &amp;quot;maxItems&amp;quot;: 5,
      &amp;quot;items&amp;quot;: {
        &amp;quot;type&amp;quot;: &amp;quot;object&amp;quot;,
        &amp;quot;required&amp;quot;: [&amp;quot;criterion&amp;quot;, &amp;quot;score&amp;quot;, &amp;quot;explanation&amp;quot;],
        &amp;quot;properties&amp;quot;: {
          &amp;quot;criterion&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
          &amp;quot;score&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;integer&amp;quot;, &amp;quot;minimum&amp;quot;: 0, &amp;quot;maximum&amp;quot;: 20},
          &amp;quot;explanation&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;}
        }
      }
    },
    &amp;quot;metric_interpretation&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
    &amp;quot;calibration_note&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
    &amp;quot;total_score&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;integer&amp;quot;, &amp;quot;minimum&amp;quot;: 0, &amp;quot;maximum&amp;quot;: 100}
  }
}&lt;/div&gt;
&lt;/div&gt;

&lt;!-- ======================== SEVİYE 6 ======================== --&gt;
&lt;h2 class="section-title"&gt;&#128204; Seviye 6 — Rubrik Bazlı Sub-Agentlar ve Python Aggregator&lt;/h2&gt;

&lt;div class="level-card"&gt;
  &lt;span class="level-badge purple"&gt;Modüler Mimarisi&lt;/span&gt;

  &lt;p&gt;Seviye 5'te tek bir ajan hem araç çağırdı hem de bütün rubriği tek başına yorumladı. Bu seviyede farklı bir mimari deniyoruz: &lt;strong&gt;tek büyük prompt yerine, her rubrik kriterini ayrı bir alt ajana (sub-agent) veriyoruz.&lt;/strong&gt;&lt;/p&gt;

  &lt;h3 class="mb-1"&gt;Mimari:&lt;/h3&gt;
  &lt;ul&gt;
    &lt;li&gt;1 generic sub-agent fonksiyonu yazılır.&lt;/li&gt;
    &lt;li&gt;Bu fonksiyon 5 farklı rubrik konfigürasyonu ile 5 kez çağrılır.&lt;/li&gt;
    &lt;li&gt;Her sub-agent yalnızca kendi kriterini değerlendirir.&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;Python aggregator&lt;/strong&gt; sonuçları doğrular, sıraya koyar ve toplam skoru hesaplar.&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;Aggregator LLM çağrısı yapmaz&lt;/strong&gt; — deterministiktir.&lt;/li&gt;
  &lt;/ul&gt;

  &lt;p&gt;&lt;strong&gt;Bu Seviyenin Pedagojik Mesajı:&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="success-message"&gt;
    ✅ "Her şeyi agent yapmayın!"&lt;br&gt;
    LLM yorumsal değerlendirme içindir. Python doğrulama, toplama, formatlama ve deterministik hesaplama içindir.
  &lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Sınırları:&lt;/strong&gt;&lt;/p&gt;
  &lt;ul&gt;
    &lt;li&gt;5 sub-agent = 5 LLM çağrısı = Seviye 5'ten daha pahalı.&lt;/li&gt;
    &lt;li&gt;Her durumda gerekli değildir; rubrik büyüdüğünde veya denetlenebilirlik kritik olduğunda anlam kazanır.&lt;/li&gt;
  &lt;/ul&gt;

  &lt;p&gt;&lt;strong&gt;Seviye 6 - Sub-Agent JSON Çıktısı (tek kriter için):&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="code-block"&gt;{
  &amp;quot;criterion&amp;quot;: &amp;quot;Ana Fikir&amp;quot;,
  &amp;quot;score&amp;quot;: 18,
  &amp;quot;explanation&amp;quot;: &amp;quot;Bu kritere özel değerlendirme&amp;quot;,
  &amp;quot;evidence&amp;quot;: &amp;quot;Metinden veya metriklerden dayanak&amp;quot;
}&lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Python Aggregator Toplam Çıktısı (tüm kriterler):&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="code-block"&gt;{
  &amp;quot;evaluations&amp;quot;: [
    {&amp;quot;criterion&amp;quot;: &amp;quot;Ana Fikir&amp;quot;, &amp;quot;score&amp;quot;: 18, &amp;quot;explanation&amp;quot;: &amp;quot;...&amp;quot;, &amp;quot;evidence&amp;quot;: &amp;quot;...&amp;quot;},
    {&amp;quot;criterion&amp;quot;: &amp;quot;Ton ve Üslup&amp;quot;, &amp;quot;score&amp;quot;: 16, &amp;quot;explanation&amp;quot;: &amp;quot;...&amp;quot;, &amp;quot;evidence&amp;quot;: &amp;quot;...&amp;quot;},
    {&amp;quot;criterion&amp;quot;: &amp;quot;Varlıklar&amp;quot;, &amp;quot;score&amp;quot;: 17, &amp;quot;explanation&amp;quot;: &amp;quot;...&amp;quot;, &amp;quot;evidence&amp;quot;: &amp;quot;...&amp;quot;},
    {&amp;quot;criterion&amp;quot;: &amp;quot;Eksik Bilgi&amp;quot;, &amp;quot;score&amp;quot;: 15, &amp;quot;explanation&amp;quot;: &amp;quot;...&amp;quot;, &amp;quot;evidence&amp;quot;: &amp;quot;...&amp;quot;},
    {&amp;quot;criterion&amp;quot;: &amp;quot;Akıcılık&amp;quot;, &amp;quot;score&amp;quot;: 14, &amp;quot;explanation&amp;quot;: &amp;quot;...&amp;quot;, &amp;quot;evidence&amp;quot;: &amp;quot;...&amp;quot;}
  ],
  &amp;quot;metric_interpretation&amp;quot;: &amp;quot;ROUGE ve hafif benzerlik metrikleri her sub-agent'a bağlam olarak verildi; her kriter kendi uzman alt ajanı tarafından yorumlandı.&amp;quot;,
  &amp;quot;aggregation_note&amp;quot;: &amp;quot;Toplam skor Python aggregator tarafından deterministik olarak hesaplandı; LLM orchestrator kullanılmadı.&amp;quot;,
  &amp;quot;total_score&amp;quot;: 80
}&lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Seviye 6 Python aggregator fonksiyon örneği:&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="code-block"&gt;def aggregate_sub_agent_results(sub_agent_results, metrics_context):
    expected_criteria = [config['criterion'] for config in RUBRIC_AGENT_CONFIGS]
    actual_criteria = [result['parsed'].get('criterion') for result in sub_agent_results]

    if actual_criteria != expected_criteria:
        raise ValueError(f"Sub-agent kriter sırası beklenenle aynı değil.")

    evaluations = []
    for result in sub_agent_results:
        parsed = result['parsed']
        score = parsed.get('score')
        if not isinstance(score, int) or not 0 &lt;= score &lt;= 20:
            raise ValueError(f"Geçersiz score: {parsed}")
        evaluations.append({
            'criterion': parsed['criterion'],
            'score': score,
            'explanation': parsed['explanation'],
            'evidence': parsed['evidence'],
        })

    total_score = sum(item['score'] for item in evaluations)
    return {
        'evaluations': evaluations,
        'metric_interpretation': '...',
        'aggregation_note': 'Toplam skor Python ile deterministik olarak hesaplandı.',
        'total_score': total_score,
    }&lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Seviye 6 — Token Maliyeti:&lt;/strong&gt;&lt;/p&gt;
  &lt;ul&gt;
    &lt;li&gt;5 sub-agent = 5 bağımsız LLM çağrısı&lt;/li&gt;
    &lt;li&gt;Her çağrı sistem promptu + kullanıcı promptu + metrik bağlamını içerir&lt;/li&gt;
    &lt;li&gt;Toplam input token = 5 × (sistem + kullanıcı promptu uzunluğu)&lt;/li&gt;
    &lt;li&gt;Aggregator token maliyeti = 0 (Python kodu çalışır)&lt;/li&gt;
  &lt;/ul&gt;

  &lt;p&gt;&lt;strong&gt;Seviye 6 — Sub-Agent JSON Şeması (build_sub_agent_system_prompt içinde):&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="code-block"&gt;{
  &amp;quot;type&amp;quot;: &amp;quot;object&amp;quot;,
  &amp;quot;required&amp;quot;: [&amp;quot;criterion&amp;quot;, &amp;quot;score&amp;quot;, &amp;quot;explanation&amp;quot;, &amp;quot;evidence&amp;quot;],
  &amp;quot;properties&amp;quot;: {
    &amp;quot;criterion&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
    &amp;quot;score&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;integer&amp;quot;, &amp;quot;minimum&amp;quot;: 0, &amp;quot;maximum&amp;quot;: 20},
    &amp;quot;explanation&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
    &amp;quot;evidence&amp;quot;: {&amp;quot;type&amp;quot;: [&amp;quot;string&amp;quot;, &amp;quot;array&amp;quot;, &amp;quot;object&amp;quot;]}
  }
}&lt;/div&gt;
&lt;/div&gt;

&lt;!-- ======================== SEVİYE 7 ======================== --&gt;
&lt;h2 class="section-title"&gt;&#128204; Seviye 7 — Orkestratör Ajanlar ve Ne Zaman Gerekli Olmadıkları&lt;/h2&gt;

&lt;div class="level-card"&gt;
  &lt;span class="level-badge slate"&gt;Mimari Karar&lt;/span&gt;

  &lt;p&gt;Seviye 6'dan sonra doğal soru: &lt;em&gt;"Bu alt ajanları bir de orkestratör ajan yönetse daha iyi olmaz mı?"&lt;/em&gt;&lt;/p&gt;

  &lt;p&gt;&lt;strong&gt;Orkestratör Ajan (Orchestrator Agent)&lt;/strong&gt; gerçek hayatta güçlü bir mimaridir. Görevi parçalara ayırabilir, hangi sub-agent'ın çalışacağına karar verebilir, araç seçebilir, eksik veya çelişkili sonuçlarda yeniden deneme başlatabilir ve farklı ajanlardan gelen sonuçları nihai karara dönüştürebilir.&lt;/p&gt;

  &lt;p&gt;Ancak bu örnekte buna &lt;strong&gt;bilinçli olarak gerek duymuyoruz&lt;/strong&gt;, çünkü:&lt;/p&gt;
  &lt;ul&gt;
    &lt;li&gt;5 rubrik kriteri önceden belli.&lt;/li&gt;
    &lt;li&gt;Her kriterin mutlaka çalışması gerekiyor.&lt;/li&gt;
    &lt;li&gt;Her sub-agent yalnızca kendi kriterini değerlendiriyor.&lt;/li&gt;
    &lt;li&gt;Eksik kriter kontrolü, sıralama ve toplam skor Python ile güvenilir biçimde yapılabiliyor.&lt;/li&gt;
  &lt;/ul&gt;

