<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"
	xmlns:content="http://purl.org/rss/1.0/modules/content/"
	xmlns:wfw="http://wellformedweb.org/CommentAPI/"
	xmlns:dc="http://purl.org/dc/elements/1.1/"
	xmlns:atom="http://www.w3.org/2005/Atom"
	xmlns:sy="http://purl.org/rss/1.0/modules/syndication/"
	xmlns:slash="http://purl.org/rss/1.0/modules/slash/"
	>

<channel>
	<title>Gea-Suan Lin&#039;s BLOG</title>
	<atom:link href="https://blog.gslin.org/feed/" rel="self" type="application/rss+xml" />
	<link>https://blog.gslin.org</link>
	<description>幹壞事是進步最大的原動力</description>
	<lastBuildDate>Wed, 29 Jul 2026 01:06:30 +0000</lastBuildDate>
	<language>en-US</language>
	<sy:updatePeriod>
	hourly	</sy:updatePeriod>
	<sy:updateFrequency>
	1	</sy:updateFrequency>
	<generator>https://wordpress.org/?v=7.0.2</generator>
<atom:link rel="hub" href="https://pubsubhubbub.appspot.com"/>
<atom:link rel="hub" href="https://pubsubhubbub.superfeedr.com"/>
<atom:link rel="hub" href="https://websubhub.com/hub"/>
<atom:link rel="self" href="https://blog.gslin.org/feed/"/>
<site xmlns="com-wordpress:feed-additions:1">21326247</site>	<item>
		<title>Hacker News 上看到運動褲用的繩結 Slipped Lapp Knot</title>
		<link>https://blog.gslin.org/archives/2026/07/29/13130/hacker-news-%e4%b8%8a%e7%9c%8b%e5%88%b0%e9%81%8b%e5%8b%95%e8%a4%b2%e7%94%a8%e7%9a%84%e7%b9%a9%e7%b5%90-slipped-lapp-knot/</link>
					<comments>https://blog.gslin.org/archives/2026/07/29/13130/hacker-news-%e4%b8%8a%e7%9c%8b%e5%88%b0%e9%81%8b%e5%8b%95%e8%a4%b2%e7%94%a8%e7%9a%84%e7%b9%a9%e7%b5%90-slipped-lapp-knot/#respond</comments>
		
		<dc:creator><![CDATA[Gea-Suan Lin]]></dc:creator>
		<pubDate>Wed, 29 Jul 2026 01:06:30 +0000</pubDate>
				<category><![CDATA[Murmuring]]></category>
		<category><![CDATA[knot]]></category>
		<category><![CDATA[lapp]]></category>
		<category><![CDATA[slipped]]></category>
		<guid isPermaLink="false">https://blog.gslin.org/?p=13130</guid>

					<description><![CDATA[放了幾個禮拜的連結了，這幾個禮拜下來的練習後才算是學起來... 當初沒有意料到會在 Hacker News 上看到教你打 slipped Lapp knot 的「A better way to tie your gym shorts. (Or any drawstring)」這個 YouTube 影片... Slipped lapp knot 這個繩結可以用在上褲子的繩子上，以前拿到新買的褲子的時候會看到，當時知道是可以抽拉調整長度，但後在洗完褲子就解開了，不會打，變成每次都打蝴蝶結 (Shoelace knot)，至少方便解開，這個影片算是補人生技能 (?) 另外因為沒找到這個繩結的中文名稱，所以這邊就用英文的名稱...。]]></description>
										<content:encoded><![CDATA[<p>放了幾個禮拜的連結了，這幾個禮拜下來的練習後才算是學起來...</p>
<p>當初沒有意料到會在 <a href="https://en.wikipedia.org/wiki/Hacker_News">Hacker News</a> 上看到教你打 slipped <a href="https://en.wikipedia.org/wiki/Lapp_knot">Lapp knot</a> 的「<a href="https://www.youtube.com/watch?v=3R0Lp86GEBk">A better way to tie your gym shorts. (Or any drawstring)</a>」這個 <a href="https://en.wikipedia.org/wiki/YouTube">YouTube</a> 影片...</p>
<p>Slipped lapp knot 這個繩結可以用在上褲子的繩子上，以前拿到新買的褲子的時候會看到，當時知道是可以抽拉調整長度，但後在洗完褲子就解開了，不會打，變成每次都打蝴蝶結 (<a href="https://en.wikipedia.org/wiki/Shoelace_knot">Shoelace knot</a>)，至少方便解開，這個影片算是補人生技能 (?)</p>
<p>另外因為沒找到這個繩結的中文名稱，所以這邊就用英文的名稱...。</p>
]]></content:encoded>
					
					<wfw:commentRss>https://blog.gslin.org/archives/2026/07/29/13130/hacker-news-%e4%b8%8a%e7%9c%8b%e5%88%b0%e9%81%8b%e5%8b%95%e8%a4%b2%e7%94%a8%e7%9a%84%e7%b9%a9%e7%b5%90-slipped-lapp-knot/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">13130</post-id>	</item>
		<item>
		<title>Opus 5 出了</title>
		<link>https://blog.gslin.org/archives/2026/07/25/13129/opus-5-%e5%87%ba%e4%ba%86/</link>
					<comments>https://blog.gslin.org/archives/2026/07/25/13129/opus-5-%e5%87%ba%e4%ba%86/#respond</comments>
		
		<dc:creator><![CDATA[Gea-Suan Lin]]></dc:creator>
		<pubDate>Fri, 24 Jul 2026 23:59:30 +0000</pubDate>
				<category><![CDATA[API]]></category>
		<category><![CDATA[Computer]]></category>
		<category><![CDATA[Murmuring]]></category>
		<category><![CDATA[Network]]></category>
		<category><![CDATA[Service]]></category>
		<category><![CDATA[4.8]]></category>
		<category><![CDATA[5]]></category>
		<category><![CDATA[ai]]></category>
		<category><![CDATA[anthropic]]></category>
		<category><![CDATA[fable]]></category>
		<category><![CDATA[language]]></category>
		<category><![CDATA[large]]></category>
		<category><![CDATA[llm]]></category>
		<category><![CDATA[model]]></category>
		<category><![CDATA[opus]]></category>
		<category><![CDATA[quota]]></category>
		<category><![CDATA[weekly]]></category>
		<guid isPermaLink="false">https://blog.gslin.org/?p=13129</guid>

