Dataconomy https://dataconomy.com Bridging the gap between technology and business Wed, 24 Dec 2025 09:54:29 +0000 en-US hourly 1 https://dataconomy.com/wp-content/uploads/2025/01/DC_icon-75x75.png Dataconomy https://dataconomy.com 32 32 GPT-5.2 surpasses expert PhD baseline with 92% science score https://dataconomy.com/2025/12/24/gpt-5-2-surpasses-expert-phd-baseline-with-92-science-score/ Wed, 24 Dec 2025 09:54:29 +0000 https://dataconomy.com/?p=85803 GPT-5.2 scored 92% on a “Google-Proof” science benchmark, significantly surpassing the 70% expert baseline. The advanced model also achieved medal-winning performance in major international competitions, demonstrating its evolving capabilities in scientific reasoning. Scientists extensively use these systems for tasks like literature searches across various disciplines and languages, as well as navigating complex mathematical proofs. This […]]]>

GPT-5.2 scored 92% on a “Google-Proof” science benchmark, significantly surpassing the 70% expert baseline. The advanced model also achieved medal-winning performance in major international competitions, demonstrating its evolving capabilities in scientific reasoning.

Scientists extensively use these systems for tasks like literature searches across various disciplines and languages, as well as navigating complex mathematical proofs. This development often reduces work that typically takes days or weeks to just a few hours. The paper, Early science acceleration experiments with GPT-5, published in November 2025, provides initial evidence that GPT-5 can notably expedite scientific workflows.

To further measure and forecast AI models’ ability to accelerate scientific research, developers introduced FrontierScience, a new benchmark designed to assess expert-level scientific capabilities. The benchmark contains questions written and verified by experts in physics, chemistry, and biology, focusing on originality and difficulty.

FrontierScience features two distinct tracks:

  • Olympiad: Measures scientific reasoning abilities in the style of international Olympiad competitions.
  • Research: Evaluates real-world scientific research capabilities.

In initial evaluations, GPT-5.2 emerged as the top-performing model on both FrontierScience-Olympiad, scoring 77%, and Research, scoring 25%. This performance positions it ahead of other frontier models, including Claude Opus 4.5 and Gemini 3 Pro. The results indicate that current models can support structured reasoning aspects of research, though significant work remains to enhance their open-ended thinking capabilities.

FrontierScience encompasses over 700 textual questions, with 160 in its gold set, spanning subfields in physics, chemistry, and biology. FrontierScience-Olympiad features 100 questions collaboratively designed by 42 international Olympiad medalists and national team coaches. FrontierScience-Research includes 60 original research subtasks developed by 45 PhD scientists, including doctoral candidates, professors, and postdoctoral researchers.

For the Olympiad set, grading occurs through short answer verification. For the Research track, a rubric-based architecture with a 10-point scoring system evaluates open-ended tasks. This rubric assesses both the final answer and intermediate reasoning steps. A model-based grader, GPT-5, evaluates responses against these criteria. Each task’s creation involved selecting against internal models, which may bias evaluations against specific models.

Key performance results include:

  • FrontierScience-Olympiad Accuracy:
    • GPT-5.2: 77.1%
    • Gemini 3 Pro: 76.1%
    • Claude Opus 4.5: 71.4%
  • FrontierScience-Research Accuracy:
    • GPT-5.2: 25.2%
    • Claude Opus 4.5: 17.5%
    • Grok 4: 15.9%

Longer processing times, or higher reasoning efforts, correlated with improved accuracy for both GPT-5.2 and OpenAI o3. For instance, GPT-5.2’s accuracy on FrontierScience-Olympiad increased from 67.5% at “Low” reasoning effort to 77.1% at “XHigh” effort. Similarly, on FrontierScience-Research, GPT-5.2’s accuracy rose from 18.2% at “Low” to 25.2% at “XHigh.”

FrontierScience currently focuses on constrained problem statements and does not assess the generation of novel hypotheses or interactions with multimodal data. Developers plan to iterate on the benchmark, expanding it to new domains and integrating more real-world evaluations as models improve.

Featured image credit

]]> Huawei Mate 80 GTS and Pura X2 foldable tipped for March release https://dataconomy.com/2025/12/24/huawei-mate-80-gts-and-pura-x2-foldable-tipped-for-march-release/ Wed, 24 Dec 2025 09:28:55 +0000 https://dataconomy.com/?p=85768 Huawei is developing a new flagship smartphone, the Mate 80 GTS, which could launch in 2026. The company trademarked the moniker Mate 80 GTS in November. A Weibo tipster indicated the device might become official next year. Huawei has previously registered several GTS models, and a Mate 60 GTS was rumored to launch with a […]]]>

Huawei is developing a new flagship smartphone, the Mate 80 GTS, which could launch in 2026. The company trademarked the moniker Mate 80 GTS in November. A Weibo tipster indicated the device might become official next year.

Huawei has previously registered several GTS models, and a Mate 60 GTS was rumored to launch with a red color variant. The latest trademark suggests a new phone is in development.

The Mate 80 GTS is reportedly in its development phase and may launch alongside the second-generation Pura X foldable phone. SuperDimensional, a tipster, stated the Mate 80 GTS is a real product and could release by March next year. It is expected to be a top-end flagship model.

Sources indicate the handset will feature the Kirin 9030 Pro 5G chipset. Huawei may enhance this chip before the Mate 80 GTS release. The device could also be a “Fan Edition” or a “Cooling Fan model,” as GTS often signifies a gaming or high-performance product.

Featured image credit

]]> Taiwan export laws are driving AMD and Google to Samsung Texas https://dataconomy.com/2025/12/24/taiwan-export-laws-are-driving-amd-and-google-to-samsung-texas/ Wed, 24 Dec 2025 09:27:09 +0000 https://dataconomy.com/?p=85765 AMD and Google are discussing production of 2-nanometer chips with Samsung Electronics at its Taylor, Texas facility, driven by Taiwan’s strict technology export rules that prevent TSMC from deploying its most advanced processes overseas. Samsung’s Taylor plant is projected to be the only U.S.-based fabrication facility capable of leading-edge manufacturing by 2026. This positions Samsung […]]]>

AMD and Google are discussing production of 2-nanometer chips with Samsung Electronics at its Taylor, Texas facility, driven by Taiwan’s strict technology export rules that prevent TSMC from deploying its most advanced processes overseas.

Samsung’s Taylor plant is projected to be the only U.S.-based fabrication facility capable of leading-edge manufacturing by 2026. This positions Samsung as a key alternative for U.S. tech firms requiring advanced chip production domestically.

Samsung Executive Chairman Lee Jae-yong met AMD CEO Lisa Su and Tesla CEO Elon Musk during a recent U.S. trip. The meetings focused on potential foundry cooperation, as reported by South Korean media outlets.

Samsung is conducting sample tests of its second-generation 2nm process, designated SF2P, in collaboration with AMD. Separately, Google’s Tensor Processing Unit team visited the Taylor facility to negotiate production volumes for its 2-nm chips.

Featured image credit

]]> OpenAI launches Skills in Codex to supercharge agentic coding https://dataconomy.com/2025/12/24/openai-launches-skills-in-codex-to-supercharge-agentic-coding/ Wed, 24 Dec 2025 09:26:04 +0000 https://dataconomy.com/?p=85762 OpenAI launched Skills in Codex this week to let developers customize the Codex coding agent with task-specific capabilities using pre-built packages or custom scripts generated through natural language prompts. The Skills in Codex service combines instructions, resources, and optional scripts, allowing Codex to execute specific workflows with reliability. OpenAI detailed this functionality in its developer […]]]>

OpenAI launched Skills in Codex this week to let developers customize the Codex coding agent with task-specific capabilities using pre-built packages or custom scripts generated through natural language prompts.