  &lt;div class="callout"&gt;
    &lt;h4&gt;&#127959;️ Orkestratör Ajan Ne Zaman Anlamlıdır?&lt;/h4&gt;
    &lt;ul&gt;
      &lt;li&gt;Hangi sub-agent'ların çalışacağı görevden göreve değişiyorsa.&lt;/li&gt;
      &lt;li&gt;Araç seçimi, veri kaynağı seçimi veya iş akışı dallanması gerekiyorsa.&lt;/li&gt;
      &lt;li&gt;Alt ajan cevapları arasında çelişki varsa ve yorumlayıcı uzlaştırma gerekiyorsa.&lt;/li&gt;
      &lt;li&gt;Kalite kontrol, yeniden deneme, eksik bilgi tamamlama veya insan onayı gibi dinamik adımlar varsa.&lt;/li&gt;
    &lt;/ul&gt;
  &lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Seviye 7'nin Mesajı:&lt;/strong&gt; Orkestratör + sub-agent mimarileri vardır ve önemlidir; fakat her problemde gerekli değildir. Bu örnekte Python aggregator doğru, sade ve öğretici tercihtir.&lt;/p&gt;

  &lt;div class="success-message"&gt;
    &#128161; &lt;strong&gt;Seviye 7 Notu:&lt;/strong&gt; Bu seviye yeni bir LLM çağrısı çalıştırmaz. Mimari karar verme ve sınır çizme bölümüdür. Performans grafikleri Seviye 0-6 deneylerinin maliyetini gösterir.
  &lt;/div&gt;
&lt;/div&gt;

&lt;!-- ======================== KARŞILAŞTIRMA TABLOSU ======================== --&gt;
&lt;h2 class="section-title"&gt;&#128202; Tüm Seviyelerin Karşılaştırması&lt;/h2&gt;

&lt;table class="comparison"&gt;
  &lt;thead&gt;
    &lt;tr&gt;
      &lt;th&gt;Seviye&lt;/th&gt;
      &lt;th&gt;Yaklaşım&lt;/th&gt;
      &lt;th&gt;Eklenen Katman&lt;/th&gt;
      &lt;th&gt;Kazanım&lt;/th&gt;
      &lt;th&gt;Sınır / Ders&lt;/th&gt;
    &lt;/tr&gt;
  &lt;/thead&gt;
  &lt;tbody&gt;
    &lt;tr&gt;
      &lt;td&gt;&lt;strong&gt;Lvl 0&lt;/strong&gt;&lt;/td&gt;
      &lt;td&gt;Persona / Kara Kutu&lt;/td&gt;
      &lt;td&gt;Basit sistem promptu&lt;/td&gt;
      &lt;td&gt;Hızlı başlangıç&lt;/td&gt;
      &lt;td&gt;Tutarsız, açıklanamaz, halüsinasyona açık&lt;/td&gt;
    &lt;/tr&gt;
    &lt;tr&gt;
      &lt;td&gt;&lt;strong&gt;Lvl 1&lt;/strong&gt;&lt;/td&gt;
      &lt;td&gt;JSON + Açıklama&lt;/td&gt;
      &lt;td&gt;Yapılandırılmış çıktı&lt;/td&gt;
      &lt;td&gt;Cevap parse edilebilir&lt;/td&gt;
      &lt;td&gt;Hâlâ subjektif&lt;/td&gt;
    &lt;/tr&gt;
    &lt;tr&gt;
      &lt;td&gt;&lt;strong&gt;Lvl 2&lt;/strong&gt;&lt;/td&gt;
      &lt;td&gt;Rubrik&lt;/td&gt;
      &lt;td&gt;Kriter bazlı değerlendirme&lt;/td&gt;
      &lt;td&gt;Daha objektif skor&lt;/td&gt;
      &lt;td&gt;Dış kanıt yok, hâlâ LLM yorumu&lt;/td&gt;
    &lt;/tr&gt;
    &lt;tr&gt;
      &lt;td&gt;&lt;strong&gt;Lvl 3&lt;/strong&gt;&lt;/td&gt;
      &lt;td&gt;One-Shot / Few-Shot Kalibrasyon&lt;/td&gt;
      &lt;td&gt;Kriter bazlı örnekler&lt;/td&gt;
      &lt;td&gt;Tutarlı skorlar&lt;/td&gt;
      &lt;td&gt;İşlem tokenı artar&lt;/td&gt;
    &lt;/tr&gt;
    &lt;tr&gt;
      &lt;td&gt;&lt;strong&gt;Lvl 4&lt;/strong&gt;&lt;/td&gt;
      &lt;td&gt;Workflow + Tools&lt;/td&gt;
      &lt;td&gt;ROUGE + hafif benzerlik metrikleri&lt;/td&gt;
      &lt;td&gt;Grounded, kanıta dayalı&lt;/td&gt;
      &lt;td&gt;Tool seçimini yazılımcı yapar&lt;/td&gt;
    &lt;/tr&gt;
    &lt;tr&gt;
      &lt;td&gt;&lt;strong&gt;Lvl 5&lt;/strong&gt;&lt;/td&gt;
      &lt;td&gt;ReAct + Tool Calling&lt;/td&gt;
      &lt;td&gt;Otomatik tool çağrısı + execution loop&lt;/td&gt;
      &lt;td&gt;Gerçek agent davranışı&lt;/td&gt;
      &lt;td&gt;Yüksek maliyet, loop yönetimi gerekli&lt;/td&gt;
    &lt;/tr&gt;
    &lt;tr&gt;
      &lt;td&gt;&lt;strong&gt;Lvl 6&lt;/strong&gt;&lt;/td&gt;
      &lt;td&gt;Sub-Agentlar + Python Aggregator&lt;/td&gt;
      &lt;td&gt;Generic sub-agent + deterministic toplama&lt;/td&gt;
      &lt;td&gt;Görev ayrıştırma, sorumluluk ayrımı&lt;/td&gt;
      &lt;td&gt;5× LLM çağrısı, daha pahalı&lt;/td&gt;
    &lt;/tr&gt;
    &lt;tr&gt;
      &lt;td&gt;&lt;strong&gt;Lvl 7&lt;/strong&gt;&lt;/td&gt;
      &lt;td&gt;Orkestratör Ajan Kararı&lt;/td&gt;
      &lt;td&gt;Mimari karar verme&lt;/td&gt;
      &lt;td&gt;İleri mimari anlayışı&lt;/td&gt;
      &lt;td&gt;Her yerde orkestratör gerekmez&lt;/td&gt;
    &lt;/tr&gt;
  &lt;/tbody&gt;
&lt;/table&gt;

&lt;!-- ======================== KAPANIŞ ======================== --&gt;
&lt;div class="final-cta"&gt;
  &lt;h3&gt;&#127891; Son Mesaj: Prompt Mühendisliği, Sistem Mühendisliğine Dönüşür&lt;/h3&gt;
  &lt;p&gt;Bir ajana yalnızca güçlü görünen bir prompt verip doğru çalışmasını beklemek yeterli değildir.&lt;br&gt;
  Gerçek Yapay Zeka Ajanı, yalnızca cevap üreten bir model değil; &lt;strong&gt;düşünme biçimi, araç kullanımı, muhakeme adımları ve veriye dayalı kanıtları&lt;/strong&gt; birlikte yönetilen bir yazılım sistemidir.&lt;/p&gt;
  &lt;p class="mt-2" style="opacity:0.8; font-style:italic;"&gt;Seviye 0'dan Seviye 7'ye ilerlerken aslında aynı fikri katman katman inşa ettik. Önce çıktıyı yapılandırdık, sonra değerlendirmeyi rubrikle parçaladık, ardından örneklerle kalibre ettik. Daha sonra metriklerle temellendirme ekledik ve ReAct fikrini gerçek araç yürütmeye bağladık. Sonunda orkestratör ajan gibi daha ileri mimarilerin var olduğunu, fakat bu örnekte ek LLM orchestrator kullanmamanın daha doğru mühendislik kararı olduğunu gördük.&lt;/p&gt;
  &lt;p class="mt-2" style="opacity:0.85;"&gt;Bir YZ Mühendisi olarak değerimiz burada ortaya çıkar: Modelden mucize beklemek yerine, modelin güçlü ve zayıf yönlerini anlayıp etrafına doğru mimariyi kurmak. &lt;strong&gt;İyi ajan tasarımı; prompt yazmak kadar, veriyle kanıtlama, araçlarla destekleme, muhakemeyi görünür kılma ve çıktıyı ölçülebilir hale getirme işidir.&lt;/strong&gt;&lt;/p&gt;
&lt;/div&gt;

&lt;!-- ======================== METİNLER (EK) ======================== --&gt;
&lt;h2 class="section-title"&gt;&#128221; Eğitimde Kullanılan Test Metinleri&lt;/h2&gt;

&lt;div class="level-card"&gt;
  &lt;p&gt;Bu iki metin özellikle seçilmiştir: &lt;strong&gt;kelime örtüşmesi düşük&lt;/strong&gt; (düşük ROUGE), ancak &lt;strong&gt;anlamsal olarak benzer mesaj&lt;/strong&gt; vermektedirler.&lt;/p&gt;

  &lt;div class="scenario"&gt;
    &lt;strong&gt;Metin 1 (Referans):&lt;/strong&gt;&lt;br&gt;
    "Yapay zeka modellerinin eğitilmesi sürecinde, yüksek kaliteli veri setlerinin kullanılması kritik bir öneme sahiptir. Eğer veri seti hatalı, yanlı veya eksik bilgiler içeriyorsa, modelin üreteceği sonuçlar da kaçınılmaz olarak kusurlu ve güvenilmez olacaktır. Bu nedenle veri temizliği, model mimarisinin karmaşıklığından bile daha öncelikli bir adımdır."
  &lt;/div&gt;

  &lt;div class="scenario"&gt;
    &lt;strong&gt;Metin 2 (Sistemden Gelen Çıktı):&lt;/strong&gt;&lt;br&gt;
    "Makine öğrenmesi algoritmalarının başarısı, büyük ölçüde beslendikleri bilginin kalitesine bağlıdır. Kirli, önyargılı ya da noksan verilerle beslenen algoritmalar, doğal olarak yanlış ve itimat edilemez çıktılar verecektir. Dolayısıyla, verilerin filtrelenmesi ve düzenlenmesi, sistemin altyapısını kurmaktan çok daha elzem bir süreçtir."
  &lt;/div&gt;
&lt;/div&gt;

&lt;!-- ======================== SEO HASHTAGLERİ ======================== --&gt;
&lt;div class="level-card" style="background: linear-gradient(135deg, #fefce8 0%, #fef9c3 100%); border-color: #fde047;"&gt;
  &lt;h4 style="margin: 0 0 12px 0; color: #854d0e; font-size: 1rem;"&gt;&#127991;️ Etiketler &amp;amp; Hashtagler&lt;/h4&gt;
  &lt;p style="margin: 0; font-size: 0.9rem; color: #713f12; line-height: 1.8;"&gt;
    #YapayZeka #AI #MachineLearning #DeepLearning #LLM #LargeLanguageModel #PromptMühendisliği #PromptEngineering #AgentDesign #ArtificialIntelligence #Ollama #ToolCalling #ReAct #SubAgent #Orchestrator #StructuredOutput #Rubric #FewShot #Grounding #Explainability #RAG #ROUGE #TokenMaliyeti #MuratKarakayaAkademi #YZMühendisliği #YZ #AIEngineering #AIAgents #TechEducation #YouTubeEğitim #Blogger
  &lt;/p&gt;
&lt;/div&gt;