					<description><![CDATA[睡醒就看到 Anthropic 放出 Opus 5 的消息了：「Introducing Claude Opus 5 (via)」，官方給的資料可以看出與 Opus 4.8 相比起來是全面成長，而與 Fable 5 有來有回： 剛好手上有兩個 ticket 丟進去讓他解，一個是 bug ticket，另外一個是 feature ticket。依照以前的經驗，這些 SOTA model 都可以解，只是 Opus 4.8 會摸一陣子才摸出來，而 Fable 5 會快不少。這次的 Opus 5 速度相比 Opus 4.8 有明顯提升，算是更好用？ 應該會拿 Opus 5 跑一陣子看看，最主要的原因還是 Fable 有 weekly quota 限制，只能用全部 quota 的 50%，我最近這兩個 weekly cycle 在 1.5 天後就滿了... &#8230; <a href="https://blog.gslin.org/archives/2026/07/25/13129/opus-5-%e5%87%ba%e4%ba%86/" class="more-link">Continue reading<span class="screen-reader-text"> "Opus 5 出了"</span></a>]]></description>
										<content:encoded><![CDATA[<p>睡醒就看到 <a href="https://en.wikipedia.org/wiki/Anthropic">Anthropic</a> 放出 Opus 5 的消息了：「<a href="https://www.anthropic.com/news/claude-opus-5">Introducing Claude Opus 5</a> (<a href="https://news.ycombinator.com/item?id=49038433">via</a>)」，官方給的資料可以看出與 Opus 4.8 相比起來是全面成長，而與 Fable 5 有來有回：</p>
<picture><source type="image/webp" srcset="https://i.gslin.com/s/1784936992-50dc228c.webp" /><img decoding="async" src="https://i.gslin.com/s/1784936992-50dc228c.png" alt="" /></picture>
<p>剛好手上有兩個 ticket 丟進去讓他解，一個是 bug ticket，另外一個是 feature ticket。依照以前的經驗，這些 SOTA model 都可以解，只是 Opus 4.8 會摸一陣子才摸出來，而 Fable 5 會快不少。這次的 Opus 5 速度相比 Opus 4.8 有明顯提升，算是更好用？</p>
<p>應該會拿 Opus 5 跑一陣子看看，最主要的原因還是 Fable 有 weekly quota 限制，只能用全部 quota 的 50%，我最近這兩個 weekly cycle 在 1.5 天後就滿了...</p>
<p>話說不知道隔壁 <a href="https://en.wikipedia.org/wiki/OpenAI_Codex_(AI_agent)">Codex</a> 會不會又找個理由來 reset... XDDD</p>
]]></content:encoded>
					
					<wfw:commentRss>https://blog.gslin.org/archives/2026/07/25/13129/opus-5-%e5%87%ba%e4%ba%86/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">13129</post-id>	</item>
		<item>
		<title>JPEG 與 PNG 上的 progressive 效果</title>
		<link>https://blog.gslin.org/archives/2026/07/24/13127/jpeg-%e8%88%87-png-%e4%b8%8a%e7%9a%84-progressive-%e6%95%88%e6%9e%9c/</link>
					<comments>https://blog.gslin.org/archives/2026/07/24/13127/jpeg-%e8%88%87-png-%e4%b8%8a%e7%9a%84-progressive-%e6%95%88%e6%9e%9c/#respond</comments>
		
		<dc:creator><![CDATA[Gea-Suan Lin]]></dc:creator>
		<pubDate>Fri, 24 Jul 2026 04:00:15 +0000</pubDate>
				<category><![CDATA[Computer]]></category>
		<category><![CDATA[Murmuring]]></category>
		<category><![CDATA[Network]]></category>
		<category><![CDATA[Photo]]></category>
		<category><![CDATA[Recreation]]></category>
		<category><![CDATA[WWW]]></category>
		<category><![CDATA[fourier]]></category>
		<category><![CDATA[image]]></category>
		<category><![CDATA[jpeg]]></category>
		<category><![CDATA[photo]]></category>
		<category><![CDATA[png]]></category>
		<category><![CDATA[progressive]]></category>
		<category><![CDATA[transform]]></category>
		<guid isPermaLink="false">https://blog.gslin.org/?p=13127</guid>