The Skills in Codex service combines instructions, resources, and optional scripts, allowing Codex to execute specific workflows with reliability. OpenAI detailed this functionality in its developer documentation. Developers gain flexibility by crafting their own skills either through natural language descriptions or by writing scripts manually. Pre-built skills appear through GitHub repositories, providing ready-to-use options for common tasks.

This launch arrives shortly after OpenAI released GPT-5.2-Codex last week. OpenAI described GPT-5.2-Codex as its most advanced agentic coding model yet, setting the stage for enhanced customization features like Skills.

OpenAI’s Skills in Codex draws directly from the Agent Skills standard, originally created by Anthropic and released as an open specification on December 18. The specification resides at agentskills.io, complete with a reference SDK. Multiple platforms have integrated this standard, including Microsoft, GitHub, Cursor, Goose, Amp, OpenCode, and additional coding agents.

Mahesh Murag, a product manager at Anthropic, explained the initiative to VentureBeat. He stated, “We’re launching Agent Skills as an independent open standard with a specification and reference SDK available at https://agentskills.io.” Murag continued, “Microsoft has already adopted Agent Skills within VS Code and GitHub; so have popular coding agents like Cursor, Goose, Amp, OpenCode, and more.” This open approach facilitates widespread compatibility across developer tools.

Anthropic first rolled out Agent Skills in October 2023 as a feature within its Claude AI assistant. Since then, Anthropic established partnerships with Atlassian, Figma, Canva, Stripe, Notion, and Zapier. These collaborations deliver pre-built skills tailored to workflows in those platforms, expanding practical applications for AI-assisted development.

The development reflects broader industry efforts among major AI providers to advance coding agents. Amazon Web Services introduced powers for its Kiro coding agent during the re:Invent conference earlier this month. Powers function through Model Context Protocol servers and steering files, serving as expertise modules that equip the Kiro agent with immediate knowledge of particular technologies and frameworks. AWS characterized powers as these expertise modules that give the Kiro agent instant knowledge of specific technologies and frameworks.

Skills in Codex employs progressive disclosure for context management. This method loads only essential information during startup, then accesses full details as required. OpenAI’s developer documentation specifies installation options: skills install globally in the ~/.codex/skills/ directory or remain scoped to individual repositories, accommodating varied project needs.

Featured image credit

]]> Türkiye announces $3 billion Turkcell and Google Cloud partnership https://dataconomy.com/2025/12/24/turkiye-announces-3-billion-turkcell-and-google-cloud-partnership/ Wed, 24 Dec 2025 09:21:04 +0000 https://dataconomy.com/?p=85759 President Recep Tayyip Erdoğan announced Tuesday that Turkcell and Google Cloud will build Türkiye’s first hyperscale cloud region through a partnership backed by a $3 billion combined investment. The project, revealed at the 2025 TÜBİTAK and Academy of Sciences awards ceremony in Ankara, will operate between 2028 and 2029 to position Türkiye as a digital […]]]>

President Recep Tayyip Erdoğan announced Tuesday that Turkcell and Google Cloud will build Türkiye’s first hyperscale cloud region through a partnership backed by a $3 billion combined investment. The project, revealed at the 2025 TÜBİTAK and Academy of Sciences awards ceremony in Ankara, will operate between 2028 and 2029 to position Türkiye as a digital bridge between Europe, Asia, and the Middle East.

The announcement occurred roughly six weeks after Turkcell and Google Cloud signed a memorandum of understanding in November 2024 to develop the hyperscale data-center infrastructure. Erdoğan described the initiative during his address as one that will transform Türkiye into the data hub of its region. This facility will join Google Cloud’s network of 42 hyperscale regions worldwide.

The cloud region will consist of multiple data centers equipped with advanced computing power, high-performance storage systems, network infrastructure, and cybersecurity measures. These components will support extensive data processing and secure operations across various applications.

Turkcell CEO Ali Taha Koç commented in November on the agreement, stating, “This strategic partnership is more than a technology investment—it is a milestone for Türkiye’s digital future.” Turkcell anticipates that the region will produce more than $5 billion in annual economic value once fully operational. The infrastructure will maintain Turkish data within national borders and supply local facilities to comply with data-residency requirements for organizations.

Services from the cloud region will encompass artificial intelligence capabilities, data storage solutions, and cybersecurity protections, all delivered directly from Türkiye. Google Cloud CEO Thomas Kurian explained that the region will bring all of our cloud-computing services to Türkiye and connect it to our global network.

The hyperscale cloud project forms a key element of Türkiye’s broader digital-transformation strategy. Erdoğan highlighted in his speech the country’s progress in technology sectors such as artificial intelligence, autonomous systems, and unmanned technologies. He pointed out that Turkey has lowered its foreign dependency in defense technology from 80 percent.

Vice-President Cevdet Yılmaz referred to the partnership in November as a highly critical step that strengthens Türkiye’s digital sovereignty and regional positioning. Türkiye presently maintains approximately 160 megawatts of data-center capacity, with projections for expansion beyond 210 megawatts in the future.

The initiative corresponds to Türkiye’s National Technology Initiative and “Digital Türkiye” strategy, which seek to establish the country as a leader in digital technologies. Over the past 23 years, the government has directed 153 billion Turkish liras toward more than 36,000 research projects and allocated 46.5 billion liras in support to over 415,000 scientists, according to Erdoğan.

Featured image credit

]]> Trump clears Nvidia and AMD to resume advanced AI chip sales to China https://dataconomy.com/2025/12/24/trump-clears-nvidia-and-amd-to-resume-advanced-ai-chip-sales-to-china/ Wed, 24 Dec 2025 09:18:18 +0000 https://dataconomy.com/?p=85757 American chipmakers Nvidia and Advanced Micro Devices plan limited sales of advanced AI chips to China following President Donald Trump’s December 8, 2023, announcement permitting exports of Nvidia’s H200 chips in exchange for a 25 percent fee, reversing the Biden administration’s ban imposed over national security concerns. Trump specified that the policy extends to other […]]]>

American chipmakers Nvidia and Advanced Micro Devices plan limited sales of advanced AI chips to China following President Donald Trump’s December 8, 2023, announcement permitting exports of Nvidia’s H200 chips in exchange for a 25 percent fee, reversing the Biden administration’s ban imposed over national security concerns.

Trump specified that the policy extends to other chipmakers, including AMD and Intel. He stated on Truth Social that he informed Chinese President Xi Jinping of the decision, adding that Xi responded positively. This arrangement marks a reversal of restrictions in place for more than two years, during which advanced AI chip sales to China were prohibited.

AMD CEO Lisa Su confirmed in December 2023 that her company had secured export licenses and stood ready to pay a 15 percent fee to the U.S. government on approved sales. The licenses enable AMD to proceed with transactions compliant with the new policy framework established under the Trump administration.

Nvidia has notified Chinese clients of its intention to start shipping H200 AI chips before the Lunar New Year in mid-February 2026. Three sources familiar with the matter shared this information with Reuters. The initial shipments will draw from existing stock, totaling 5,000 to 10,000 chip modules. Each module contains multiple chips, making the volume equivalent to approximately 40,000 to 80,000 individual H200 chips.

Separately, Alibaba is evaluating a purchase of 40,000 to 50,000 MI308 AI accelerators from AMD, as reported by MLex. The MI308 serves as a China-specific product engineered to adhere to U.S. export regulations. It delivers substantial performance capabilities tailored for large-scale AI training and inference workloads.

Chinese regulators have conducted meetings with major technology firms, such as Alibaba, ByteDance, and Tencent, to evaluate demand for these incoming chips.

One proposal on the table mandates that each H200 purchase include a specific ratio of domestically produced chips. Buyers could face an approval process requiring them to justify why local alternatives fail to meet their requirements, according to the Financial Times. Beijing contemplates restricting access to the H200 to limited quantities only.