&lt;/body&gt;
&lt;/html&gt;
</description><link>https://www.muratkarakaya.net/2026/06/yapay-zeka-ajannz-neden-basarsz-seviye.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-6141736189068464413</guid><pubDate>Thu, 18 Jun 2026 20:30:24 +0000</pubDate><atom:updated>2026-06-18T23:30:24.285+03:00</atom:updated><category domain="http://www.blogger.com/atom/ns#">Agents</category><category domain="http://www.blogger.com/atom/ns#">Prompt Engineering</category><title>How to Design Real AI Agents (Agents) </title><description>&lt;!DOCTYPE html&gt;
&lt;html lang="en"&gt;
&lt;head&gt;
&lt;meta charset="UTF-8"&gt;
&lt;meta name="viewport" content="width=device-width, initial-scale=1.0"&gt;
&lt;title&gt;How to Design Real AI Agents (Agents) — A Step-by-Step Guide from Level 0 to Level 7&lt;/title&gt;
&lt;style&gt;
/* Blogger-compatible styles — page-level, post content classes */
body {
  font-family: 'Segoe UI', 'Helvetica Neue', Arial, sans-serif;
  line-height: 1.75;
  color: #1e293b;
  max-width: 900px;
  margin: 0 auto;
  padding: 24px;
  background: #f8fafc;
}

.post-header {
  background: linear-gradient(135deg, #1e3a5f 0%, #0f766e 100%);
  color: #ffffff;
  padding: 40px 32px;
  border-radius: 16px;
  margin-bottom: 36px;
  box-shadow: 0 4px 24px rgba(15,118,110,0.18);
}
.post-header h1 {
  font-size: 1.8rem;
  margin: 0 0 12px 0;
  line-height: 1.3;
}
.post-header .subtitle {
  font-size: 1.05rem;
  opacity: 0.92;
  margin: 0;
}
.post-header .meta {
  font-size: 0.85rem;
  opacity: 0.75;
  margin-top: 16px;
}

h2.section-title {
  font-size: 1.45rem;
  color: #1e3a5f;
  border-left: 5px solid #0f766e;
  padding-left: 16px;
  margin-top: 48px;
  margin-bottom: 20px;
}

h3.level-title {
  font-size: 1.2rem;
  color: #0f766e;
  margin-top: 32px;
  margin-bottom: 12px;
}

.level-card {
  background: #ffffff;
  border: 1px solid #e2e8f0;
  border-radius: 12px;
  padding: 24px 28px;
  margin-bottom: 28px;
  box-shadow: 0 1px 6px rgba(0,0,0,0.05);
}
.level-card .level-badge {
  display: inline-block;
  background: #0f766e;
  color: #fff;
  font-weight: 700;
  font-size: 0.8rem;
  padding: 4px 14px;
  border-radius: 999px;
  margin-bottom: 12px;
  letter-spacing: 0.03em;
  text-transform: uppercase;
}
.level-card .level-badge.green { background: #059669; }
.level-card .level-badge.blue  { background: #2563eb; }
.level-card .level-badge.purple{ background: #7c3aed; }
.level-card .level-badge.amber { background: #d97706; }
.level-card .level-badge.red   { background: #dc2626; }
.level-card .level-badge.slate { background: #475569; }

.level-card .scenario {
  background: #f0fdf4;
  border-left: 4px solid #22c55e;
  padding: 12px 16px;
  border-radius: 0 8px 8px 0;
  margin: 16px 0;
  font-size: 0.95rem;
}
.level-card .prompt-box {
  background: #0f172a;
  color: #e2e8f0;
  padding: 16px 20px;
  border-radius: 10px;
  font-family: 'Fira Code', 'Consolas', monospace;
  font-size: 0.85rem;
  overflow-x: auto;
  margin: 16px 0;
  white-space: pre-wrap;
  line-height: 1.6;
}
.level-card .why-better {
  background: #eff6ff;
  border-left: 4px solid #3b82f6;
  padding: 12px 16px;
  border-radius: 0 8px 8px 0;
  margin: 16px 0;
  font-size: 0.95rem;
}
.level-card .why-better ul,
.level-card .why-better ol {
  margin: 8px 0 0 20px;
  padding: 0;
}
.level-card .why-better li {
  margin-bottom: 6px;
}
.level-card .missing {
  background: #fffbeb;
  border-left: 4px solid #f59e0b;
  padding: 12px 16px;
  border-radius: 0 8px 8px 0;
  margin: 16px 0;
  font-size: 0.95rem;
}

table.comparison {
  width: 100%;
  border-collapse: collapse;
  margin: 28px 0;
  font-size: 0.88rem;
}
table.comparison th {
  background: #1e3a5f;
  color: #fff;
  padding: 12px 14px;
  text-align: left;
}
table.comparison td {
  border: 1px solid #e2e8f0;
  padding: 10px 14px;
  vertical-align: top;
}
table.comparison tr:nth-child(even) td {
  background: #f8fafc;
}

.text-center { text-align: center; }
.mt-2 { margin-top: 16px; }
.mb-1 { margin-bottom: 8px; }

.code-block {
  background: #0f172a;
  color: #e2e8f0;
  padding: 16px 20px;
  border-radius: 10px;
  font-family: 'Fira Code', 'Consolas', monospace;
  font-size: 0.83rem;
  overflow-x: auto;
  white-space: pre;
  line-height: 1.55;
  margin: 16px 0;
}

.callout {
  background: linear-gradient(135deg, #eff6ff 0%, #f0fdf4 100%);
  border: 1px solid #bfdbfe;
  border-radius: 12px;
  padding: 24px;
  margin: 28px 0;
}
.callout h4 {
  margin: 0 0 10px 0;
  color: #1e3a5f;
}
.callout p {
  margin: 0;
  font-size: 0.95rem;
}

.final-cta {
  background: linear-gradient(135deg, #1e3a5f 0%, #0f766e 100%);
  color: #fff;
  border-radius: 14px;
  padding: 32px;
  margin: 40px 0 24px 0;
  text-align: center;
}
.final-cta h3 {
  color: #fff;
  margin: 0 0 12px 0;
}
.final-cta p {
  opacity: 0.92;
  font-size: 1rem;
  margin: 0;
}

.rubric-list {
  list-style: none;
  padding: 0;
  margin: 12px 0;
}
.rubric-list li {
  background: #f8fafc;
  border: 1px solid #e2e8f0;
  border-radius: 8px;
  padding: 10px 16px;
  margin-bottom: 8px;
  font-size: 0.95rem;
}
.rubric-list li strong {
  color: #0f766e;
}

.success-message {
  background: #ecfdf5;
  border: 1px solid #a7f3d0;
  border-radius: 10px;
  padding: 18px 22px;
  margin: 20px 0;
  text-align: center;
  font-weight: 600;
  color: #065f46;
}

@media (max-width: 600px) {
  body { padding: 12px; }
  .post-header { padding: 24px 16px; }
  .post-header h1 { font-size: 1.35rem; }
  .level-card { padding: 16px 18px; }
  table.comparison { font-size: 0.78rem; }
}
&lt;/style&gt;
&lt;/head&gt;
&lt;body&gt;

&lt;!-- ======================== HEADER ======================== --&gt;
&lt;div class="post-header"&gt;
  &lt;h1&gt;How to Design Real AI Agents (Agents)?&lt;/h1&gt;
  &lt;p class="subtitle"&gt;&lt;strong&gt;Transitioning from "Persona" to Tool-Using Decision Mechanisms&lt;/strong&gt;&lt;br&gt;
  A step-by-step senior-level AI Engineer guide from Level 0 to Level 7.&lt;/p&gt;
  &lt;p class="meta"&gt;Published: June 2026 &amp;nbsp;|&amp;nbsp; Reading time: ~25 min &amp;nbsp;|&amp;nbsp; Category: AI Engineering&lt;/p&gt;
&lt;/div&gt;

&lt;!-- ======================== INTRODUCTION ======================== --&gt;
&lt;h2 class="section-title"&gt;&#127919; Introduction: Why "You Are the Best AI Engineer in the World" Is Not Enough&lt;/h2&gt;

&lt;div class="level-card"&gt;
  &lt;p&gt;90% of online tutorials tell you to write this to create an agent:&lt;/p&gt;
  &lt;div class="prompt-box"&gt;"You are the best software engineer in the world..."&lt;/div&gt;
  &lt;p&gt;If you are deploying an application to production, prompts written with only persona will turn your system into a &lt;strong&gt;hallucination bomb&lt;/strong&gt;.&lt;/p&gt;
  &lt;p&gt;The real power of agents does not come from the roles you assign them; it comes from &lt;strong&gt;Tool Calling&lt;/strong&gt;, &lt;strong&gt;Grounding&lt;/strong&gt;, and &lt;strong&gt;Explainability&lt;/strong&gt;.&lt;/p&gt;

  &lt;div class="callout"&gt;
    &lt;h4&gt;&#128161; What Will You Learn in This Post?&lt;/h4&gt;
    &lt;p&gt;In this post we will take a simple problem — &lt;strong&gt;"measuring text similarity"&lt;/strong&gt; — and show you step by step how to evolve it from a novice approach to an autonomous level designed by a senior AI Engineer. At each level we will compare prompts, architecture, and cost.&lt;/p&gt;
  &lt;/div&gt;

  &lt;div class="callout" style="background: linear-gradient(135deg, #fef2f2 0%, #fff7ed 100%); border-color: #fca5a5;"&gt;
    &lt;h4&gt;&#127916; Learn with Video: Murat Karakaya Akademi&lt;/h4&gt;
    &lt;p&gt;You can also watch this training on the &lt;strong&gt;Murat Karakaya Akademi&lt;/strong&gt; YouTube channel. Follow the same journey from Level 0 to Level 7 with step-by-step live demos, code explanations, and architectural analysis.&lt;/p&gt;
    &lt;p class="mt-2"&gt;&lt;a href="https://youtube.com/live/cM4Bkon5q4U" target="_blank" rel="noopener" style="display: inline-block; background: #dc2626; color: #fff; padding: 10px 24px; border-radius: 8px; text-decoration: none; font-weight: 700; font-size: 0.95rem;"&gt;▶ Watch on YouTube — Murat Karakaya Akademi&lt;/a&gt;&lt;/p&gt;
  &lt;/div&gt;
&lt;/div&gt;

&lt;!-- ======================== LEVEL 0 ======================== --&gt;
&lt;h2 class="section-title"&gt;&#128204; Level 0 — Novice Approach: Black Box &amp;amp; Hallucination&lt;/h2&gt;

&lt;div class="level-card"&gt;
  &lt;span class="level-badge red"&gt;Beginner&lt;/span&gt;

  &lt;p&gt;&lt;strong&gt;Scenario:&lt;/strong&gt; We need to find how similar Text 1 and Text 2 are on a &lt;strong&gt;0–100&lt;/strong&gt; scale. (For example, in a RAG system, we measure how faithfully a generated answer stays true to the reference text).&lt;/p&gt;