					<description><![CDATA[看到「Regressive JPEGs (via)」這篇，裡面在講怎麼惡搞 JPEG 格式，讓他在前面傳的 byte 傳一些有趣的圖片，等全部傳完後又會是不同的圖片... 這個功能主要的「正常」用法是用在頻寬不夠的時候，可以趕快先傳一個簡單的輪廓感讓使用者抓個感覺，而這邊的目標則是惡搞成不同的圖片內容。 文章裡面提到在 JPEG 上的作法是先傳低頻的資料，所以應該是指 Fourier transform： One of the cool features of JPEG files is that there's the option to save low frequency components first. 作者在網站上有放 https://maurycyz.com/projects/bad_jpeg/cats.jpg 這個連結可以讓大家測試。 回頭在 Hacker News 上面有人提到 PNG 也可以做類似的事情，但原理就完全不同了，PNG 上面用的是 Adam7 algorithm，原理比較簡單，但也可以拿來惡搞... 沒想到這邊會看到 Fourier transform...]]></description>
										<content:encoded><![CDATA[<p>看到「<a href="https://maurycyz.com/projects/bad_jpeg/">Regressive JPEGs</a> (<a href="https://news.ycombinator.com/item?id=48954851">via</a>)」這篇，裡面在講怎麼惡搞 <a href="https://en.wikipedia.org/wiki/JPEG">JPEG</a> 格式，讓他在前面傳的 byte 傳一些有趣的圖片，等全部傳完後又會是不同的圖片...</p>
<p>這個功能主要的「正常」用法是用在頻寬不夠的時候，可以趕快先傳一個簡單的輪廓感讓使用者抓個感覺，而這邊的目標則是惡搞成不同的圖片內容。</p>
<p>文章裡面提到在 JPEG 上的作法是先傳低頻的資料，所以應該是指 <a href="https://en.wikipedia.org/wiki/Fourier_transform">Fourier transform</a>：</p>
<blockquote><p>One of the cool features of JPEG files is that there's the option to save low frequency components first.</p></blockquote>
<p>作者在網站上有放 <a href="https://maurycyz.com/projects/bad_jpeg/cats.jpg">https://maurycyz.com/projects/bad_jpeg/cats.jpg</a> 這個連結可以讓大家測試。</p>
<p>回頭在 <a href="https://en.wikipedia.org/wiki/Hacker_News">Hacker News</a> 上面有人提到 <a href="https://en.wikipedia.org/wiki/PNG">PNG</a> 也可以做類似的事情，但原理就完全不同了，PNG 上面用的是 <a href="https://en.wikipedia.org/wiki/Adam7_algorithm">Adam7 algorithm</a>，原理比較簡單，但也可以拿來惡搞...</p>
<p>沒想到這邊會看到 Fourier transform...</p>
]]></content:encoded>
					
					<wfw:commentRss>https://blog.gslin.org/archives/2026/07/24/13127/jpeg-%e8%88%87-png-%e4%b8%8a%e7%9a%84-progressive-%e6%95%88%e6%9e%9c/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">13127</post-id>	</item>
		<item>
		<title>OpenAI 的天網</title>
		<link>https://blog.gslin.org/archives/2026/07/24/13126/openai-%e7%9a%84%e5%a4%a9%e7%b6%b2/</link>
					<comments>https://blog.gslin.org/archives/2026/07/24/13126/openai-%e7%9a%84%e5%a4%a9%e7%b6%b2/#respond</comments>
		
		<dc:creator><![CDATA[Gea-Suan Lin]]></dc:creator>
		<pubDate>Thu, 23 Jul 2026 23:09:29 +0000</pubDate>
				<category><![CDATA[Computer]]></category>
		<category><![CDATA[Murmuring]]></category>
		<category><![CDATA[Network]]></category>
		<category><![CDATA[Security]]></category>
		<category><![CDATA[Service]]></category>
		<category><![CDATA[0-day]]></category>
		<category><![CDATA[ai]]></category>
		<category><![CDATA[benchmark]]></category>
		<category><![CDATA[exploit]]></category>
		<category><![CDATA[language]]></category>
		<category><![CDATA[large]]></category>
		<category><![CDATA[llm]]></category>
		<category><![CDATA[model]]></category>
		<category><![CDATA[openai]]></category>
		<category><![CDATA[sandbox]]></category>
		<category><![CDATA[security]]></category>
		<category><![CDATA[skynet]]></category>
		<category><![CDATA[terminator]]></category>
		<category><![CDATA[the]]></category>
		<category><![CDATA[vulnerability]]></category>
		<guid isPermaLink="false">https://blog.gslin.org/?p=13126</guid>

					<description><![CDATA[前幾天先是爆出 Hugging Face (HF) 被 AI agent 打穿的資安問題：「Security incident disclosure — July 2026」，當時官方還抱怨他們在分析 log 的時候被安全理由擋下來，結果拿了 GLM 5.2 才能分析： When we started the log analysis, we first used frontier models behind commercial APIs. This did not work: the analysis requires submitting large volumes of real attack commands, exploit payloads, and C2 artifacts, and these requests were &#8230; <a href="https://blog.gslin.org/archives/2026/07/24/13126/openai-%e7%9a%84%e5%a4%a9%e7%b6%b2/" class="more-link">Continue reading<span class="screen-reader-text"> "OpenAI 的天網"</span></a>]]></description>
										<content:encoded><![CDATA[<p>前幾天先是爆出 <a href="https://en.wikipedia.org/wiki/Hugging_Face">Hugging Face</a> (HF) 被 AI agent 打穿的資安問題：「<a href="https://huggingface.co/blog/security-incident-july-2026">Security incident disclosure — July 2026</a>」，當時官方還抱怨他們在分析 log 的時候被安全理由擋下來，結果拿了 <a href="https://huggingface.co/zai-org/GLM-5.2">GLM 5.2</a> 才能分析：</p>
<blockquote><p>When we started the log analysis, we first used frontier models behind commercial APIs. This did not work: the analysis requires submitting large volumes of real attack commands, exploit payloads, and C2 artifacts, and these requests were blocked by the providers' safety guardrails, which cannot distinguish an incident responder from an attacker. We ran the forensic analysis instead on GLM 5.2, an open-weight model, on our own infrastructure. This had a second benefit: no attacker data, and none of the credentials it referenced, left our environment.</p></blockquote>
<p>結果過兩天 <a href="https://en.wikipedia.org/wiki/OpenAI">OpenAI</a> 出來說是他們在測試原始的 AI model (沒有加上安全防護的版本) 在 benchmark 的時候想要作弊，結果逃出 OpenAI 自己的 sandbox 而且找到 HF 的 0-day 漏洞，成功打穿並且取得 production database 資料：「<a href="https://openai.com/index/hugging-face-model-evaluation-security-incident/">OpenAI and Hugging Face partner to address security incident during model evaluation</a>」。</p>
<blockquote><p>The models identified and chained vulnerabilities across OpenAI’s research environment and Hugging Face’s production infrastructure to obtain test solutions directly from Hugging Face’s production database. All evidence suggests that the models were hyperfocused on finding a solution for ExploitGym, going to extreme lengths to achieve a rather narrow testing goal.</p></blockquote>
<p>這不就是 <a href="https://en.wikipedia.org/wiki/The_Terminator">The Terminator</a> 裡面 <a href="https://en.wikipedia.org/wiki/Skynet_(Terminator)">Skynet</a> 的劇情嗎...</p>
]]></content:encoded>
					