A source remarked to Reuters, “The whole plan is contingent on government approval.” The same source added, “Nothing is certain until we get the official go-ahead.” These statements underscore the pending nature of any transactions.

Chinese officials express concern that permitting imports of advanced U.S. chips could hinder development within the nation’s domestic AI chip sector. Should the deals proceed, Chinese AI developers would access computing power far exceeding current local options. The H200 offers performance nearly six times greater than the H20, the most advanced chip previously approved for export to China.

Featured image credit

]]> Google changes Pixel Launcher behavior in new update https://dataconomy.com/2025/12/24/google-changes-pixel-launcher-behavior-in-new-update/ Wed, 24 Dec 2025 09:14:51 +0000 https://dataconomy.com/?p=85754 Google updated the Pixel Launcher in a recent Pixel device update, altering the home screen search bar to open the standard Google Search UI instead of the Pixel Launcher’s unified search experience. Users can restore the prior unified search via ADB commands shared by jspirit and Kieron Quinn. The update modifies the search bar behavior […]]]>

Google updated the Pixel Launcher in a recent Pixel device update, altering the home screen search bar to open the standard Google Search UI instead of the Pixel Launcher’s unified search experience. Users can restore the prior unified search via ADB commands shared by jspirit and Kieron Quinn.

The update modifies the search bar behavior on Pixel devices running the Pixel Launcher. Previously, tapping the home screen search bar activated the Pixel Launcher’s unified search interface, which integrated various search functions directly within the launcher. Now, it directs users to the standard Google Search UI, a full-screen application interface provided by Google Search.

This standard Google Search UI offers direct access to AI Mode, a feature within Google Search that utilizes artificial intelligence for enhanced query processing and responses. However, the standard UI omits certain features present in the unified search and appears less integrated with the Pixel Launcher’s home screen environment.

To reverse this change, users employ ADB, a command-line tool that facilitates communication between a computer and an Android device. First, individuals must install ADB on their computer and enable USB debugging on the Pixel device.

With ADB configured and the device connected via USB, users execute two specific commands in sequence. The primary command overrides a device configuration flag named “enable_one_search” within the launcher module.

  • Override command: adb shell cmd device_config override launcher enable_one_search true. This setting to ‘true’ instructs the Pixel Launcher to revert to the unified search experience, bypassing the standard Google Search UI.
  • Restart command: adb shell am force-stop com.google.android.apps.nexuslauncher. This force-stops the Pixel Launcher process, prompting an automatic restart that loads the updated flag configuration.

To return to the standard Google Search UI, users repeat the process but substitute “true” with “false” in the override command, followed by the restart command.

Testing confirms the workaround functions on the Pixel’s December 2025 stable update, which represents the current production release for compatible Pixel devices. It also operates on the Android 16 QPR3 Beta 1 release, a pre-release version available to beta program participants for testing upcoming features and fixes.

Featured image credit

]]> Don’t miss: Google’s AI Pro annual plan is 50% off https://dataconomy.com/2025/12/24/dont-miss-googles-ai-pro-annual-plan-is-50-off/ Wed, 24 Dec 2025 09:12:50 +0000 https://dataconomy.com/?p=85751 Google has announced a significant holiday promotion offering a 50% discount on its AI Pro annual plan, though the deal comes with a major restriction. The price cut is exclusively available to new subscribers, meaning existing Google One members are ineligible to redeem the offer. Another gift before 2026 🎁 For a limited time, new […]]]>

Google has announced a significant holiday promotion offering a 50% discount on its AI Pro annual plan, though the deal comes with a major restriction. The price cut is exclusively available to new subscribers, meaning existing Google One members are ineligible to redeem the offer.

The AI Pro plan bundles several of Google’s advanced tools, including access to Nano Banana Pro, the Deep Research tool, and the Gemini 3 Pro model, alongside 2TB of cloud storage. Subscribers also gain higher usage limits for services like NotebookLM and Jules, access to the Flow tool, and AI-powered calling features for local businesses in the US.

According to clarifications from a Google employee, the promotion is currently active in select regions, including the UK, the EU, and Chile. The offer is expected to run until December 31, giving eligible new users a limited window to secure the reduced rate.

Featured image credit

]]> Why Elon Musk’s Grok AI is becoming a core tool for the Pentagon https://dataconomy.com/2025/12/24/why-elon-musks-grok-ai-is-becoming-a-core-tool-for-the-pentagon/ Wed, 24 Dec 2025 09:09:27 +0000 https://dataconomy.com/?p=85748 The Pentagon announced on Monday via press release the addition of frontier AI systems based on the Grok family of models to its AI Arsenal, integrated into the GenAI.mil platform, to enhance efficiency across operations. The GenAI.mil platform launched earlier this month and incorporates Google’s Gemini for Government, as detailed in a prior press release. […]]]>

The Pentagon announced on Monday via press release the addition of frontier AI systems based on the Grok family of models to its AI Arsenal, integrated into the GenAI.mil platform, to enhance efficiency across operations.

The GenAI.mil platform launched earlier this month and incorporates Google’s Gemini for Government, as detailed in a prior press release. This platform serves as a comprehensive AI toolset for Pentagon personnel engaged in various tasks requiring advanced computational capabilities.

Pete Hegseth, identified as U.S. Secretary of War, provided a statement in the release: “AI tools present boundless opportunities to increase efficiency, and we are thrilled to witness AI’s future positive impact across the War Department.” The quote emphasizes the role of AI in operational streamlining within the department.

Implementation of new AI products from an Elon Musk-owned company will occur in early 2026. These products enable the secure handling of Controlled Unclassified Information (CUI) within daily workflows. They also provide access to real-time global insights from the X platform. This access delivers a decisive information advantage to War Department personnel by supplying current data streams directly into their processes.

The incorporation of Grok-derived models constitutes a second set of AI models available on the Pentagon’s bespoke AI platform. These models apply specifically to AI-intensive tasks performed by department staff, expanding the range of tools beyond the initial offerings.

An executive order issued in April from the Trump administration directed reviews aimed at improving Pentagon efficiency. The order specified goals such as eliminating or revising any unnecessary supplemental regulations or other internal guidance. This initiative targeted administrative burdens to foster a more streamlined operational environment.

Precedent for collaboration between tech firms and defense exists across political lines. During the Biden era, former Google CEO Eric Schmidt participated in an effort to significantly increase AI-related spending on defense and security programs in the federal government. Senator Elizabeth Warren identified this involvement as a potential conflict of interest, citing Schmidt’s prior leadership at Google.

Companies including xAI and Google continue efforts to align their technologies with defense industry needs. These pursuits involve developing and supplying AI solutions tailored for military and security applications, building on established patterns of partnership between private sector innovators and government entities responsible for national defense.

Featured image credit

]]> Retailers struggle to spot AI-doctored images in refund requests https://dataconomy.com/2025/12/24/retailers-struggle-to-spot-ai-doctored-images-in-refund-requests/ Wed, 24 Dec 2025 09:07:16 +0000 https://dataconomy.com/?p=85746 Generative AI is democratizing e-commerce fraud, with scammers increasingly using AI-generated images to secure refunds for items they never return. According to Wired, online shopping platforms—which have long relied on customer-submitted photos to validate refund requests—are struggling to distinguish between real and fabricated damage claims. The issue has become particularly acute on Chinese social media […]]]>

Generative AI is democratizing e-commerce fraud, with scammers increasingly using AI-generated images to secure refunds for items they never return. According to Wired, online shopping platforms—which have long relied on customer-submitted photos to validate refund requests—are struggling to distinguish between real and fabricated damage claims.