  &lt;div class="scenario"&gt;
    &lt;strong&gt;Novice Prompt:&lt;/strong&gt;&lt;br&gt;
    "You are a text analyst. Compare the given Text 1 and Text 2 and decide how similar they are. Return the similarity as a numeric value between 0–100. 0: They do not match at all, 100: They are completely identical word for word."
  &lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Why Is This Bad?&lt;/strong&gt;&lt;/p&gt;
  &lt;div class="why-better"&gt;
    &lt;ol&gt;
      &lt;li&gt;&lt;strong&gt;Black Box:&lt;/strong&gt; The model only returns "75". Why 75? Not 74 or 80? We don't know.&lt;/li&gt;
      &lt;li&gt;&lt;strong&gt;Subjectivity:&lt;/strong&gt; The same texts might get 60 one day and 85 the next.&lt;/li&gt;
      &lt;li&gt;&lt;strong&gt;Hallucination Risk:&lt;/strong&gt; LLMs cannot perform mathematical measurements; they predict words. "75" is a fabricated (hallucinated) value.&lt;/li&gt;
    &lt;/ol&gt;
  &lt;/div&gt;
&lt;/div&gt;

&lt;!-- ======================== LEVEL 1 ======================== --&gt;
&lt;h2 class="section-title"&gt;&#128204; Level 1 — Structured Output &amp;amp; Chain of Thought&lt;/h2&gt;

&lt;div class="level-card"&gt;
  &lt;span class="level-badge blue"&gt;Improvement&lt;/span&gt;

  &lt;p&gt;The first thing we want is &lt;strong&gt;Explainability&lt;/strong&gt;. We tell the model: "Don't just give me the result, show me your thinking process."&lt;/p&gt;

  &lt;div class="prompt-box"&gt;"You are a text analyst. Your goal is to measure the semantic similarity between two texts.

Your output MUST be a valid JSON object. Do not use Markdown or additional text.

Required JSON fields:
- reasoning: The specific justification for similarity and differences between these two texts.
- similarity_score: An integer between 0-100.

Expected JSON schema:
{
  &amp;quot;type&amp;quot;: &amp;quot;object&amp;quot;,
  &amp;quot;required&amp;quot;: [&amp;quot;reasoning&amp;quot;, &amp;quot;similarity_score&amp;quot;],
  &amp;quot;properties&amp;quot;: {
    &amp;quot;reasoning&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
    &amp;quot;similarity_score&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;integer&amp;quot;, &amp;quot;minimum&amp;quot;: 0, &amp;quot;maximum&amp;quot;: 100}
  }
}"&lt;/div&gt;

  &lt;div class="why-better"&gt;
    &lt;strong&gt;Why Is This Better?&lt;/strong&gt;
    &lt;ul&gt;
      &lt;li&gt;&lt;strong&gt;Chain of Thought:&lt;/strong&gt; We did not ask for the score first. We asked it to fill the reasoning field first. The model justifies its own inference while writing the explanation.&lt;/li&gt;
      &lt;li&gt;&lt;strong&gt;Debugging:&lt;/strong&gt; If the score is 40, we can look at the logs and say "The model missed this detail, that's why it gave a low score."&lt;/li&gt;
      &lt;li&gt;&lt;strong&gt;Integration:&lt;/strong&gt; The JSON output can be parsed by application code.&lt;/li&gt;
    &lt;/ul&gt;
  &lt;/div&gt;

  &lt;div class="missing"&gt;
    &lt;strong&gt;⚠️ What Is Missing:&lt;/strong&gt; The score is still subjective. The 0–100 scale is too broad and unstructured for an LLM.
  &lt;/div&gt;
&lt;/div&gt;

&lt;!-- ======================== LEVEL 2 ======================== --&gt;
&lt;h2 class="section-title"&gt;&#128204; Level 2 — Rubric and Objectivity (Divide &amp;amp; Conquer)&lt;/h2&gt;

&lt;div class="level-card"&gt;
  &lt;span class="level-badge blue"&gt;Increased Autonomy&lt;/span&gt;

  &lt;p&gt;We are not allowing the agent to interpret the abstract concept of "similarity" on its own. We give it &lt;strong&gt;rules (rubric)&lt;/strong&gt;. We break the problem into parts.&lt;/p&gt;

  &lt;p&gt;&lt;strong&gt;Rubric:&lt;/strong&gt;&lt;/p&gt;
  &lt;ul class="rubric-list"&gt;
    &lt;li&gt;&lt;strong&gt;1. Main Idea (0-20):&lt;/strong&gt; Do both texts convey the same core message?&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;2. Tone and Style (0-20):&lt;/strong&gt; Do the texts have the same formality and emotion?&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;3. Entities (0-20):&lt;/strong&gt; Do the names, dates, and numbers in the texts match?&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;4. Missing Information (0-20):&lt;/strong&gt; Does Text 2 omit an important detail from Text 1?&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;5. Fluency (0-20):&lt;/strong&gt; How structurally coherent is Text 2?&lt;/li&gt;
  &lt;/ul&gt;

  &lt;div class="prompt-box"&gt;"You are an expert evaluator. Evaluate the two texts according to the 5 criteria below. Each criterion is scored 0-20.

Your output MUST be a valid JSON object. Do not use Markdown or additional text.

Required JSON fields:
- evaluations: A list of 5 items [criterion, score, explanation]
- total_score: The sum of the five criterion scores.

Expected JSON schema:
{
  &amp;quot;type&amp;quot;: &amp;quot;object&amp;quot;,
  &amp;quot;required&amp;quot;: [&amp;quot;evaluations&amp;quot;, &amp;quot;total_score&amp;quot;],
  &amp;quot;properties&amp;quot;: {
    &amp;quot;evaluations&amp;quot;: {
      &amp;quot;type&amp;quot;: &amp;quot;array&amp;quot;,
      &amp;quot;minItems&amp;quot;: 5,
      &amp;quot;maxItems&amp;quot;: 5,
      &amp;quot;items&amp;quot;: {
        &amp;quot;type&amp;quot;: &amp;quot;object&amp;quot;,
        &amp;quot;required&amp;quot;: [&amp;quot;criterion&amp;quot;, &amp;quot;score&amp;quot;, &amp;quot;explanation&amp;quot;],
        &amp;quot;properties&amp;quot;: {
          &amp;quot;criterion&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
          &amp;quot;score&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;integer&amp;quot;, &amp;quot;minimum&amp;quot;: 0, &amp;quot;maximum&amp;quot;: 20},
          &amp;quot;explanation&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;}
        }
      }
    },
    &amp;quot;total_score&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;integer&amp;quot;, &amp;quot;minimum&amp;quot;: 0, &amp;quot;maximum&amp;quot;: 100}
  }
}"&lt;/div&gt;

  &lt;div class="why-better"&gt;
    &lt;strong&gt;Why Is This Better?&lt;/strong&gt;
    &lt;ul&gt;
      &lt;li&gt;&lt;strong&gt;Grounding:&lt;/strong&gt; We removed the model from abstract evaluation on a scale of 100.&lt;/li&gt;
      &lt;li&gt;&lt;strong&gt;Objectivity:&lt;/strong&gt; Scores will be much more consistent even if you run them at different times (variance decreases).&lt;/li&gt;
      &lt;li&gt;&lt;strong&gt;Comparability:&lt;/strong&gt; We can evaluate different prompts against the same rubric.&lt;/li&gt;
    &lt;/ul&gt;
  &lt;/div&gt;
&lt;/div&gt;

&lt;!-- ======================== LEVEL 3 ======================== --&gt;
&lt;h2 class="section-title"&gt;&#128204; Level 3 — Example-Based Rubric Calibration (One-Shot / Few-Shot)&lt;/h2&gt;

&lt;div class="level-card"&gt;
  &lt;span class="level-badge amber"&gt;Calibration&lt;/span&gt;

  &lt;p&gt;In Level 2 we gave the rubric; however, this was still &lt;strong&gt;zero-shot prompting&lt;/strong&gt;. The model read the criteria but never saw from examples what "When do I give 20 points?", "What is the boundary case for 10?", "When is 0 appropriate?"&lt;/p&gt;

  &lt;p&gt;At this level we provide small, representative-examples for each criterion.&lt;/p&gt;

  &lt;p&gt;&lt;strong&gt;Scoring Calibration Examples:&lt;/strong&gt;&lt;/p&gt;

  &lt;ul class="rubric-list"&gt;
    &lt;li&gt;&lt;strong&gt;Main Idea — 20 points:&lt;/strong&gt; "Data cleaning is critical for model success" and "Model quality heavily depends on clean data" convey the same core message.&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;Main Idea — 10 points:&lt;/strong&gt; Both texts discuss data quality but one focuses on security risks while the other focuses on model performance.&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;Main Idea — 0 points:&lt;/strong&gt; One discusses data cleaning, the other discusses a sports match result.&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;Tone and Style — 20 points:&lt;/strong&gt; Both texts are written in an academic and formal tone.&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;Tone and Style — 10 points:&lt;/strong&gt; One is formal, the other more conversational, but the meaning is preserved.&lt;/li&gt;
  &lt;/ul&gt;

  &lt;div class="why-better"&gt;
    &lt;strong&gt;Why Is This Better?&lt;/strong&gt;
    &lt;ul&gt;
      &lt;li&gt;&lt;strong&gt;Consistency:&lt;/strong&gt; The model uses the same score ranges more reliably.&lt;/li&gt;
      &lt;li&gt;&lt;strong&gt;Teachability:&lt;/strong&gt; The rubric is now supported by behavioral examples, not just an abstract list.&lt;/li&gt;
      &lt;li&gt;&lt;strong&gt;Cost:&lt;/strong&gt; Few-shot examples increase input tokens — so examples must be short and clear.&lt;/li&gt;
    &lt;/ul&gt;
  &lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Expected JSON Schema:&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="code-block"&gt;{
  &amp;quot;type&amp;quot;: &amp;quot;object&amp;quot;,
  &amp;quot;required&amp;quot;: [&amp;quot;evaluations&amp;quot;, &amp;quot;calibration_note&amp;quot;, &amp;quot;total_score&amp;quot;],
  &amp;quot;properties&amp;quot;: {
    &amp;quot;evaluations&amp;quot;: {
      &amp;quot;type&amp;quot;: &amp;quot;array&amp;quot;,
      &amp;quot;minItems&amp;quot;: 5,
      &amp;quot;maxItems&amp;quot;: 5,
      &amp;quot;items&amp;quot;: {
        &amp;quot;type&amp;quot;: &amp;quot;object&amp;quot;,
        &amp;quot;required&amp;quot;: [&amp;quot;criterion&amp;quot;, &amp;quot;score&amp;quot;, &amp;quot;explanation&amp;quot;],
        &amp;quot;properties&amp;quot;: {
          &amp;quot;criterion&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
          &amp;quot;score&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;integer&amp;quot;, &amp;quot;minimum&amp;quot;: 0, &amp;quot;maximum&amp;quot;: 20},
          &amp;quot;explanation&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;}
        }
      }
    },
    &amp;quot;calibration_note&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
    &amp;quot;total_score&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;integer&amp;quot;, &amp;quot;minimum&amp;quot;: 0, &amp;quot;maximum&amp;quot;: 100}
  }
}&lt;/div&gt;
&lt;/div&gt;