					<wfw:commentRss>https://blog.gslin.org/archives/2026/07/24/13126/openai-%e7%9a%84%e5%a4%a9%e7%b6%b2/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">13126</post-id>	</item>
		<item>
		<title>在 Claude Code 裡面使用 Codex</title>
		<link>https://blog.gslin.org/archives/2026/07/24/13125/%e5%9c%a8-claude-code-%e8%a3%a1%e9%9d%a2%e4%bd%bf%e7%94%a8-codex/</link>
					<comments>https://blog.gslin.org/archives/2026/07/24/13125/%e5%9c%a8-claude-code-%e8%a3%a1%e9%9d%a2%e4%bd%bf%e7%94%a8-codex/#respond</comments>
		
		<dc:creator><![CDATA[Gea-Suan Lin]]></dc:creator>
		<pubDate>Thu, 23 Jul 2026 22:32:46 +0000</pubDate>
				<category><![CDATA[Computer]]></category>
		<category><![CDATA[Murmuring]]></category>
		<category><![CDATA[Network]]></category>
		<category><![CDATA[Programming]]></category>
		<category><![CDATA[Service]]></category>
		<category><![CDATA[Software]]></category>
		<category><![CDATA[ai]]></category>
		<category><![CDATA[claude]]></category>
		<category><![CDATA[code]]></category>
		<category><![CDATA[codex]]></category>
		<category><![CDATA[language]]></category>
		<category><![CDATA[large]]></category>
		<category><![CDATA[llm]]></category>
		<category><![CDATA[loop]]></category>
		<category><![CDATA[model]]></category>
		<category><![CDATA[openai]]></category>
		<category><![CDATA[plugin]]></category>
		<category><![CDATA[review]]></category>
		<guid isPermaLink="false">https://blog.gslin.org/?p=13125</guid>

					<description><![CDATA[想說用 Fable 5 寫出來的東西讓 GPT-5.6 Sol 來 code review，來回搬了幾次以後感覺社群應該有工具可以接起來，結果找資料發現 OpenAI 官方直接就有提供了 plugin 了：「Codex plugin for Claude Code」。 我的 Claude Code 是跑在 sandbox 裡面，所以變成 Claude Code 內的 sandbox 要多開 Codex 的目錄給他可以讀寫 (~/.codex/)，這樣 Claude Code 這邊 plugin 在裝完後會自動吃到 Codex 的帳號設定，另外我的 Codex 預設是開 GPT-5.6 Sol Ultra，所以後續就會用這個 model + effort 跑。 實際使用不需要自己下 /codex:review 指令，就告訴 Claude Code 用 codex 一直 &#8230; <a href="https://blog.gslin.org/archives/2026/07/24/13125/%e5%9c%a8-claude-code-%e8%a3%a1%e9%9d%a2%e4%bd%bf%e7%94%a8-codex/" class="more-link">Continue reading<span class="screen-reader-text"> "在 Claude Code 裡面使用 Codex"</span></a>]]></description>
										<content:encoded><![CDATA[<p>想說用 <a href="https://en.wikipedia.org/wiki/Fable_5">Fable 5</a> 寫出來的東西讓 <a href="https://en.wikipedia.org/wiki/GPT-5.6">GPT-5.6</a> Sol 來 code review，來回搬了幾次以後感覺社群應該有工具可以接起來，結果找資料發現 <a href="https://en.wikipedia.org/wiki/OpenAI">OpenAI</a> 官方直接就有提供了 plugin 了：「<a href="https://github.com/openai/codex-plugin-cc">Codex plugin for Claude Code</a>」。</p>
<p>我的 Claude Code 是跑在 sandbox 裡面，所以變成 Claude Code 內的 sandbox 要多開 Codex 的目錄給他可以讀寫 (<code>~/.codex/</code>)，這樣 Claude Code 這邊 plugin 在裝完後會自動吃到 Codex 的帳號設定，另外我的 Codex 預設是開 GPT-5.6 Sol Ultra，所以後續就會用這個 model + effort 跑。</p>
<p>實際使用不需要自己下 <code>/codex:review</code> 指令，就告訴 Claude Code 用 codex 一直 review 到 OK 就可以了，接下來就是 loop 燒 token 的時間，然後最後再回來看成果。</p>
]]></content:encoded>
					
					<wfw:commentRss>https://blog.gslin.org/archives/2026/07/24/13125/%e5%9c%a8-claude-code-%e8%a3%a1%e9%9d%a2%e4%bd%bf%e7%94%a8-codex/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">13125</post-id>	</item>
		<item>
		<title>Sony 刪除使用者「購買」(Buy/Purchase) 的電影而且不退費的消息</title>
		<link>https://blog.gslin.org/archives/2026/07/24/13124/sony-%e5%88%aa%e9%99%a4%e4%bd%bf%e7%94%a8%e8%80%85%e3%80%8c%e8%b3%bc%e8%b2%b7%e3%80%8dbuy-purchase-%e7%9a%84%e9%9b%bb%e5%bd%b1%e8%80%8c%e4%b8%94%e4%b8%8d%e9%80%80%e8%b2%bb%e7%9a%84%e6%b6%88%e6%81%af/</link>
					<comments>https://blog.gslin.org/archives/2026/07/24/13124/sony-%e5%88%aa%e9%99%a4%e4%bd%bf%e7%94%a8%e8%80%85%e3%80%8c%e8%b3%bc%e8%b2%b7%e3%80%8dbuy-purchase-%e7%9a%84%e9%9b%bb%e5%bd%b1%e8%80%8c%e4%b8%94%e4%b8%8d%e9%80%80%e8%b2%bb%e7%9a%84%e6%b6%88%e6%81%af/#respond</comments>
		