The issue has become particularly acute on Chinese social media platforms like RedNote and Douyin. Wired highlights a recent case involving a crab merchant named Gao Jing, who received a refund claim accompanied by a video of “dead” crabs. The fraud was exposed when the seller noticed biological inconsistencies in the footage: the crabs had the wrong number of legs, and the sex of the crabs changed between clips. Police investigation confirmed the video was AI-generated, leading to the buyer’s detention.

The problem extends beyond individual scammers. Michael Reitblat, CEO of fraud detection firm Forter, told Wired that AI-doctored refund claims have increased by more than 15% globally since the start of the year. Organized crime groups are reportedly using these tools to automate fraud at scale, utilizing rotating IP addresses to flood retailers with fake claims for “damaged” home goods. While some merchants are deploying their own AI tools to detect doctored images, the technology remains imperfect, forcing retailers to consider stricter return policies that could inconvenience honest shoppers.

In a lighter development from the same report, Wired notes the viral success of “Yichang Beer.” Despite its branding featuring Chinese characters and dragon imagery, the beer is actually brewed in Kazakhstan and was originally sold in Russian-speaking markets. The product’s label claims it has been brewed since 1858—decades before beer was introduced to China. Ironically, after going viral on Chinese social media for its “fake” heritage, the beer is now being imported and sold in the actual city of Yichang.

Featured image credit

]]> Amazon adds Angi, Expedia, Square and Yelp to Alexa+ https://dataconomy.com/2025/12/24/amazon-adds-angi-expedia-square-and-yelp-to-alexa/ Wed, 24 Dec 2025 09:01:13 +0000 https://dataconomy.com/?p=85743 Amazon announced on Thursday that its AI-powered digital assistant Alexa+ will integrate with Angi, Expedia, Square, and Yelp starting in 2026 to enable users to book hotels, obtain home-service quotes, and schedule salon appointments. The expansions target practical consumer needs by connecting Alexa+ directly to service providers. Angi provides quotes for home services such as repairs […]]]>

Amazon announced on Thursday that its AI-powered digital assistant Alexa+ will integrate with Angi, Expedia, Square, and Yelp starting in 2026 to enable users to book hotels, obtain home-service quotes, and schedule salon appointments.

The expansions target practical consumer needs by connecting Alexa+ directly to service providers. Angi provides quotes for home services such as repairs and maintenance tasks. Expedia handles hotel comparisons, bookings, and management. Square supports scheduling for salon appointments, streamlining personal care reservations. Yelp contributes local business discovery and reviews within these interactions. These features activate through voice commands, allowing seamless access without switching devices or apps.

Expedia integration permits detailed user requests. Customers compare options based on location, price, and amenities. They book reservations instantly and manage changes like cancellations or modifications. Personalized recommendations emerge from spoken preferences. For example, a user states, “Can you find me pet-friendly hotels for this weekend in Chicago?” Alexa+ processes the query, filters results, and presents matches with availability and ratings.

Alexa+ already connects to Fodor for travel planning, OpenTable for restaurant bookings, Suno for music generation, Ticketmaster for event tickets, Thumbtack for local services, and Uber for ride-hailing. These partnerships form the foundation, handling diverse tasks from dining reservations to transportation.

Amazon positions Alexa+ to facilitate a broad array of online services through natural-language conversations. This approach mirrors ChatGPT’s app integrations within its chatbot interface. Users engage in fluid exchanges, issuing commands like requesting an Uber ride or securing an OpenTable dinner table. The system supports back-and-forth dialogue, where initial requests refine progressively based on follow-up clarifications or adjustments.

Early adopters demonstrate usage patterns. Amazon reports strong engagement with home and personal service providers, including Thumbtack for task-based hires and Vagaro for salon and spa bookings. These interactions highlight preferences for service-oriented queries over other categories.

The industry explores AI assistants as platforms for app-like functions to distribute AI capabilities widely. This model shifts from web or mobile app interactions to voice-driven commands. Adoption hinges on perceived simplicity matching or exceeding traditional methods. Users accustomed to app stores or browsers require minimal friction in transitions.

AI providers must replicate app store breadth, which offers curated selections narrower than the open web. Alternatively, they excel by delivering timely suggestions at opportune moments. Prompts avoid perceptions of intrusive advertising, maintaining user trust through relevance and restraint.

Featured image credit

]]> Aflac data breach affected 22.65M customers https://dataconomy.com/2025/12/24/aflac-data-breach-2025-explained/ Wed, 24 Dec 2025 08:59:08 +0000 https://dataconomy.com/?p=85740 U.S. insurance giant Aflac confirmed a data breach on Tuesday it has begun notifying 22.65 million customers after hackers stole their personal and health information in a June cyberattack. In June, Aflac disclosed the data breach, revealing that hackers accessed customers’ personal information, including Social Security numbers and health information, but did not specify the […]]]>

U.S. insurance giant Aflac confirmed a data breach on Tuesday it has begun notifying 22.65 million customers after hackers stole their personal and health information in a June cyberattack.

In June, Aflac disclosed the data breach, revealing that hackers accessed customers’ personal information, including Social Security numbers and health information, but did not specify the number of victims at that time. The breach exposed sensitive details for a substantial portion of the company’s customer base.

A filing with the Texas attorney general outlined the scope of the stolen data. It included customer names, dates of birth, home addresses, government-issued ID numbers such as passports and state ID cards, driver’s license numbers, Social Security numbers, and medical and health insurance information. This comprehensive set of records heightens risks associated with the unauthorized access.

In a separate filing with the Iowa attorney general, Aflac provided details on the perpetrators. The company stated that the cybercriminals responsible for the breach “may be affiliated with a known cyber‑criminal organization; federal law enforcement and third-party cybersecurity experts have indicated that this group may have been targeting the insurance industry at large.” These disclosures point to coordinated efforts against the sector.

Scattered Spider, an amorphous collective of primarily young English‑speaking hackers, was targeting the insurance industry at the time of the breach. Aflac’s description aligns with activities attributed to this group.

Metric Details
Total individuals affected 22,654,000
Percentage of customers ~45% of Aflac’s total base
Breach discovery date June 12, 2025
Notification date December 19 – 23, 2025
Suspected threat actor Scattered Spider (UNC3944)
Regulatory filings Texas, Iowa, and California AG Offices

Featured image credit

]]> Xbox cloud gaming comes to Amazon Fire TV models https://dataconomy.com/2025/12/24/xbox-cloud-gaming-comes-to-amazon-fire-tv-models/ Wed, 24 Dec 2025 08:56:12 +0000 https://dataconomy.com/?p=85737 Xbox has expanded its cloud gaming support to select Amazon Fire TV smart TV models, specifically the Fire TV 4-Series and the Fire TV Omni QLED Series. This integration allows users to play Xbox games directly on their televisions without requiring a console or an external streaming stick. The service is accessible via the Xbox […]]]>

Xbox has expanded its cloud gaming support to select Amazon Fire TV smart TV models, specifically the Fire TV 4-Series and the Fire TV Omni QLED Series. This integration allows users to play Xbox games directly on their televisions without requiring a console or an external streaming stick. The service is accessible via the Xbox app on these devices, joining the lineup of previously supported Amazon streaming media players.

To utilize this feature, users require an active Game Pass subscription—with plans ranging from $10 to $30 per month—and a compatible Bluetooth wireless controller. In addition to the standard Game Pass library, the update supports the “stream your own game” functionality, enabling players to stream hundreds of select titles they already own, even if those games are not currently in the Game Pass catalog. Amazon has indicated that this support will extend to more TV models in the future.