&lt;!-- ======================== LEVEL 4 ======================== --&gt;
&lt;h2 class="section-title"&gt;&#128204; Level 4 — Tool Calling and Workflow&lt;/h2&gt;

&lt;div class="level-card"&gt;
  &lt;span class="level-badge purple"&gt;Grounding Begins&lt;/span&gt;

  &lt;p&gt;In Level 3 we calibrated the rubric with examples; however, the evaluation still relied solely on LLM interpretation. Now we add &lt;strong&gt;deterministic metrics from external systems&lt;/strong&gt; as evidence.&lt;/p&gt;

  &lt;p&gt;&lt;strong&gt;Metrics Used:&lt;/strong&gt;&lt;/p&gt;
  &lt;ul class="rubric-list"&gt;
    &lt;li&gt;&lt;strong&gt;ROUGE-L F1:&lt;/strong&gt; Measures word-sequence overlap.&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;Lightweight Similarity Score:&lt;/strong&gt; A combination of Token cosine + Token Jaccard + Character 3-gram cosine + Sequence ratio.&lt;/li&gt;
  &lt;/ul&gt;

  &lt;div class="scenario"&gt;
    &lt;strong&gt;Metric Interpretation Rules:&lt;/strong&gt;
    &lt;ul&gt;
      &lt;li&gt;If ROUGE-L F1 is low, this indicates low word-sequence overlap; it is not alone evidence of low semantic similarity.&lt;/li&gt;
      &lt;li&gt;If the Lightweight Similarity Score is higher than ROUGE-L, the texts may convey a similar message with different words.&lt;/li&gt;
      &lt;li&gt;The Lightweight Similarity Score is not a real semantic embedding; it should be used as a decision-support signal, not as the sole decision-maker.&lt;/li&gt;
    &lt;/ul&gt;
  &lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;&#127757; Why Is This Used in the Real World?&lt;/strong&gt;&lt;/p&gt;
  &lt;div class="why-better"&gt;
    &lt;ul&gt;
      &lt;li&gt;&lt;strong&gt;Reliability:&lt;/strong&gt; ROUGE and lightweight similarity scores are deterministic — they give the same scores to the same text pair every time.&lt;/li&gt;
      &lt;li&gt;&lt;strong&gt;Traceability:&lt;/strong&gt; Since the LLM's opinion is grounded in external evidence, evaluation becomes more auditable.&lt;/li&gt;
      &lt;li&gt;&lt;strong&gt;Cost Control:&lt;/strong&gt; Lightweight metrics are fast and do not require heavy model dependencies.&lt;/li&gt;
    &lt;/ul&gt;
  &lt;/div&gt;

  &lt;div class="callout"&gt;
    &lt;h4&gt;&#128300; Experiment Hygiene Note&lt;/h4&gt;
    &lt;p&gt;Level 4 uses the same rubric text as Level 3. This is a deliberate decision: the difference between the two levels is not a rubric change, but only &lt;strong&gt;external metric context&lt;/strong&gt;.&lt;/p&gt;
  &lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Additional Required JSON Fields in Level 4:&lt;/strong&gt;&lt;/p&gt;
  &lt;ul&gt;
    &lt;li&gt;&lt;code&gt;metric_interpretation&lt;/code&gt;: Explain how you interpreted the ROUGE-L F1 and Lightweight Similarity Score values.&lt;/li&gt;
    &lt;li&gt;&lt;code&gt;calibration_note&lt;/code&gt;: Explain how the rubric calibration examples affected your scoring.&lt;/li&gt;
  &lt;/ul&gt;

  &lt;p&gt;&lt;strong&gt;Expected JSON Schema (Level 4):&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="code-block"&gt;{
  &amp;quot;type&amp;quot;: &amp;quot;object&amp;quot;,
  &amp;quot;required&amp;quot;: [&amp;quot;evaluations&amp;quot;, &amp;quot;metric_interpretation&amp;quot;, &amp;quot;calibration_note&amp;quot;, &amp;quot;total_score&amp;quot;],
  &amp;quot;properties&amp;quot;: {
    &amp;quot;evaluations&amp;quot;: { /* Same 5-item list as Level 3 */ },
    &amp;quot;metric_interpretation&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
    &amp;quot;calibration_note&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
    &amp;quot;total_score&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;integer&amp;quot;, &amp;quot;minimum&amp;quot;: 0, &amp;quot;maximum&amp;quot;: 100}
  }
}&lt;/div&gt;
&lt;/div&gt;

&lt;!-- ======================== LEVEL 5 ======================== --&gt;
&lt;h2 class="section-title"&gt;&#128204; Level 5 — ReAct and Ollama Tool Calling with Real Agent Loop&lt;/h2&gt;

&lt;div class="level-card"&gt;
  &lt;span class="level-badge purple"&gt;Real Agentic Loop&lt;/span&gt;

  &lt;p&gt;We built a strong workflow in Level 4, but was our system a real agent? Not exactly. Because we calculated the metrics with Python. In real agent behavior, the &lt;strong&gt;model determines what evidence it needs&lt;/strong&gt;, the software layer executes the tool, and the result returns to the model.&lt;/p&gt;

  &lt;h3 class="mb-1"&gt;ReAct Loop:&lt;/h3&gt;
  &lt;ul&gt;
    &lt;li&gt;&lt;strong&gt;Reasoning:&lt;/strong&gt; The model determines what external evidence it needs to evaluate text similarity.&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;Action:&lt;/strong&gt; Instead of writing plain text, the model produces Ollama's native &lt;code&gt;tool_calls&lt;/code&gt; field.&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;Observation:&lt;/strong&gt; Python executes the relevant function and returns the result to the model as &lt;code&gt;role="tool"&lt;/code&gt;.&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;Final Answer:&lt;/strong&gt; The model uses the tool results to produce the rubric-based JSON evaluation.&lt;/li&gt;
  &lt;/ul&gt;

  &lt;p&gt;&lt;strong&gt;Ollama Native Tool Calling Flow:&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="code-block"&gt;1. Functions are presented to the model via the `tools=[...]` list.
2. If needed, the model produces `response.message.tool_calls`.
3. Python executes these tool calls.
4. Results are added to the conversation as `role="tool"` messages.
5. The model produces the final answer based on tool results.&lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Tool Definitions (Python):&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="code-block"&gt;def calculate_rouge_tool(reference_text: str = "", candidate_text: str = "") -&amp;gt; str:
    """Calculate ROUGE scores."""
    return json.dumps(calculate_rouge(METIN_1, METIN_2), ensure_ascii=False)

def calculate_lightweight_similarity_tool(reference_text: str = "", candidate_text: str = "") -&amp;gt; str:
    """Calculate lightweight similarity score for the fixed Text 1 and Text 2 in the notebook.
    
    Args:
        reference_text: Not considered in training demo; tool uses fixed METIN_1.
        candidate_text: Not considered in training demo; tool uses fixed METIN_2.
    
    Returns:
        JSON string containing Token cosine, Token Jaccard, character 3-gram cosine, sequence ratio,
        and combined score.
    """
    return json.dumps(calculate_lightweight_similarity(METIN_1, METIN_2), ensure_ascii=False)&lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Tool Calling Loop (Pseudo-Python):&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="code-block"&gt;messages = [
    {"role": "system", "content": TOOL_CALLING_SYSTEM_PROMPT},
    {"role": "user", "content": "Text 1: ..., Text 2: ..."},
]

while True:
    response = client.chat(
        model=MODEL,
        messages=messages,
        tools=[calculate_rouge_tool, calculate_lightweight_similarity_tool],
    )
    messages.append(response.message)

    if not response.message.tool_calls:
        print(response.message.content)  # Final answer
        break

    for tool_call in response.message.tool_calls:
        tool_name = tool_call.function.name
        result = available_tools[tool_name](**tool_call.function.arguments)
        messages.append({
            "role": "tool",
            "tool_name": tool_name,
            "content": str(result),
        })&lt;/div&gt;

  &lt;div class="why-better"&gt;
    &lt;strong&gt;Why Is This a Mastery-Level Skill?&lt;/strong&gt;
    &lt;ul&gt;
      &lt;li&gt;ReAct is not just writing &lt;code&gt;Thought / Action / Observation&lt;/code&gt;; it is connecting thought to real tool execution.&lt;/li&gt;
      &lt;li&gt;An agent is the joint design of &lt;strong&gt;prompt&lt;/strong&gt;, &lt;strong&gt;tool calling&lt;/strong&gt;, &lt;strong&gt;execution loop&lt;/strong&gt;, &lt;strong&gt;grounding&lt;/strong&gt;, and &lt;strong&gt;error control&lt;/strong&gt; layers.&lt;/li&gt;
      &lt;li&gt;The model determines tool needs, Python executes the tool, and the final evaluation is supported by external evidence.&lt;/li&gt;
    &lt;/ul&gt;
  &lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;&#128202; Level 5 Token Cost:&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="scenario"&gt;
    &lt;p&gt;&lt;em&gt;"Since Level 5 is a multi-turn agent loop, the input token is not just the length of the first user prompt. With each &lt;code&gt;client.chat&lt;/code&gt; call, the system message, user message, previous assistant messages, and tool results are re-injected into context. Therefore, the input token total in Level 5 is not a unique token count, but a &lt;strong&gt;cumulative processed token / cost indicator&lt;/strong&gt;."&lt;/em&gt;&lt;/p&gt;
  &lt;/div&gt;

  &lt;div class="callout"&gt;
    &lt;h4&gt;&#128737;️ In the Real World: Production Guardrails&lt;/h4&gt;
    &lt;p&gt;In real production, these protections are added: schema validation, maximum step limit, tool allowlist, retry, and tracing/logging.&lt;/p&gt;
  &lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Level 5 Tool Calling JSON Schema:&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="code-block"&gt;{
  &amp;quot;type&amp;quot;: &amp;quot;object&amp;quot;,
  &amp;quot;required&amp;quot;: [&amp;quot;evaluations&amp;quot;, &amp;quot;metric_interpretation&amp;quot;, &amp;quot;calibration_note&amp;quot;, &amp;quot;total_score&amp;quot;],
  &amp;quot;properties&amp;quot;: {
    &amp;quot;evaluations&amp;quot;: {
      &amp;quot;type&amp;quot;: &amp;quot;array&amp;quot;,
      &amp;quot;minItems&amp;quot;: 5,
      &amp;quot;maxItems&amp;quot;: 5,
      &amp;quot;items&amp;quot;: {
        &amp;quot;type&amp;quot;: &amp;quot;object&amp;quot;,
        &amp;quot;required&amp;quot;: [&amp;quot;criterion&amp;quot;, &amp;quot;score&amp;quot;, &amp;quot;explanation&amp;quot;],
        &amp;quot;properties&amp;quot;: {
          &amp;quot;criterion&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
          &amp;quot;score&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;integer&amp;quot;, &amp;quot;minimum&amp;quot;: 0, &amp;quot;maximum&amp;quot;: 20},
          &amp;quot;explanation&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;}
        }
      }
    },
    &amp;quot;metric_interpretation&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
    &amp;quot;calibration_note&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
    &amp;quot;total_score&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;integer&amp;quot;, &amp;quot;minimum&amp;quot;: 0, &amp;quot;maximum&amp;quot;: 100}
  }
}&lt;/div&gt;
&lt;/div&gt;