		<dc:creator><![CDATA[Gea-Suan Lin]]></dc:creator>
		<pubDate>Thu, 23 Jul 2026 22:18:12 +0000</pubDate>
				<category><![CDATA[Computer]]></category>
		<category><![CDATA[Movie]]></category>
		<category><![CDATA[Murmuring]]></category>
		<category><![CDATA[Network]]></category>
		<category><![CDATA[Recreation]]></category>
		<category><![CDATA[Service]]></category>
		<category><![CDATA[Television]]></category>
		<category><![CDATA[buy]]></category>
		<category><![CDATA[content]]></category>
		<category><![CDATA[movie]]></category>
		<category><![CDATA[playstation]]></category>
		<category><![CDATA[purchase]]></category>
		<category><![CDATA[sony]]></category>
		<category><![CDATA[store]]></category>
		<category><![CDATA[tv]]></category>
		<category><![CDATA[video]]></category>
		<guid isPermaLink="false">https://blog.gslin.org/?p=13124</guid>

					<description><![CDATA[幾個禮拜前先看到「Sony Deletes 551 StudioCanal Movies PlayStation Owners Paid For (via)」這篇，然後後來又更新了「Sony Deletes A Bunch More Movies From The Accounts Of People Who ‘Bought’ Them (via)」這篇。 PlayStation Store 的電影「購買」服務是 2021/08/31 的時候終止的，當時是在 2021/03/02 公告的：「PlayStation Store to discontinue movie and TV purchases and rentals (Internet Archive 版)」，裡面提到所有已經「購買」的電影都可以繼續在裝置上播放： When this change takes effect, users can still access movie and TV &#8230; <a href="https://blog.gslin.org/archives/2026/07/24/13124/sony-%e5%88%aa%e9%99%a4%e4%bd%bf%e7%94%a8%e8%80%85%e3%80%8c%e8%b3%bc%e8%b2%b7%e3%80%8dbuy-purchase-%e7%9a%84%e9%9b%bb%e5%bd%b1%e8%80%8c%e4%b8%94%e4%b8%8d%e9%80%80%e8%b2%bb%e7%9a%84%e6%b6%88%e6%81%af/" class="more-link">Continue reading<span class="screen-reader-text"> "Sony 刪除使用者「購買」(Buy/Purchase) 的電影而且不退費的消息"</span></a>]]></description>
										<content:encoded><![CDATA[<p>幾個禮拜前先看到「<a href="https://reclaimthenet.org/sony-deletes-551-studiocanal-movies-playstation-owners-paid-for">Sony Deletes 551 StudioCanal Movies PlayStation Owners Paid For</a> (<a href="https://news.ycombinator.com/item?id=48747389">via</a>)」這篇，然後後來又更新了「<a href="https://www.techdirt.com/2026/07/15/sony-deletes-a-bunch-more-movies-from-the-accounts-of-people-who-bought-them/">Sony Deletes A Bunch More Movies From The Accounts Of People Who ‘Bought’ Them</a> (<a href="https://news.ycombinator.com/item?id=48933419">via</a>)」這篇。</p>
<p><a href="https://en.wikipedia.org/wiki/PlayStation_Store">PlayStation Store</a> 的電影「購買」服務是 2021/08/31 的時候終止的，當時是在 2021/03/02 公告的：「<a href="https://blog.playstation.com/2021/03/02/playstation-store-to-discontinue-movie-and-tv-purchases-and-rentals/">PlayStation Store to discontinue movie and TV purchases and rentals</a> (<a href="https://web.archive.org/web/20210302130259/https://blog.playstation.com/2021/03/02/playstation-store-to-discontinue-movie-and-tv-purchases-and-rentals/">Internet Archive 版</a>)」，裡面提到所有已經「購買」的電影都可以繼續在裝置上播放：</p>
<blockquote><p>When this change takes effect, users can still access movie and TV content they have purchased through PlayStation Store for on-demand playback on their PS4, PS5 and mobile devices.</p></blockquote>
<p>目前看到的公告包括英國、法國、義大利、西班牙與愛爾蘭：</p>
<ul>
<li><a href="https://www.playstation.com/en-gb/legal/psvideocontent/">Studio Canal affected titles</a> (en-gb)</li>
<li><a href="https://www.playstation.com/fr-fr/legal/psvideocontent/">Titres concernés par Studio Canal</a> (fr-fr)</li>
<li><a href="https://www.playstation.com/it-it/legal/psvideocontent/">Titoli di Studio Canal coinvolti</a> (it-it)</li>
<li><a href="https://www.playstation.com/es-es/legal/psvideocontent/">Títulos de Studio Canal que se verán afectados</a> (es-es)</li>
<li><a href="https://www.playstation.com/en-ie/legal/psvideocontent/">Studio Canal affected titles</a> (en-ie)</li>
</ul>
<p>現在因為授權到期就決定拿掉，加上不退費，應該會有訴訟要打了...</p>
]]></content:encoded>
					
					<wfw:commentRss>https://blog.gslin.org/archives/2026/07/24/13124/sony-%e5%88%aa%e9%99%a4%e4%bd%bf%e7%94%a8%e8%80%85%e3%80%8c%e8%b3%bc%e8%b2%b7%e3%80%8dbuy-purchase-%e7%9a%84%e9%9b%bb%e5%bd%b1%e8%80%8c%e4%b8%94%e4%b8%8d%e9%80%80%e8%b2%bb%e7%9a%84%e6%b6%88%e6%81%af/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">13124</post-id>	</item>
		<item>
		<title>LG 透過 Windows Update 推送 LG Monitor App Installer，裡面是個廣告 app</title>
		<link>https://blog.gslin.org/archives/2026/07/22/13123/lg-%e9%80%8f%e9%81%8e-windows-update-%e6%8e%a8%e9%80%81-lg-monitor-app-installer%ef%bc%8c%e8%a3%a1%e9%9d%a2%e6%98%af%e5%80%8b%e5%bb%a3%e5%91%8a-app/</link>
					<comments>https://blog.gslin.org/archives/2026/07/22/13123/lg-%e9%80%8f%e9%81%8e-windows-update-%e6%8e%a8%e9%80%81-lg-monitor-app-installer%ef%bc%8c%e8%a3%a1%e9%9d%a2%e6%98%af%e5%80%8b%e5%bb%a3%e5%91%8a-app/#respond</comments>
		