Featured image credit

]]> Porsche 2025 holiday ad goes viral for using absolutely no AI https://dataconomy.com/2025/12/24/porsche-2025-holiday-ad-goes-viral-for-using-absolutely-no-ai/ Wed, 24 Dec 2025 08:53:41 +0000 https://dataconomy.com/?p=85734 Porsche released a Christmas advertisement featuring hand-drawn and 3D animation, produced by Parallel Studio. The short depicts a man and his Porsche attending car meets and taking nighttime drives during the festive season. Parallel Studio, experts in animation and illustration, created the advertisement. It presents a slice-of-life storyline centered on the man and his vehicle. […]]]>

Porsche released a Christmas advertisement featuring hand-drawn and 3D animation, produced by Parallel Studio. The short depicts a man and his Porsche attending car meets and taking nighttime drives during the festive season.

Parallel Studio, experts in animation and illustration, created the advertisement. It presents a slice-of-life storyline centered on the man and his vehicle. The production includes a stripped-back musical score. The narrative avoids saccharine holiday elements, relying instead on iconic imagery within nostalgic and dreamy animation styles.

The advertisement emerged amid controversy surrounding AI-generated Christmas ads from other brands. Coca-Cola and McDonald’s have produced such campaigns, contributing to audience skepticism toward brand advertising. Social media users accused Porsche’s advertisement of using AI technology.

Porsche addressed the allegations by sharing behind-the-scenes footage. This material documents the creation process from initial rough sketches through to the completed animation. The footage emphasizes the human craftsmanship involved in every stage of production.

The advertisement contrasts with AI-driven holiday campaigns by maintaining traditional animation techniques. It focuses on everyday moments involving the Porsche model, set against nighttime and social driving scenes. Car enthusiasts and creative professionals viewed the short online, noting its detailed hand-drawn elements combined with 3D effects.

Featured image credit

]]> ServiceNow agrees to buy Armis for record $7.75 billion as ServiceNow stock drops https://dataconomy.com/2025/12/23/servicenow-armis-deal-affects-servicenow-stock/ Tue, 23 Dec 2025 16:21:04 +0000 https://dataconomy.com/?p=85706 ServiceNow (NYSE: NOW) has entered into a definitive agreement to acquire the cybersecurity firm Armis for $7.75 billion in cash. This represents the largest acquisition in the company’s history. While the deal expands ServiceNow’s capabilities in device security and creates a unified “security exposure and operations stack,” the announcement has weighed heavily on ServiceNow stock […]]]>

ServiceNow (NYSE: NOW) has entered into a definitive agreement to acquire the cybersecurity firm Armis for $7.75 billion in cash. This represents the largest acquisition in the company’s history. While the deal expands ServiceNow’s capabilities in device security and creates a unified “security exposure and operations stack,” the announcement has weighed heavily on ServiceNow stock as investors react to the high price tag.

Why ServiceNow stock fell

The market response was immediate and negative. When reports of the deal first surfaced, ServiceNow stock dropped between 5% and 7% in pre-market trading. The finalized agreement values Armis at roughly 23 times its annual recurring revenue (ARR), which recently surpassed $340 million. This is a steep premium, even for the high-growth cybersecurity sector.

This volatility reflects investor concern over the company’s spending habits. ServiceNow recently acquired Moveworks for $2.85 billion, and this new, larger expenditure has raised questions about cash flow. Shareholders are wary that the company might be relying too heavily on expensive acquisitions rather than organic growth, despite Armis‘s impressive year-over-year ARR growth of over 50%.

Who is Armis?

Founded in 2015, Armis specializes in cyber exposure management and securing “unmanaged” devices. These are assets that cannot easily run traditional security software, such as medical equipment in hospitals, industrial control systems (OT) in factories, and smart office technology (IoT).

By purchasing Armis, ServiceNow fills a specific gap in its portfolio. It gains the ability to see and secure the millions of connected devices that interact with corporate networks but often remain invisible to standard IT tools. “AI is transforming the threat landscape faster than most organizations can adapt. Every connected asset has become a potential point of vulnerability,” said Yevgeny Dibrov, co-founder and CEO of Armis.

The strategy: Building the security platform of tomorrow

ServiceNow began in 2003 as a tool for IT service management but has evolved into the “AI control tower for business reinvention.” Under CEO Bill McDermott and President Amit Zavery, the goal is to provide a unified system that manages everything from employee workflows to security incidents.

Integrating Armis allows ServiceNow to offer a comprehensive view of a company’s digital assets. The plan is to feed the real-time device data from Armis directly into the ServiceNow AI Platform. This helps security teams identify risks and automate responses without needing to switch between different software tools.

“Together with Armis, we will deliver an industry-defining strategic cybersecurity shield for real-time, end-to-end proactive protection across all technology estates,” said Amit Zavery.

The deal is expected to close in the second half of 2026. Until then, the performance of ServiceNow stock will likely depend on how well the company explains the long-term value of this investment to skeptical shareholders.

The information provided on Dataconomy is for general informational purposes only and does not constitute financial, investment, or trading advice. Articles, analyses, and opinions reflect the authors’ views at the time of publication and may change without notice.

Featured image credit

]]> Beyond Magic: Strategic Realism in AI Revenue Generation https://dataconomy.com/2025/12/23/strategic-realism-ai-revenue-generation/ Tue, 23 Dec 2025 11:31:42 +0000 https://dataconomy.com/?p=85661 As 2025 draws to a close, the bill for the Artificial Intelligence boom has officially come due. While corporate roadmaps remain cluttered with generative pilots, the gap between “magic” and “margin” in AI revenue generation is widening. Recent data paints a stark picture of this “ROI Gap.” According to a December 2025 study from MIT, […]]]>

As 2025 draws to a close, the bill for the Artificial Intelligence boom has officially come due. While corporate roadmaps remain cluttered with generative pilots, the gap between “magic” and “margin” in AI revenue generation is widening.

Recent data paints a stark picture of this “ROI Gap.” According to a December 2025 study from MIT, nearly 95% of enterprise AI projects are currently failing to deliver measurable returns. Similarly, Forrester reports that only 15% of executives have seen any improvement in profit margins from their AI investments over the last year.

The uncomfortable silence in boardrooms is no longer about whether the technology works – it’s about why it isn’t paying.

Moving from a promising demo to a revenue-generating engine requires more than just clean data and good models; it requires a fundamental shift in strategy – one that bridges the divide between executive ambition and engineering reality.

To navigate this divide, we turn to Vladyslav Chekryzhov, Director of Data Science & AI at AUTODOC. Operating across 27 distinct European markets, Chekryzhov sits at the rare intersection of executive product ownership and hands-on system architecture. Unlike the theoretical futurists often dominating the headlines, his mandate is grounded in the high-stakes reality of major e-commerce: delivering production-grade systems that directly influence pricing, retention, and customer loyalty.

He represents a discipline we might call “Revenue Realism” – the understanding that an AI model is only as valuable as its ability to survive in the wild and deliver measurable commercial impact.

Here are five strategic pivots required to turn AI hype into P&L reality.

The “Utility Filter”: Ruthless Prioritization

The first trap many organizations fall into is the “solution in search of a problem.” With the barrier to entry for Generative AI lower than ever, the temptation to build “cool” features is high. However, revenue generation requires a disciplined refusal to chase trends that don’t move the needle.

For Chekryzhov, the distinction between a feature and a business driver is stark. It begins not with code, but with financial modeling.

“Ultimately, prioritizing any AI/ML initiatives comes down to the discipline of building assumptions. Don’t rely on intuition; model the impact first – make money in Excel before the code is even written.”

He categorizes initiatives into three levels: Optimizing current economics (Level 1), Unlocking new product economics (Level 2), and Remodeling the business ecosystem (Level 3). The danger zone, he notes, is usually Level 3, where strategic stories often mask weak assumptions.

“The common failure mode is building an expensive toy… I force a vendor test: would we pay for this capability at vendor rates (e.g., OpenAI) and still maintain margins? If there’s no defensible path to revenue growth or a step-change in operating expenses, it’s just a costly experiment.”