&lt;!-- ======================== LEVEL 6 ======================== --&gt;
&lt;h2 class="section-title"&gt;&#128204; Level 6 — Rubric-Based Sub-Agents and Python Aggregator&lt;/h2&gt;

&lt;div class="level-card"&gt;
  &lt;span class="level-badge purple"&gt;Modular Architecture&lt;/span&gt;

  &lt;p&gt;In Level 5, a single agent both called tools and interpreted the entire rubric on its own. At this level we try a different architecture: &lt;strong&gt;instead of one large prompt, we give each rubric criterion to a separate sub-agent&lt;/strong&gt;.&lt;/p&gt;

  &lt;h3 class="mb-1"&gt;Architecture:&lt;/h3&gt;
  &lt;ul&gt;
    &lt;li&gt;1 generic sub-agent function is written.&lt;/li&gt;
    &lt;li&gt;This function is called 5 times with 5 different rubric configurations.&lt;/li&gt;
    &lt;li&gt;Each sub-agent evaluates only its own criterion.&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;Python aggregator&lt;/strong&gt; validates, sorts, and calculates the total score.&lt;/li&gt;
    &lt;li&gt;&lt;strong&gt;The aggregator makes no LLM calls&lt;/strong&gt; — it is deterministic.&lt;/li&gt;
  &lt;/ul&gt;

  &lt;p&gt;&lt;strong&gt;The Pedagogical Message of This Level:&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="success-message"&gt;
    ✅ "Don't make everything an agent!"&lt;br&gt;
    LLM is for subjective evaluation. Python is for validation, aggregation, formatting, and deterministic computation.
  &lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Limitations:&lt;/strong&gt;&lt;/p&gt;
  &lt;ul&gt;
    &lt;li&gt;5 sub-agents = 5 LLM calls = more expensive than Level 5.&lt;/li&gt;
    &lt;li&gt;Not necessary in every case; it makes sense when the rubric grows or when audibility is critical.&lt;/li&gt;
  &lt;/ul&gt;

  &lt;p&gt;&lt;strong&gt;Level 6 — Sub-Agent JSON Output (single criterion):&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="code-block"&gt;{
  &amp;quot;criterion&amp;quot;: &amp;quot;Main Idea&amp;quot;,
  &amp;quot;score&amp;quot;: 18,
  &amp;quot;explanation&amp;quot;: &amp;quot;Specific evaluation for this criterion&amp;quot;,
  &amp;quot;evidence&amp;quot;: &amp;quot;Basis from text or metrics&amp;quot;
}&lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Python Aggregator Total Output (all criteria):&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="code-block"&gt;{
  &amp;quot;evaluations&amp;quot;: [
    {&amp;quot;criterion&amp;quot;: &amp;quot;Main Idea&amp;quot;, &amp;quot;score&amp;quot;: 18, &amp;quot;explanation&amp;quot;: &amp;quot;...&amp;quot;, &amp;quot;evidence&amp;quot;: &amp;quot;...&amp;quot;},
    {&amp;quot;criterion&amp;quot;: &amp;quot;Tone and Style&amp;quot;, &amp;quot;score&amp;quot;: 16, &amp;quot;explanation&amp;quot;: &amp;quot;...&amp;quot;, &amp;quot;evidence&amp;quot;: &amp;quot;...&amp;quot;},
    {&amp;quot;criterion&amp;quot;: &amp;quot;Entities&amp;quot;, &amp;quot;score&amp;quot;: 17, &amp;quot;explanation&amp;quot;: &amp;quot;...&amp;quot;, &amp;quot;evidence&amp;quot;: &amp;quot;...&amp;quot;},
    {&amp;quot;criterion&amp;quot;: &amp;quot;Missing Information&amp;quot;, &amp;quot;score&amp;quot;: 15, &amp;quot;explanation&amp;quot;: &amp;quot;...&amp;quot;, &amp;quot;evidence&amp;quot;: &amp;quot;...&amp;quot;},
    {&amp;quot;criterion&amp;quot;: &amp;quot;Fluency&amp;quot;, &amp;quot;score&amp;quot;: 14, &amp;quot;explanation&amp;quot;: &amp;quot;...&amp;quot;, &amp;quot;evidence&amp;quot;: &amp;quot;...&amp;quot;}
  ],
  &amp;quot;metric_interpretation&amp;quot;: &amp;quot;ROUGE and lightweight similarity metrics were provided as context to each sub-agent; each criterion was interpreted by its own expert sub-agent.&amp;quot;,
  &amp;quot;aggregation_note&amp;quot;: &amp;quot;Total score was computed deterministically by the Python aggregator; no LLM orchestrator was used.&amp;quot;,
  &amp;quot;total_score&amp;quot;: 80
}&lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Level 6 Python Aggregator Function Example:&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="code-block"&gt;def aggregate_sub_agent_results(sub_agent_results, metrics_context):
    expected_criteria = [config['criterion'] for config in RUBRIC_AGENT_CONFIGS]
    actual_criteria = [result['parsed'].get('criterion') for result in sub_agent_results]

    if actual_criteria != expected_criteria:
        raise ValueError(f"Sub-agent criterion order does not match expected.")

    evaluations = []
    for result in sub_agent_results:
        parsed = result['parsed']
        score = parsed.get('score')
        if not isinstance(score, int) or not 0 &lt;= score &lt;= 20:
            raise ValueError(f"Invalid score: {parsed}")
        evaluations.append({
            'criterion': parsed['criterion'],
            'score': score,
            'explanation': parsed['explanation'],
            'evidence': parsed['evidence'],
        })

    total_score = sum(item['score'] for item in evaluations)
    return {
        'evaluations': evaluations,
        'metric_interpretation': '...',
        'aggregation_note': 'Total score was computed deterministically by Python.',
        'total_score': total_score,
    }&lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Level 6 — Token Cost:&lt;/strong&gt;&lt;/p&gt;
  &lt;ul&gt;
    &lt;li&gt;5 sub-agents = 5 independent LLM calls&lt;/li&gt;
    &lt;li&gt;Each call includes system prompt + user prompt + metric context&lt;/li&gt;
    &lt;li&gt;Total input token = 5 × (system + user prompt length)&lt;/li&gt;
    &lt;li&gt;Aggregator token cost = 0 (Python code runs)&lt;/li&gt;
  &lt;/ul&gt;

  &lt;p&gt;&lt;strong&gt;Level 6 — Sub-Agent JSON Schema (inside build_sub_agent_system_prompt):&lt;/strong&gt;&lt;/p&gt;

  &lt;div class="code-block"&gt;{
  &amp;quot;type&amp;quot;: &amp;quot;object&amp;quot;,
  &amp;quot;required&amp;quot;: [&amp;quot;criterion&amp;quot;, &amp;quot;score&amp;quot;, &amp;quot;explanation&amp;quot;, &amp;quot;evidence&amp;quot;],
  &amp;quot;properties&amp;quot;: {
    &amp;quot;criterion&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
    &amp;quot;score&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;integer&amp;quot;, &amp;quot;minimum&amp;quot;: 0, &amp;quot;maximum&amp;quot;: 20},
    &amp;quot;explanation&amp;quot;: {&amp;quot;type&amp;quot;: &amp;quot;string&amp;quot;},
    &amp;quot;evidence&amp;quot;: {&amp;quot;type&amp;quot;: [&amp;quot;string&amp;quot;, &amp;quot;array&amp;quot;, &amp;quot;object&amp;quot;]}
  }
}&lt;/div&gt;
&lt;/div&gt;

&lt;!-- ======================== LEVEL 7 ======================== --&gt;
&lt;h2 class="section-title"&gt;&#128204; Level 7 — Orchestrator Agents and When They Are Not Needed&lt;/h2&gt;

&lt;div class="level-card"&gt;
  &lt;span class="level-badge slate"&gt;Architectural Decision&lt;/span&gt;

  &lt;p&gt;After Level 6 the natural question is: &lt;em&gt;"Wouldn't it be better if an orchestrator agent managed these sub-agents?"&lt;/em&gt;&lt;/p&gt;

  &lt;p&gt;&lt;strong&gt;Orchestrator Agent&lt;/strong&gt; is a powerful architecture in the real world. An orchestrator can break down tasks, decide which sub-agent to run, select appropriate tools, initiate retries on missing or contradictory results, and convert results from different agents into a final decision.&lt;/p&gt;

  &lt;p&gt;However, in this example we &lt;strong&gt;deliberately do not need it&lt;/strong&gt;, because:&lt;/p&gt;
  &lt;ul&gt;
    &lt;li&gt;The 5 rubric criteria are predetermined.&lt;/li&gt;
    &lt;li&gt;Every criterion must run.&lt;/li&gt;
    &lt;li&gt;Each sub-agent evaluates only its own criterion.&lt;/li&gt;
    &lt;li&gt;Missing criterion check, sorting, and total score can be reliably done with Python.&lt;/li&gt;
  &lt;/ul&gt;

  &lt;div class="callout"&gt;
    &lt;h4&gt;&#127959;️ When Does an Orchestrator Agent Make Sense?&lt;/h4&gt;
    &lt;ul&gt;
      &lt;li&gt;When which sub-agents to run changes from task to task.&lt;/li&gt;
      &lt;li&gt;When tool selection, data source selection, or workflow branching is needed.&lt;/li&gt;
      &lt;li&gt;When there are contradictions between sub-agent responses and an interpretive reconciliation is needed.&lt;/li&gt;
      &lt;li&gt;When there are dynamic steps such as quality control, retry, missing information completion, or human approval.&lt;/li&gt;
    &lt;/ul&gt;
  &lt;/div&gt;

  &lt;p&gt;&lt;strong&gt;Level 7's Message:&lt;/strong&gt; Orchestrator + sub-agent architectures exist and are important; however, they are not necessary in every problem. In this example, the Python aggregator is the correct, simple, and instructive choice.&lt;/p&gt;

  &lt;div class="success-message"&gt;
    &#128161; &lt;strong&gt;Level 7 Note:&lt;/strong&gt; This level does not make a new LLM call. It is an architectural decision-making and boundary-setting section. Performance graphs show the cost of Level 0-6 experiments.
  &lt;/div&gt;
&lt;/div&gt;

&lt;!-- ======================== COMPARISON TABLE ======================== --&gt;
&lt;h2 class="section-title"&gt;&#128202; Comparison of All Levels&lt;/h2&gt;