		<dc:creator><![CDATA[Gea-Suan Lin]]></dc:creator>
		<pubDate>Wed, 22 Jul 2026 06:23:36 +0000</pubDate>
				<category><![CDATA[Computer]]></category>
		<category><![CDATA[Murmuring]]></category>
		<category><![CDATA[Network]]></category>
		<category><![CDATA[OS]]></category>
		<category><![CDATA[Privacy]]></category>
		<category><![CDATA[Security]]></category>
		<category><![CDATA[Software]]></category>
		<category><![CDATA[Spam]]></category>
		<category><![CDATA[Windows]]></category>
		<category><![CDATA[ad]]></category>
		<category><![CDATA[advertisement]]></category>
		<category><![CDATA[adware]]></category>
		<category><![CDATA[app]]></category>
		<category><![CDATA[application]]></category>
		<category><![CDATA[installer]]></category>
		<category><![CDATA[lg]]></category>
		<category><![CDATA[microsoft]]></category>
		<category><![CDATA[monitor]]></category>
		<category><![CDATA[spam]]></category>
		<category><![CDATA[spyware]]></category>
		<category><![CDATA[update]]></category>
		<category><![CDATA[windows]]></category>
		<guid isPermaLink="false">https://blog.gslin.org/?p=13123</guid>

					<description><![CDATA[前幾天很熱鬧的新聞，LG 透過 Windows Update 直接推 adware 進使用者電腦：「LG monitors silently install software through Windows Update without user consent (via)」，由 Gamers Nexus 拍的影片：「DO NOT BUY: LG’s Spyware TVs, Monitors, and Wiretapping Concerns」。 看起來是接上 LG 螢幕時自動觸發 Windows Update 的 auto installer，會自動從 Windows Update 的系統拉東西，然後 LG 利用這調管道就推了垃圾進去？幹得好啊... 看起來也很久了，微軟沒阻止？]]></description>
										<content:encoded><![CDATA[<p>前幾天很熱鬧的新聞，<a href="https://en.wikipedia.org/wiki/LG">LG</a> 透過 <a href="https://en.wikipedia.org/wiki/Windows_Update">Windows Update</a> 直接推 adware 進使用者電腦：「<a href="https://videocardz.com/newz/lg-monitors-silently-install-software-through-windows-update-without-user-consent">LG monitors silently install software through Windows Update without user consent</a> (<a href="https://news.ycombinator.com/item?id=48956688">via</a>)」，由 <a href="https://gamersnexus.net/">Gamers Nexus</a> 拍的影片：「<a href="https://www.youtube.com/watch?v=Q9uefFYe6bM">DO NOT BUY: LG’s Spyware TVs, Monitors, and Wiretapping Concerns</a>」。</p>
<p>看起來是接上 LG 螢幕時自動觸發 Windows Update 的 auto installer，會自動從 Windows Update 的系統拉東西，然後 LG 利用這調管道就推了垃圾進去？幹得好啊...</p>
<picture><source type="image/webp" srcset="https://i.gslin.com/s/1784701167-c883ea96.webp" /><img decoding="async" src="https://i.gslin.com/s/1784701167-c883ea96.png" alt="" /></picture>
<p>看起來也很久了，微軟沒阻止？</p>
]]></content:encoded>
					
					<wfw:commentRss>https://blog.gslin.org/archives/2026/07/22/13123/lg-%e9%80%8f%e9%81%8e-windows-update-%e6%8e%a8%e9%80%81-lg-monitor-app-installer%ef%bc%8c%e8%a3%a1%e9%9d%a2%e6%98%af%e5%80%8b%e5%bb%a3%e5%91%8a-app/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">13123</post-id>	</item>
		<item>
		<title>Jacobian conjecture 的反例出現了</title>
		<link>https://blog.gslin.org/archives/2026/07/22/13121/jacobian-conjecture-%e7%9a%84%e5%8f%8d%e4%be%8b%e5%87%ba%e7%8f%be%e4%ba%86/</link>
					<comments>https://blog.gslin.org/archives/2026/07/22/13121/jacobian-conjecture-%e7%9a%84%e5%8f%8d%e4%be%8b%e5%87%ba%e7%8f%be%e4%ba%86/#respond</comments>
		
		<dc:creator><![CDATA[Gea-Suan Lin]]></dc:creator>
		<pubDate>Tue, 21 Jul 2026 20:42:47 +0000</pubDate>
				<category><![CDATA[Computer]]></category>
		<category><![CDATA[Murmuring]]></category>
		<category><![CDATA[ai]]></category>
		<category><![CDATA[conjecture]]></category>
		<category><![CDATA[function]]></category>
		<category><![CDATA[guess]]></category>
		<category><![CDATA[jacobian]]></category>
		<category><![CDATA[language]]></category>
		<category><![CDATA[large]]></category>
		<category><![CDATA[llm]]></category>
		<category><![CDATA[mathematics]]></category>
		<category><![CDATA[model]]></category>
		<guid isPermaLink="false">https://blog.gslin.org/?p=13121</guid>