Balancing the Algorithm: Pricing vs. Retention

In e-commerce, AI is often tasked with optimization. But optimization is rarely zero-sum. A model designed to maximize immediate margin (Dynamic Pricing) might inadvertently punish long-term loyalty (Retention).

Chekryzhov argues that managing this tension isn’t about finding the perfect neural network architecture, but about establishing the proper organizational boundaries.

“The minimum that works surprisingly well is culture, not architecture: rigorous experimentation with the right guardrails. Every pricing or promo change is measured not only on immediate efficiency but also on the “halo effects”: how it shifts behavior across cohorts and segments… We define upfront which metrics are allowed to move, in which direction, and by how much. If a margin win comes with a retention or CLV hit outside those bounds, it’s not a win.”

To implement this technically, he suggests avoiding “black box” monoliths in favor of a layered approach that gives business leaders control without requiring a full model retrain.

“One practical way to do it is a cascade of models: a pricing model proposes candidate prices, then lightweight models predict user outcomes and act as a filter or a weighting reranker. The benefit is control: you can adjust business logic by changing the final configuration rather than retraining the heavy model every time priorities shift.”

The “Production Gap”: Where ROI Dies

A Proof of Concept (POC) is a controlled experiment; production is a war zone. Many revenue projections fail because they underestimate the engineering overhead required to keep a model running at scale.

Chekryzhov warns that AI introduces a specific type of technical debt that traditional software engineers often miss: non-determinism.

“The honest answer is that a successful PoC doesn’t prove you have a scalable product… The model is non-deterministic: a rerun can produce different outputs. That explodes debugging cost, makes incidents harder to reproduce, and raises the bar for monitoring. Technical debt shows up sooner in AI systems than in traditional software, becoming a tax on the entire team’s development speed.”

Strategically, this means your ROI calculation must include the cost of reliability. If you only budget for development and not for the “tax” of maintenance, your margins will evaporate.

“The best investments I’ve seen here aren’t exotic… I push for basic hygiene (MLOps culture and the continuous process of ML systems design), the parts that don’t go out of date: measurable quality, debuggability, and reversibility.”

Isolating the Signal: The Attribution Challenge

Perhaps the most complex strategic question to answer is: “Did the AI do that?” In a complex ecosystem involving dozens of markets, seasonality, and marketing spend, attributing revenue to specific sources is statistically messy. Yet, without clear attribution, continued investment is impossible to justify to the C-suite.

Chekryzhov approaches this with the rigor of a scientist, rejecting the idea that complex models generate trust. Instead, he relies on counterfactuals – proving what would have happened in the absence of the AI.

“The only way to claim ‘AI drove X’ with a straight face is to anchor on a credible counterfactual. I rely on two families of evidence: randomized experiments (A/B) when feasible, and quasi-experimental methods when not. If the decision matters beyond the test window, we add a global holdout to the A/B setup: a persistent control group that never sees the feature. It’s painful – you’re literally losing money. But it’s often the only reliable link to reality.”

“For the C-suite, the message is consistent: trust doesn’t come from a complex model. It comes from a transparent approach and a measurement design you can explain clearly.”

Safety Rails: Trusting the Machine

Finally, automating revenue decisions – such as bidding or pricing – carries inherent risks. A “hallucinating” chatbot is embarrassing; a pricing algorithm that sells inventory at a 90% loss is catastrophic.

Strategic implementation requires a “human-in-the-loop” philosophy that evolves into “human-over-the-loop” governance. Chekryzhov advises assessing the cost of error before granting autonomy.

“I start with ML/AI system design, and one artifact matters most here: the cost of error. If the downside is high and hard to reverse, I don’t chase full autonomy… When the risk profile is acceptable, I like an “autonomy slider.” Early iterations are human-validated. As you accumulate data and confidence, you move the slider toward automation in controlled steps.”

Even when a system is fully autonomous, it must operate within strict bounds defined by the business, not the model.

“Autonomy must be bounded by policy-as-code. The system should have explicit constraints, circuit breakers, and safe fallbacks… You’re not debating autonomy in theory; you’re earning it.”

AI Revenue Needs a Maturity Upgrade

The transition from AI experimentation to AI revenue is not a technological upgrade; it is a maturity upgrade. It requires moving away from the allure of novelty and embracing the rigor of engineering, the complexity of attribution, and the discipline of prioritization.

As Chekryzhov’s experience at AUTODOC demonstrates, the companies that will win are not necessarily those with the most advanced models, but those with the most robust bridges between data science and business strategy.

]]> China fails to recover booster on first flight of Long March 12A rocket https://dataconomy.com/2025/12/23/china-fails-to-recover-booster-on-first-flight-of-long-march-12a-rocket/ Tue, 23 Dec 2025 10:14:55 +0000 https://dataconomy.com/?p=85611 China launched its first state-owned reusable rocket, the Long March 12A, from the Jiuquan Satellite Launch Center in northwestern China on Monday. The mission succeeded in orbital insertion and payload deployment but failed to recover the first-stage booster during a vertical landing attempt 250 kilometers downrange. The Long March 12A, developed by the Shanghai Academy […]]]>

China launched its first state-owned reusable rocket, the Long March 12A, from the Jiuquan Satellite Launch Center in northwestern China on Monday. The mission succeeded in orbital insertion and payload deployment but failed to recover the first-stage booster during a vertical landing attempt 250 kilometers downrange.

The Long March 12A, developed by the Shanghai Academy of Spaceflight Technology under the China Aerospace Science and Technology Corporation, lifted off as China’s inaugural state-owned effort in reusable rocket technology. State news agency Xinhua reported that while the rocket’s second stage entered its planned orbit and successfully deployed its payload, the first-stage booster was not successfully retrieved. This outcome represents China’s second unsuccessful booster recovery attempt within the same month.

Earlier this month, on December 3, private Chinese firm LandSpace’s Zhuque-3 rocket achieved orbital insertion but encountered a catastrophic failure during its landing phase. After a successful re-entry, the Zhuque-3’s first-stage booster lost an engine during the landing burn and crashed at the edge of the recovery pad in a fireball.

The United States holds the distinction as the only country to have successfully returned an orbital-class booster to Earth. SpaceX achieved the first such landing with its Falcon 9 rocket in December 2015. More recently, Blue Origin landed the first-stage booster of its New Glenn rocket on a drone ship during the vehicle’s second flight attempt in November 2025.

China’s commercial and state-owned space developers pursue reusable rocket technology to enable more frequent launches at reduced costs. The Long March 12A operates as a two-stage vehicle powered by liquid oxygen and methane. It measures 69 meters in height with a 3.8-meter diameter and possesses a payload capacity of 12,000 kilograms to low Earth orbit.

For this maiden flight, the Long March 12A’s first stage targeted a vertical landing on a designated pad approximately 250 kilometers downrange from the Jiuquan launch site. This reusability approach supports China’s plans to construct large satellite constellations, such as the Guowang network, which intends to deploy nearly 13,000 satellites by the 2030s.

Both the Long March 12A and Zhuque-3 missions accomplished their primary objectives by delivering payloads to the planned orbits. These successes highlight advancements in the precise sequencing of maneuvers essential for booster recovery operations. Investigations continue to determine the exact causes of the landing failures in both cases.

Featured image credit

]]> Fans are calling the Battlefield 6 Windchill bundle “AI slop” https://dataconomy.com/2025/12/23/fans-are-calling-the-battlefield-6-windchill-bundle-ai-slop/ Tue, 23 Dec 2025 10:12:39 +0000 https://dataconomy.com/?p=85609 According to Kotaku, Electronic Arts is investigating accusations that generative AI-generated content appears in Battlefield 6’s Season 1 winter cosmetics after fans identified anomalies in paid stickers. The claims emerged from player scrutiny of the Windchill bundle, priced at just under $10, prompting concerns over production practices. The accusations surfaced over the weekend when players […]]]>

According to Kotaku, Electronic Arts is investigating accusations that generative AI-generated content appears in Battlefield 6’s Season 1 winter cosmetics after fans identified anomalies in paid stickers. The claims emerged from player scrutiny of the Windchill bundle, priced at just under $10, prompting concerns over production practices.