&lt;table class="comparison"&gt;
  &lt;thead&gt;
    &lt;tr&gt;
      &lt;th&gt;Level&lt;/th&gt;
      &lt;th&gt;Approach&lt;/th&gt;
      &lt;th&gt;Added Layer&lt;/th&gt;
      &lt;th&gt;Gain&lt;/th&gt;
      &lt;th&gt;Limitation / Lesson&lt;/th&gt;
    &lt;/tr&gt;
  &lt;/thead&gt;
  &lt;tbody&gt;
    &lt;tr&gt;
      &lt;td&gt;&lt;strong&gt;Lvl 0&lt;/strong&gt;&lt;/td&gt;
      &lt;td&gt;Persona / Black Box&lt;/td&gt;
      &lt;td&gt;Simple system prompt&lt;/td&gt;
      &lt;td&gt;Fast start&lt;/td&gt;
      &lt;td&gt;Inconsistent, unexplainable, hallucination-prone&lt;/td&gt;
    &lt;/tr&gt;
    &lt;tr&gt;
      &lt;td&gt;&lt;strong&gt;Lvl 1&lt;/strong&gt;&lt;/td&gt;
      &lt;td&gt;JSON + Explanation&lt;/td&gt;
      &lt;td&gt;Structured output&lt;/td&gt;
      &lt;td&gt;Answer becomes parseable&lt;/td&gt;
      &lt;td&gt;Still subjective&lt;/td&gt;
    &lt;/tr&gt;
    &lt;tr&gt;
      &lt;td&gt;&lt;strong&gt;Lvl 2&lt;/strong&gt;&lt;/td&gt;
      &lt;td&gt;Rubric&lt;/td&gt;
      &lt;td&gt;Criteria-based evaluation&lt;/td&gt;
      &lt;td&gt;More objective score&lt;/td&gt;
      &lt;td&gt;No external evidence, still LLM opinion&lt;/td&gt;
    &lt;/tr&gt;
    &lt;tr&gt;
      &lt;td&gt;&lt;strong&gt;Lvl 3&lt;/strong&gt;&lt;/td&gt;
      &lt;td&gt;One-Shot / Few-Shot Calibration&lt;/td&gt;
      &lt;td&gt;Criteria-based examples&lt;/td&gt;
      &lt;td&gt;More consistent scores&lt;/td&gt;
      &lt;td&gt;Input token cost increases&lt;/td&gt;
    &lt;/tr&gt;
    &lt;tr&gt;
      &lt;td&gt;&lt;strong&gt;Lvl 4&lt;/strong&gt;&lt;/td&gt;
      &lt;td&gt;Workflow + Tools&lt;/td&gt;
      &lt;td&gt;ROUGE + lightweight similarity metrics&lt;/td&gt;
      &lt;td&gt;Grounded, evidence-based&lt;/td&gt;
      &lt;td&gt;Developer selects the tools&lt;/td&gt;
    &lt;/tr&gt;
    &lt;tr&gt;
      &lt;td&gt;&lt;strong&gt;Lvl 5&lt;/strong&gt;&lt;/td&gt;
      &lt;td&gt;ReAct + Tool Calling&lt;/td&gt;
      &lt;td&gt;Automatic tool calling + execution loop&lt;/td&gt;
      &lt;td&gt;Real agent behavior&lt;/td&gt;
      &lt;td&gt;High cost, loop management needed&lt;/td&gt;
    &lt;/tr&gt;
    &lt;tr&gt;
      &lt;td&gt;&lt;strong&gt;Lvl 6&lt;/strong&gt;&lt;/td&gt;
      &lt;td&gt;Sub-Agents + Python Aggregator&lt;/td&gt;
      &lt;td&gt;Generic sub-agent + deterministic aggregation&lt;/td&gt;
      &lt;td&gt;Task decomposition, responsibility separation&lt;/td&gt;
      &lt;td&gt;5× LLM calls, more expensive&lt;/td&gt;
    &lt;/tr&gt;
    &lt;tr&gt;
      &lt;td&gt;&lt;strong&gt;Lvl 7&lt;/strong&gt;&lt;/td&gt;
      &lt;td&gt;Orchestrator Agent Decision&lt;/td&gt;
      &lt;td&gt;Architectural decision-making&lt;/td&gt;
      &lt;td&gt;Understanding of advanced architectures&lt;/td&gt;
      &lt;td&gt;Orchestrator not needed everywhere&lt;/td&gt;
    &lt;/tr&gt;
  &lt;/tbody&gt;
&lt;/table&gt;

&lt;!-- ======================== CLOSING ======================== --&gt;
&lt;div class="final-cta"&gt;
  &lt;h3&gt;&#127891; Final Message: Prompt Engineering Becomes Systems Engineering&lt;/h3&gt;
  &lt;p&gt;Simply giving an agent a powerful-sounding prompt and expecting it to work correctly is not enough.&lt;br&gt;
  A Real AI Agent is not just a model that produces answers; it is a &lt;strong&gt;software system that jointly manages thinking patterns, tool usage, reasoning steps, and data-driven evidence&lt;/strong&gt;.&lt;/p&gt;
  &lt;p class="mt-2" style="opacity:0.8; font-style:italic;"&gt;As we progressed from Level 0 to Level 7, we actually built the same idea layer by layer. First we structured the output, then we broke evaluation into rubric parts, then we calibrated with examples. Then we added grounding with metrics and connected the ReAct idea to real tool execution. Finally, we saw that more advanced architectures like orchestrator agents exist, but in this example, not adding an extra LLM orchestrator was the more correct engineering decision.&lt;/p&gt;
  &lt;p class="mt-2" style="opacity:0.85;"&gt;Our value as AI Engineers emerges here: instead of expecting miracles from the model, understand the model's strengths and weaknesses and build the right architecture around them. &lt;strong&gt;Good agent design is not just about writing prompts; it is about proving with data, supporting with tools, making reasoning visible, and making outputs measurable.&lt;/strong&gt;&lt;/p&gt;
&lt;/div&gt;

&lt;!-- ======================== TEXTS (APPENDIX) ======================== --&gt;
&lt;h2 class="section-title"&gt;&#128221; Test Texts Used in the Training&lt;/h2&gt;

&lt;div class="level-card"&gt;
  &lt;p&gt;These two texts were specifically selected: &lt;strong&gt;low word overlap&lt;/strong&gt; (low ROUGE), yet they &lt;strong&gt;convey a semantically similar message&lt;/strong&gt;.&lt;/p&gt;

  &lt;div class="scenario"&gt;
    &lt;strong&gt;Text 1 (Reference):&lt;/strong&gt;&lt;br&gt;
    "In the process of training artificial intelligence models, the use of high-quality datasets is of critical importance. If the dataset contains incorrect, biased, or incomplete information, the results produced by the model will inevitably be flawed and unreliable. Therefore, data cleaning is a more prioritized step than the complexity of the model architecture."
  &lt;/div&gt;

  &lt;div class="scenario"&gt;
    &lt;strong&gt;Text 2 (System Output):&lt;/strong&gt;&lt;br&gt;
    "The success of machine learning algorithms heavily depends on the quality of the information they are fed. Algorithms fed with dirty, biased, or incomplete data will naturally produce incorrect and untrustworthy outputs. Therefore, filtering and organizing data is a much more essential process than building the system's infrastructure."
  &lt;/div&gt;
&lt;/div&gt;

&lt;!-- ======================== SEO HASHTAGS ======================== --&gt;
&lt;div class="level-card" style="background: linear-gradient(135deg, #fefce8 0%, #fef9c3 100%); border-color: #fde047;"&gt;
  &lt;h4 style="margin: 0 0 12px 0; color: #854d0e; font-size: 1rem;"&gt;&#127991;️ Tags &amp;amp; Hashtags&lt;/h4&gt;
  &lt;p style="margin: 0; font-size: 0.9rem; color: #713f12; line-height: 1.8;"&gt;
    #ArtificialIntelligence #AI #MachineLearning #DeepLearning #LLM #LargeLanguageModel #PromptEngineering #AgentDesign #AI Agents #Ollama #ToolCalling #ReAct #SubAgent #Orchestrator #StructuredOutput #Rubric #FewShot #Grounding #Explainability #RAG #ROUGE #TokenCost #MuratKarakayaAkademi #AIEngineering #TechEducation #YouTubeEducation #Blogger
  &lt;/p&gt;
&lt;/div&gt;

&lt;/body&gt;
&lt;/html&gt;
</description><link>https://www.muratkarakaya.net/2026/06/how-to-design-real-ai-agents-agents.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item><item><guid isPermaLink="false">tag:blogger.com,1999:blog-7258374652086543351.post-7344101731772598060</guid><pubDate>Fri, 12 Jun 2026 18:23:12 +0000</pubDate><atom:updated>2026-06-12T21:23:12.996+03:00</atom:updated><title>Are University Capstone Projects Still Necessary?</title><description>&lt;!DOCTYPE html&gt;
&lt;html lang="en"&gt;
&lt;head&gt;
&lt;meta charset="UTF-8"&gt;
&lt;meta name="viewport" content="width=device-width, initial-scale=1.0"&gt;
&lt;title&gt;Are University Capstone Projects Still Necessary?&lt;/title&gt;
&lt;style&gt;
  /* Container and General Typography */
  body {
    background-color: #f4f7f6;
    margin: 0;
    padding: 20px 0;
  }
  .mk-blog-container {
    font-family: 'Segoe UI', Roboto, Helvetica, Arial, sans-serif;
    line-height: 1.8;
    color: #34495e;
    max-width: 900px;
    margin: 0 auto;
    padding: 40px;
    background-color: #ffffff;
    box-shadow: 0 10px 30px rgba(0,0,0,0.05);
    border-radius: 12px;
  }

  /* Headings */
  .mk-blog-container h1 {
    color: #2c3e50;
    font-size: 2.5em;
    font-weight: 800;
    line-height: 1.2;
    margin-bottom: 20px;
    letter-spacing: -0.5px;
  }
  
  .mk-blog-container h2 {
    color: #2980b9;
    font-size: 1.8em;
    margin-top: 40px;
    margin-bottom: 20px;
    border-bottom: 2px solid #ecf0f1;
    padding-bottom: 10px;
  }

  /* Image */
  .mk-blog-container .hero-image {
    width: 100%;
    height: 450px;
    object-fit: cover;
    border-radius: 12px;
    margin-bottom: 30px;
    box-shadow: 0 5px 15px rgba(0,0,0,0.1);
  }

  /* Lead Paragraph */
  .mk-blog-container .lead {
    font-size: 1.25em;
    color: #555;
    font-weight: 500;
    margin-bottom: 30px;
  }

  /* Paragraphs */
  .mk-blog-container p {
    font-size: 1.1em;
    margin-bottom: 20px;
    text-align: justify;
  }

  /* Highlight Box */
  .mk-blog-container .highlight-box {
    background-color: #f8f9fa;
    border-left: 5px solid #e74c3c;
    padding: 20px 25px;
    margin: 30px 0;
    border-radius: 0 8px 8px 0;
    font-style: italic;
    color: #2c3e50;
  }