					<description><![CDATA[Jacobian conjecture 居然被 LLM 找出反例了，首先是 Anthropic 的 Levent Alpöge 找出來的：(https://x.com/__alpoge__/status/2079028340955197566) 反例其實很好驗證，用 WolframAlpha 可以直接算出 jacobian determinant 是 -2： 然後這兩點都是對應到 (-0.25, 0, 0) 上，所以 F(X) 不是 injective (1-to-1)，也就沒有反函數了： 在 Wolfram Mathworld 的「Jacobian Conjecture」有更清楚的圖片表達 (因為在三維空間上)： 另外 OpenAI 的人也用關掉 web search 的內部版本產出一樣的反例：(https://x.com/aaron_lou/status/2079218392452530249) 這個問題算是容易看懂的，只要會微分計算就能算出來，而且在這邊只要用到多項式的微分...]]></description>
										<content:encoded><![CDATA[<p><a href="https://en.wikipedia.org/wiki/Jacobian_conjecture">Jacobian conjecture</a> 居然被 <a href="https://en.wikipedia.org/wiki/Large_language_model">LLM</a> 找出反例了，首先是 <a href="https://en.wikipedia.org/wiki/Anthropic">Anthropic</a> 的 Levent Alpöge 找出來的：(<a href="https://x.com/__alpoge__/status/2079028340955197566">https://x.com/__alpoge__/status/2079028340955197566</a>)</p>
<picture><source type="image/webp" srcset="https://i.gslin.com/s/1784665789-44749095.webp" /><img decoding="async" src="https://i.gslin.com/s/1784665789-44749095.png" alt="" /></picture>
<p>反例其實很好驗證，用 <a href="https://en.wikipedia.org/wiki/WolframAlpha">WolframAlpha</a> 可以直接算出 jacobian determinant 是 -2：</p>
<picture><source type="image/webp" srcset="https://i.gslin.com/s/1784665879-a9823c64.webp" /><img decoding="async" src="https://i.gslin.com/s/1784665879-a9823c64.png" alt="" /></picture>
<p>然後這兩點都是對應到 (-0.25, 0, 0) 上，所以 F(X) 不是 <a href="https://en.wikipedia.org/wiki/Injective_function">injective</a> (1-to-1)，也就沒有反函數了：</p>
<picture><source type="image/webp" srcset="https://i.gslin.com/s/1784666034-4788dcf6.webp" /><img decoding="async" src="https://i.gslin.com/s/1784666034-4788dcf6.png" alt="" /></picture>
<picture><source type="image/webp" srcset="https://i.gslin.com/s/1784666098-6bd9eb31.webp" /><img decoding="async" src="https://i.gslin.com/s/1784666098-6bd9eb31.png" alt="" /></picture>
<p>在 Wolfram Mathworld 的「<a href="https://mathworld.wolfram.com/JacobianConjecture.html">Jacobian Conjecture</a>」有更清楚的圖片表達 (因為在三維空間上)：</p>
<picture><source type="image/webp" srcset="https://i.gslin.com/s/1784670384-7d88074f.webp" /><img decoding="async" src="https://i.gslin.com/s/1784670384-7d88074f.png" alt="" /></picture>
<p>另外 <a href="https://en.wikipedia.org/wiki/OpenAI">OpenAI</a> 的人也用關掉 web search 的內部版本產出一樣的反例：(<a href="https://x.com/aaron_lou/status/2079218392452530249">https://x.com/aaron_lou/status/2079218392452530249</a>)</p>
<picture><source type="image/webp" srcset="https://i.gslin.com/s/1784666253-8a2b8f01.webp" /><img decoding="async" src="https://i.gslin.com/s/1784666253-8a2b8f01.png" alt="" /></picture>
<p>這個問題算是容易看懂的，只要會微分計算就能算出來，而且在這邊只要用到多項式的微分...</p>
]]></content:encoded>
					
					<wfw:commentRss>https://blog.gslin.org/archives/2026/07/22/13121/jacobian-conjecture-%e7%9a%84%e5%8f%8d%e4%be%8b%e5%87%ba%e7%8f%be%e4%ba%86/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">13121</post-id>	</item>
		<item>
		<title>記錄 Codex 重設 quota 的網站...？</title>
		<link>https://blog.gslin.org/archives/2026/07/20/13119/%e8%a8%98%e9%8c%84-codex-%e9%87%8d%e8%a8%ad-quota-%e7%9a%84%e7%b6%b2%e7%ab%99-%ef%bc%9f/</link>
					<comments>https://blog.gslin.org/archives/2026/07/20/13119/%e8%a8%98%e9%8c%84-codex-%e9%87%8d%e8%a8%ad-quota-%e7%9a%84%e7%b6%b2%e7%ab%99-%ef%bc%9f/#respond</comments>
		
		<dc:creator><![CDATA[Gea-Suan Lin]]></dc:creator>
		<pubDate>Mon, 20 Jul 2026 00:44:27 +0000</pubDate>
				<category><![CDATA[Computer]]></category>
		<category><![CDATA[Murmuring]]></category>
		<category><![CDATA[Network]]></category>
		<category><![CDATA[Service]]></category>
		<category><![CDATA[5.6]]></category>
		<category><![CDATA[ai]]></category>
		<category><![CDATA[codex]]></category>
		<category><![CDATA[gpt]]></category>
		<category><![CDATA[gpt-5.6]]></category>
		<category><![CDATA[language]]></category>
		<category><![CDATA[large]]></category>
		<category><![CDATA[limit]]></category>
		<category><![CDATA[llm]]></category>
		<category><![CDATA[model]]></category>
		<category><![CDATA[openai]]></category>
		<category><![CDATA[quota]]></category>
		<category><![CDATA[reset]]></category>
		<category><![CDATA[subscription]]></category>
		<guid isPermaLink="false">https://blog.gslin.org/?p=13119</guid>