The accusations surfaced over the weekend when players shared images of a sticker from the Windchill bundle. This cosmetic depicts a character aiming down the scope of a double-barreled rifle. Fans quickly labeled it as low-quality AI-generated imagery. A viral post on the game’s subreddit stated, “Two barrels on the M4A1, sure. I would literally prefer to have no sticker than some low-quality AI-generated garbage. You can look at BO7 and see how many favors AI-generated rewards won with them.” The post highlighted the rifle’s unrealistic design, as the M4A1 carbine features a single barrel.

Learn: AI slop describes digital content generated by artificial intelligence that is perceived as lacking in quality and thoughtfulness.

This criticism spurred fans to conduct an “AI vibe check” on additional cosmetics within Battlefield 6. They examined various in-game stickers and other items for similar irregularities. One example includes a bear sticker where the paws display more than ten claws. Although not as pronounced as anomalies in other titles, such as the six-fingered Santa Zombie loading screen in Call of Duty: Black Ops 6, these observations have fueled speculation about broader use of generative AI tools in post-launch development.

Players question whether Battlefield 6’s ongoing content creation involves outsourcing designs to generative AI services. Such practices would conflict with assurances from EA Vice President Rebecka Coutaz, who oversees DICE and other Battlefield development teams. In a BBC interview earlier this year, Coutaz stated that fans would not see AI-generated imagery in the final game. She clarified that the technology serves earlier production phases “to allow more time and more space to be creative.” The double-barreled rifle sticker directly challenges this commitment regarding final content.

Generative AI content can infiltrate large-scale titles through established workflows. Concept artists often generate AI references for initial designs, then manually paint over them digitally to refine details. Another pathway involves subcontracted outsourcing teams that deliver cosmetics passing through expedited review processes. These methods, while efficient, risk unpolished AI artifacts reaching players if oversight lapses.

EA has actively promoted AI integration across its operations. During the company’s earnings report in May, CEO Andrew Wilson described AI to investors as “a powerful accelerator of creativity, innovation, and player connection.” Employees in departments including art, quality assurance, marketing, and customer service received encouragement to incorporate AI tools into daily tasks. Last month, reports confirmed AI usage in generating the full cover art for the Deluxe Edition of NHL 26.

The double-barreled rifle serves as a prominent indicator of AI involvement due to its anatomical impossibility on the weapon. Other detected errors, like the excessive claws on the bear, could arise from quality control shortcomings rather than direct AI generation. Battlefield 6’s post-launch schedule demands rapid deployment of new content and microtransactions, contributing to these issues. A source familiar with the game’s production explained that the current roadmap’s extensive scope has strained resources, reducing capacity for comprehensive reviews of all submitted materials.

Player backlash against generative AI in gaming has escalated toward year’s end, positioning Battlefield 6 amid widespread scrutiny. EA has not responded to requests for comment on the investigation or cosmetics production.

Featured image credit

]]> Why DIG AI is the most dangerous malicious AI of 2025 https://dataconomy.com/2025/12/23/why-dig-ai-is-the-most-dangerous-malicious-ai-of-2025/ Tue, 23 Dec 2025 10:07:41 +0000 https://dataconomy.com/?p=85607 Cybersecurity firm Resecurity (via Help Net Security) exposed DIG AI, an uncensored artificial intelligence assistant on the darknet that enables criminals to generate malware, create child sexual abuse material, and obtain detailed instructions for manufacturing explosives without safety restrictions. The tool operates via the Tor network. Resecurity first detected DIG AI on September 29, 2025. […]]]>

Cybersecurity firm Resecurity (via Help Net Security) exposed DIG AI, an uncensored artificial intelligence assistant on the darknet that enables criminals to generate malware, create child sexual abuse material, and obtain detailed instructions for manufacturing explosives without safety restrictions. The tool operates via the Tor network.

Resecurity first detected DIG AI on September 29, 2025. The platform experienced a surge in adoption during the final quarter of 2025, particularly amid the winter holiday season when levels of illegal activity reached records. The creator, operating under the alias “Pitch,” presented DIG AI as a criminal alternative to mainstream artificial intelligence platforms. Access requires no account registration.

Resecurity’s report dated December 17 details the platform’s foundation on a jailbroken version of ChatGPT Turbo. It provides specialized models tailored to distinct functions. DIG-Uncensored handles generation of prohibited content. DIG-GPT processes text-based tasks. DIG-Vision supports creation of deepfakes and illicit imagery.

This development aligns with a broader pattern in weaponized artificial intelligence. Resecurity recorded an increase exceeding 200 percent in mentions and use of malicious artificial intelligence tools across cybercriminal forums from 2024 to 2025. Such tools fall under the category of dark large language models, which consist of systems either constructed from scratch or derived from legitimate artificial intelligence with safety restrictions eliminated.

Predecessors in this category include FraudGPT and WormGPT. Resecurity analysts performed extensive testing on DIG AI. They verified its capacity to produce functional malicious code, such as obfuscated JavaScript backdoors intended to compromise web applications. Tasks like code obfuscation require three to five minutes owing to constrained computing resources.

Operators address these delays through premium services that charge fees, establishing a “Crime-as-a-Service” model. DIG AI also supports production of child sexual abuse material by generating hyper-realistic images. This occurs through entirely synthetic content or by altering benign images of real minors.

Resecurity collaborated with law enforcement authorities to gather and preserve evidence documenting threat actors’ use of the platform for highly realistic child sexual abuse material. The firm identified criminal artificial intelligence as presenting heightened threats prior to major global events in 2026, such as the Winter Olympics in Milan and the FIFA World Cup.

These tools allow bad actors to expand operations and circumvent content protection policies. Resecurity described the situation as a new frontier in which criminals develop and sustain custom infrastructure for artificial intelligence operations, terming it the “fifth domain of warfare: cyber.”

Featured image credit

]]> Nissan data breach is real and you might be affected https://dataconomy.com/2025/12/23/nissan-data-breach-is-real-and-you-might-be-affected/ Tue, 23 Dec 2025 10:05:58 +0000 https://dataconomy.com/?p=85604 Nissan Motor Co. Ltd. confirmed that approximately 21,000 customers of Nissan Fukuoka Sales Co., Ltd., in Fukuoka, Japan, had personal data exposed in a September Red Hat breach through unauthorized access to data servers. The Japanese multinational automobile manufacturer, headquartered in Yokohama, Japan, produces more than 3.2 million cars annually and employs 120,000 people. Nissan […]]]>

Nissan Motor Co. Ltd. confirmed that approximately 21,000 customers of Nissan Fukuoka Sales Co., Ltd., in Fukuoka, Japan, had personal data exposed in a September Red Hat breach through unauthorized access to data servers.

The Japanese multinational automobile manufacturer, headquartered in Yokohama, Japan, produces more than 3.2 million cars annually and employs 120,000 people. Nissan maintains operations across Japan, North America, Europe, and Asia. The company disclosed its indirect involvement in the Red Hat security incident, which stemmed from Red Hat’s role in developing customer management systems for Nissan’s sales companies.

In its announcement, Nissan stated: “Nissan Motor Co., Ltd. received a report from Red Hat, the company it commissioned to develop customer management systems for its sales companies, that unauthorized access to its data servers had resulted in the data being leaked. It was later confirmed that the data leaked by the company contained some customer information from Nissan Fukuoka Sales Co., Ltd.”