  /* Custom List Cards (Problems) */
  .problem-grid {
    display: flex;
    flex-direction: column;
    gap: 20px;
    margin: 40px 0;
  }
  .problem-card {
    background: #ffffff;
    border: 1px solid #e0e6ed;
    border-radius: 10px;
    padding: 25px;
    box-shadow: 0 4px 6px rgba(0,0,0,0.02);
    transition: all 0.3s ease;
  }
  .problem-card:hover {
    box-shadow: 0 8px 15px rgba(0,0,0,0.08);
    transform: translateY(-3px);
    border-color: #bdc3c7;
  }
  .problem-card h3 {
    margin-top: 0;
    color: #c0392b;
    display: flex;
    align-items: center;
    font-size: 1.4em;
    margin-bottom: 15px;
  }
  .problem-icon {
    font-size: 1.5em;
    margin-right: 12px;
  }

  /* Questions Section */
  .questions-section {
    background: linear-gradient(to right bottom, #2c3e50, #3498db);
    color: white;
    padding: 40px;
    border-radius: 12px;
    margin: 50px 0;
  }
  .questions-section h2 {
    color: #ffffff;
    border-bottom: none;
    margin-top: 0;
    text-align: center;
    font-size: 2em;
  }
  .question-item {
    background: rgba(255,255,255,0.1);
    padding: 20px;
    border-radius: 8px;
    margin-bottom: 15px;
    display: flex;
    align-items: flex-start;
  }
  .question-number {
    background: #e74c3c;
    color: white;
    font-weight: bold;
    width: 35px;
    height: 35px;
    display: flex;
    align-items: center;
    justify-content: center;
    border-radius: 50%;
    margin-right: 20px;
    flex-shrink: 0;
    font-size: 1.1em;
  }
  .question-text {
    font-size: 1.15em;
    line-height: 1.6;
    font-weight: 500;
  }

  /* Tags */
  .tags {
    display: flex;
    flex-wrap: wrap;
    gap: 10px;
    margin-top: 50px;
    padding-top: 30px;
    border-top: 1px solid #eee;
  }
  .tag {
    background: #ecf0f1;
    color: #7f8c8d;
    padding: 6px 14px;
    border-radius: 20px;
    font-size: 0.85em;
    font-weight: 600;
    text-decoration: none;
  }
  .tag:hover {
    background: #3498db;
    color: white;
  }

  /* Responsive */
  @media (max-width: 768px) {
    .mk-blog-container {
      padding: 20px;
    }
    .mk-blog-container h1 {
      font-size: 2em;
    }
    .questions-section {
      padding: 20px;
    }
  }
&lt;/style&gt;
&lt;/head&gt;
&lt;body&gt;

&lt;div class="mk-blog-container"&gt;
  
  &lt;img class="hero-image" src="https://images.unsplash.com/photo-1522071820081-009f0129c71c?q=80&amp;w=1600&amp;auto=format&amp;fit=crop" alt="Students and Computer - Capstone Project"&gt;

  &lt;h1&gt;&#127891; Are University Capstone Projects "Obsolete"? What Are We Really Grading?&lt;/h1&gt;

  &lt;p class="lead"&gt;
    &#127963; In university engineering faculties, we are facing a staggering situation regarding "capstone projects"—often seen as the pinnacle of a student's educational life and where they are expected to put in serious effort for one or two semesters—that we can no longer ignore.
  &lt;/p&gt;

  &lt;p&gt;
    Engineering education, by its nature, aims to impart skills in problem-solving, analytical thinking, and designing a system from scratch. Traditionally, the capstone project is the ultimate test where a student puts all the theoretical knowledge acquired over 4 years into practice and proves they can stand on their own feet. I certainly exclude our students who focus on their projects ethically, with genuine hard work, turning night into day. However, the general landscape in the field and during jury defenses shows that we are &lt;strong&gt;frighteningly rapidly moving away from the core purpose of engineering education&lt;/strong&gt;.
  &lt;/p&gt;

  &lt;div class="highlight-box"&gt;
    The issue has now moved far beyond simply copying ready-made code snippets from StackOverflow or GitHub. The process now includes the unchecked use of artificial intelligence tools in a way that "adds zero vision and skill to the student."
  &lt;/div&gt;

  &lt;h2&gt;The Next-Generation Landscape Juries Face&lt;/h2&gt;
  &lt;p&gt;
    When you sit down to evaluate a project, the anatomy of the work presented to you now looks very similar. We can group the flaws of this new era under three main headings:
  &lt;/p&gt;

  &lt;div class="problem-grid"&gt;
    
    &lt;div class="problem-card"&gt;
      &lt;h3&gt;&lt;span class="problem-icon"&gt;&#129302;&lt;/span&gt; Frankenstein Projects&lt;/h3&gt;
      &lt;p&gt;
        When you open the hood and look inside projects that seem highly complicated and appear to use modern architectures from the outside or on the presentation screen, you encounter utter chaos. In reality, there is no "engineering design"; you see that many ready-made infrastructures are somehow "stitched together" using Large Language Models (LLMs) or API requests. There is no logical coherence between code blocks. Systems that fully meet no technical requirements (performance, security, scalability) and run almost by &lt;strong&gt;pure luck&lt;/strong&gt; are brought before the jury as a "success."
      &lt;/p&gt;
    &lt;/div&gt;

    &lt;div class="problem-card"&gt;
      &lt;h3&gt;&lt;span class="problem-icon"&gt;&#128196;&lt;/span&gt; Candidates Alien to Their Own Reports&lt;/h3&gt;
      &lt;p&gt;
        In the past, even if projects were incomplete, the student knew where they made a mistake in the code they wrote or the circuit they built. Today, however, when we ask students about the most fundamental components of their project or their engineering choices at critical decision points, they cannot explain them. Questions like &lt;em&gt;"Why did you choose this algorithm?"&lt;/em&gt; or &lt;em&gt;"How will this database architecture handle this load?"&lt;/em&gt; remain unanswered. The saddest part is realizing they haven't even read or understood the statements, numbers, and tables in the brilliantly academic-sounding report (written by AI) they claim to have "prepared themselves."
      &lt;/p&gt;
    &lt;/div&gt;

    &lt;div class="problem-card"&gt;
      &lt;h3&gt;&lt;span class="problem-icon"&gt;&#128735;&lt;/span&gt; Passing the Buck in Defenses&lt;/h3&gt;
      &lt;p&gt;
        An engineer is someone who puts their signature under their work and takes responsibility. Yet, the answers we get to our questions during the jury usually point to a "third party": &lt;strong&gt;"It wrote it like this"&lt;/strong&gt;, &lt;strong&gt;"It gave these results"&lt;/strong&gt;, &lt;strong&gt;"When I fed the data to the system, it generated this graph"&lt;/strong&gt;. We frequently and sadly see that AI has ceased to be a "tool" that accelerates the learning process and has become a "subcontractor" onto which all cognitive action and responsibility are dumped.
      &lt;/p&gt;
    &lt;/div&gt;

  &lt;/div&gt;

  &lt;p&gt;
    We could multiply these examples for pages with anecdotes from the field. Artificial intelligence will not leave our lives; on the contrary, it will become even more integrated. Rejecting this is nothing more than Luddism. However, there are crucial questions that academia and the industry need to ask themselves and urgently open up for discussion.
  &lt;/p&gt;

  &lt;div class="questions-section"&gt;
    &lt;h2&gt;❓️ The Big Questions&lt;/h2&gt;
    
    &lt;div class="question-item"&gt;
      &lt;div class="question-number"&gt;1&lt;/div&gt;
      &lt;div class="question-text"&gt;
        Under these circumstances, how realistic and useful to them is it to still give students &lt;strong&gt;2 semesters of time&lt;/strong&gt; for projects that an AI assistant can prototype in a few hours, and to take the resulting code seriously and grade it at the end of these processes?
      &lt;/div&gt;
    &lt;/div&gt;

    &lt;div class="question-item"&gt;
      &lt;div class="question-number"&gt;2&lt;/div&gt;
      &lt;div class="question-text"&gt;
        Can our traditional evaluation methods (reading reports, listening to slide presentations) truly measure the student's &lt;strong&gt;engineering reflexes, problem-solving skills, and ethical understanding&lt;/strong&gt; in an age where AI "solves" and perfectly packages everything?
      &lt;/div&gt;
    &lt;/div&gt;

    &lt;div class="question-item"&gt;
      &lt;div class="question-number"&gt;3&lt;/div&gt;
      &lt;div class="question-text"&gt;
        If the professors on the jury are also using another AI to read the reports, at the end of the day, are we grading the student's capacity or &lt;strong&gt;the outputs of the AI model (ChatGPT, Claude, etc.) they used?&lt;/strong&gt;
      &lt;/div&gt;
    &lt;/div&gt;

    &lt;div class="question-item" style="background: rgba(231, 76, 60, 0.4); border: 1px solid #e74c3c;"&gt;
      &lt;div class="question-number" style="background: white; color: #e74c3c;"&gt;4&lt;/div&gt;
      &lt;div class="question-text" style="font-weight: 700;"&gt;
        So what is the solution? How should we update this system? How can we transition to evaluating students as individuals who "design and manage systems" rather than just "writing code"?
      &lt;/div&gt;
    &lt;/div&gt;
  &lt;/div&gt;

  &lt;p&gt;
    We must approach this problem not with outdated rules like "ban artificial intelligence," but by completely changing our evaluation metrics. Perhaps capstone projects should no longer be months-long coding assignments; they should transform into &lt;strong&gt;System Design interviews&lt;/strong&gt; where the candidate defends an architecture by drawing it on a board live in front of the jury and explains how they used AI "correctly and ethically."
  &lt;/p&gt;

  &lt;p style="font-weight: bold; color: #2c3e50; font-size: 1.1em; text-align: center; margin-top: 40px;"&gt;
    I am very curious about the observations, experiences, and most importantly, the proposed solutions of my esteemed colleagues in academia and the industry regarding this issue. How do you think we will get out of this situation? Let's meet in the comments! &#128071;
  &lt;/p&gt;

  &lt;div class="tags"&gt;
    &lt;a href="#" class="tag"&gt;#MuratKarakayaAkademi&lt;/a&gt;
    &lt;a href="#" class="tag"&gt;#Engineering&lt;/a&gt;
    &lt;a href="#" class="tag"&gt;#ArtificialIntelligence&lt;/a&gt;
    &lt;a href="#" class="tag"&gt;#Academia&lt;/a&gt;
    &lt;a href="#" class="tag"&gt;#Education&lt;/a&gt;
    &lt;a href="#" class="tag"&gt;#CapstoneProject&lt;/a&gt;
    &lt;a href="#" class="tag"&gt;#HigherEducation&lt;/a&gt;
    &lt;a href="#" class="tag"&gt;#EngineeringEthics&lt;/a&gt;
    &lt;a href="#" class="tag"&gt;#FutureEngineers&lt;/a&gt;
  &lt;/div&gt;

&lt;/div&gt;

&lt;/body&gt;
&lt;/html&gt;</description><link>https://www.muratkarakaya.net/2026/06/are-university-capstone-projects-still.html</link><author>noreply@blogger.com (Murat Karakaya)</author><thr:total>0</thr:total></item></channel></rss>