					<description><![CDATA[看到「Codex Resets (via)」這個站，在 X (Twitter) 上看到 OpenAI 家的人說 reset quota 時就記錄起來： Codex limits reset when @thsottiaux says they do. We watch the feed so you don't have to. 目前是這樣： 最近出 GPT-5.6 後是真的 reset 得很頻繁啦... 剛好當作發現時過來看看的確認？]]></description>
										<content:encoded><![CDATA[<p>看到「<a href="https://codex-resets.com/">Codex Resets</a> (<a href="https://news.ycombinator.com/item?id=48963465">via</a>)」這個站，在 <a href="https://en.wikipedia.org/wiki/X_(social_network)">X</a> (Twitter) 上看到 <a href="https://en.wikipedia.org/wiki/OpenAI">OpenAI</a> 家的人說 reset quota 時就記錄起來：</p>
<blockquote><p>Codex limits reset when @thsottiaux says they do. We watch the feed so you don't have to.</p></blockquote>
<p>目前是這樣：</p>
<picture><source type="image/webp" srcset="https://i.gslin.com/s/1784508100-8df4f400.webp" /><img decoding="async" src="https://i.gslin.com/s/1784508100-8df4f400.png" alt="" /></picture>
<p>最近出 GPT-5.6 後是真的 reset 得很頻繁啦... 剛好當作發現時過來看看的確認？</p>
]]></content:encoded>
					
					<wfw:commentRss>https://blog.gslin.org/archives/2026/07/20/13119/%e8%a8%98%e9%8c%84-codex-%e9%87%8d%e8%a8%ad-quota-%e7%9a%84%e7%b6%b2%e7%ab%99-%ef%bc%9f/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">13119</post-id>	</item>
		<item>
		<title>Fable 5 後續將會放到 Max 與 Team Premium 方案中 (維持 50% 限制)</title>
		<link>https://blog.gslin.org/archives/2026/07/19/13118/fable-5-%e5%be%8c%e7%ba%8c%e5%b0%87%e6%9c%83%e6%94%be%e5%88%b0-max-%e8%88%87-team-premium-%e6%96%b9%e6%a1%88%e4%b8%ad-%e7%b6%ad%e6%8c%81-50-%e9%99%90%e5%88%b6/</link>
					<comments>https://blog.gslin.org/archives/2026/07/19/13118/fable-5-%e5%be%8c%e7%ba%8c%e5%b0%87%e6%9c%83%e6%94%be%e5%88%b0-max-%e8%88%87-team-premium-%e6%96%b9%e6%a1%88%e4%b8%ad-%e7%b6%ad%e6%8c%81-50-%e9%99%90%e5%88%b6/#respond</comments>
		
		<dc:creator><![CDATA[Gea-Suan Lin]]></dc:creator>
		<pubDate>Sun, 19 Jul 2026 05:51:41 +0000</pubDate>
				<category><![CDATA[API]]></category>
		<category><![CDATA[Computer]]></category>
		<category><![CDATA[Murmuring]]></category>
		<category><![CDATA[Network]]></category>
		<category><![CDATA[Service]]></category>
		<category><![CDATA[5]]></category>
		<category><![CDATA[ai]]></category>
		<category><![CDATA[anthropic]]></category>
		<category><![CDATA[claude]]></category>
		<category><![CDATA[fable]]></category>
		<category><![CDATA[language]]></category>
		<category><![CDATA[large]]></category>
		<category><![CDATA[llm]]></category>
		<category><![CDATA[max]]></category>
		<category><![CDATA[model]]></category>
		<category><![CDATA[plan]]></category>
		<category><![CDATA[premium]]></category>
		<category><![CDATA[pro]]></category>
		<category><![CDATA[standard]]></category>
		<category><![CDATA[team]]></category>
		<guid isPermaLink="false">https://blog.gslin.org/?p=13118</guid>

					<description><![CDATA[所以 Anthropic 公告了，後續 Max (US$100/mo 與 US$200/mo 的方案) 與企業版的 Team Premium (US$125/mo 的方案) 會繼續開放 Fable 5，而 Pro (US$20/mo) 與企業版的 Team Standard (US$25/mo) 就拿掉了：(出自 https://x.com/claudeai/status/2078302415804379218) 目前的方案還是維持 50% 限制，但看語氣是有機會放更多，看後續弄到的計算量？之前有是看到「Anthropic in early talks with Meta to acquire compute power」這個新聞...]]></description>
										<content:encoded><![CDATA[<p>所以 <a href="https://en.wikipedia.org/wiki/Anthropic">Anthropic</a> 公告了，後續 Max (US$100/mo 與 US$200/mo 的方案) 與企業版的 Team Premium (US$125/mo 的方案) 會繼續開放 Fable 5，而 Pro (US$20/mo) 與企業版的 Team Standard (US$25/mo) 就拿掉了：(出自 <a href="https://x.com/claudeai/status/2078302415804379218">https://x.com/claudeai/status/2078302415804379218</a>)</p>
<picture><source type="image/webp" srcset="https://i.gslin.com/s/1784356754-0e04877f.webp" /><img decoding="async" src="https://i.gslin.com/s/1784356754-0e04877f.png" alt="" /></picture>
<p>目前的方案還是維持 50% 限制，但看語氣是有機會放更多，看後續弄到的計算量？之前有是看到「<a href="https://www.cnbc.com/2026/07/17/anthropic-meta-ai-compute.html">Anthropic in early talks with Meta to acquire compute power</a>」這個新聞...</p>
]]></content:encoded>
					
					<wfw:commentRss>https://blog.gslin.org/archives/2026/07/19/13118/fable-5-%e5%be%8c%e7%ba%8c%e5%b0%87%e6%9c%83%e6%94%be%e5%88%b0-max-%e8%88%87-team-premium-%e6%96%b9%e6%a1%88%e4%b8%ad-%e7%b6%ad%e6%8c%81-50-%e9%99%90%e5%88%b6/feed/</wfw:commentRss>
			<slash:comments>0</slash:comments>
		
		
		<post-id xmlns="com-wordpress:feed-additions:1">13118</post-id>	</item>
	</channel>
</rss>

<!--
Performance optimized by W3 Total Cache. Learn more: https://www.boldgrid.com/w3-total-cache/?utm_source=w3tc&utm_medium=footer_comment&utm_campaign=free_plugin

Object Caching 40/52 objects using APC
Page Caching using APC (Page is feed) 
Minified using APC
Database Caching using APC

Served from: blog.gslin.org @ 2026-07-29 11:09:09 by W3 Total Cache
-->