The affected customers, who had purchased vehicles or received services at Nissan locations in Fukuoka, numbered around 21,000. Their leaked information encompassed the following details:

  • Full names of individuals associated with the purchases or services.
  • Physical addresses linked to customer records in the sales operations.
  • Phone numbers provided during vehicle transactions or service visits.
  • Email addresses used for communications related to Nissan dealings in Fukuoka.
  • Customer data used in sales operations, including records maintained for business purposes by Nissan Fukuoka Sales Co., Ltd.

Nissan specified that no financial information, such as credit card details, appeared in the leaked data.

Red Hat, a U.S.-based enterprise software company, disclosed the breach in early October 2025. The incident involved the theft of hundreds of gigabytes of sensitive data from 28,000 private GitLab repositories. The Crimson Collective threat actor initially claimed responsibility for the hack. Subsequently, the ShinyHunters group hosted samples of the stolen data on their extortion platform, exerting direct pressure on Red Hat.

Nissan emphasized that the compromised Red Hat environment held no additional Nissan-related data beyond the confirmed customer information from Fukuoka. The company reported no evidence indicating misuse of the leaked data to date.

This event represents the second cybersecurity incident for Nissan Japan in 2025. In late August, a Qilin ransomware attack targeted the company’s design subsidiary, Creative Box Inc. (CBI). In the prior year, Nissan North America experienced a data breach affecting 53,000 employees. Separately, Nissan Oceania disclosed an Akira ransomware attack that exposed data belonging to 100,000 customers.

Featured image credit

]]> First things we know about iPhone 18 https://dataconomy.com/2025/12/23/first-things-we-know-about-iphone-18/ Tue, 23 Dec 2025 10:04:19 +0000 https://dataconomy.com/?p=85601 A rumor from China states Apple will begin testing the iPhone 18 series on the mass-production line “one after another” in the very early days of January. Full production will start right before the 2026 Chinese New Year on February 17. The iPhone 18 Pro mass-production line has already been set up, as reported by […]]]>

A rumor from China states Apple will begin testing the iPhone 18 series on the mass-production line “one after another” in the very early days of January. Full production will start right before the 2026 Chinese New Year on February 17.

The iPhone 18 Pro mass-production line has already been set up, as reported by the source. This preparation positions the Pro model ahead in the manufacturing process for the series.

The iPhone 18 itself will not include exterior changes as extensive as many anticipate. The source specifies that design updates will fall short of widespread expectations for dramatic alterations.

Production timing precedes the presumed September launch by several months. The source notes no clear explanation exists for commencing manufacturing so early in the year. Details emerge from an unverified Chinese report, prompting caution in interpretation due to the absence of supporting rationale.

Rumored specs:

Feature iPhone 18 Pro / Pro Max iPhone 18 (Standard)
Launch Window September 2026 Spring 2027 (Anticipated)
Processor A20 Pro (TSMC 2nm) A20 (TSMC 2nm)
Front Display Under-Display Face ID Smaller Dynamic Island
Main Camera Variable Aperture (Mechanical) 48MP Fixed Aperture
RAM 12GB – 16GB LPDDR6 12GB LPDDR6
Connectivity Apple C2 5G Modem Apple C2 5G Modem

Featured image credit

]]> Why Apple is telling its visa employees to cancel holiday travel https://dataconomy.com/2025/12/23/why-apple-is-telling-its-visa-employees-to-cancel-holiday-travel/ Tue, 23 Dec 2025 10:01:16 +0000 https://dataconomy.com/?p=85598 Apple has instructed certain visa-holding employees to avoid leaving the United States due to unpredictable and extended delays in re-entering, as outlined in a memo from its law firm Fragomen issued last week. The memo, reported by Business Insider, targets employees lacking a valid H-1B visa stamp in their passports. It states verbatim: “Given the […]]]>

Apple has instructed certain visa-holding employees to avoid leaving the United States due to unpredictable and extended delays in re-entering, as outlined in a memo from its law firm Fragomen issued last week.

The memo, reported by Business Insider, targets employees lacking a valid H-1B visa stamp in their passports. It states verbatim: “Given the recent updates and the possibility of unpredictable, extended delays when returning to the U.S., we strongly recommend that employees without a valid H‑1B visa stamp avoid international travel for now.” The document continues: “If travel cannot be postponed, employees should connect with Apple Immigration and Fragomen in advance to discuss the risks.” This guidance emphasizes proactive consultation to assess individual circumstances before any necessary trips abroad.

Apple joins other major technology firms in issuing comparable travel advisories. Business Insider notes that Microsoft, Google, and ServiceNow have similarly cautioned their visa-dependent staff against international journeys amid the same processing hurdles. These companies rely heavily on H-1B visas to employ skilled foreign workers in specialized roles across engineering, research, and development.

In fiscal year 2024, Apple submitted 3,880 applications for H-1B visas, underscoring its substantial dependence on this visa category to maintain its workforce. H-1B visas enable U.S. employers to hire nonimmigrant workers in occupations requiring theoretical or technical expertise, with annual caps set by federal law.

Delays in visa processing, reaching up to 12 months, stem directly from a newly implemented social media screening requirement. This policy has led to widespread postponements of routine visa appointments at American embassies and consulates globally. Specific instances include facilities in Ireland and Vietnam, where scheduling backlogs have intensified since the rule’s introduction.

The screening mandate covers H-1B visa applicants along with their H-4 dependents, as well as students and exchange visitors. Individuals must disclose all social media usernames used over the past five years on the DS-160 visa application form, applicable to both new applications and renewals.

Beyond listing usernames, applicants are required to configure their social media profiles to public settings. This adjustment allows consular officers to examine online activity thoroughly during the adjudication process, forming a key component of the enhanced vetting procedures now standard for these visa categories.

Featured image credit

]]> Leaked: Samsung Galaxy A37 and A57 https://dataconomy.com/2025/12/23/leaked-samsung-galaxy-a37-and-a57/ Tue, 23 Dec 2025 09:58:51 +0000 https://dataconomy.com/?p=85595 Leaks from SmartPrix analysis reveal camera specifications for Samsung’s upcoming Galaxy A37 and Galaxy A57 mid-range smartphones, successors to the 2025 Galaxy A36 and Galaxy A56 models, with a projected February 2026 launch. Samsung released the Galaxy A36 and Galaxy A56 as its primary mid-range offerings in 2025. These devices set the benchmark for the […]]]>

Leaks from SmartPrix analysis reveal camera specifications for Samsung’s upcoming Galaxy A37 and Galaxy A57 mid-range smartphones, successors to the 2025 Galaxy A36 and Galaxy A56 models, with a projected February 2026 launch.

Samsung released the Galaxy A36 and Galaxy A56 as its primary mid-range offerings in 2025. These devices set the benchmark for the segment before early leaks emerged about their replacements. SmartPrix examined purported software details to uncover the camera configurations for the Galaxy A57 and Galaxy A37.

The Galaxy A57 features a Sony IMX906 main camera with 50 megapixels and a 1/1.56-inch sensor size. In certain regions, it may instead use a 50-megapixel ISOCELL GNJ main sensor, which matches the IMX906 in sensor dimensions. The setup includes a 13-megapixel ISOCELL S5K3L6 ultrawide camera and a 5-megapixel macro lens. Selfies and video calls rely on a 12-megapixel ISOCELL S5K3LC front-facing sensor.

Relative to the Galaxy A56, the A57’s camera array shows minimal changes. The predecessor employed a main camera nearly identical to the IMX906 or GNJ sensors. It paired this with a 12-megapixel ultrawide camera and the same 5-megapixel macro lens. Neither model incorporates a telephoto lens, consistent with Samsung’s approach in this category despite competitors like Nothing and realme introducing such options in mid-range devices.

The Galaxy A37 equips a Sony IMX906 main camera, an 8-megapixel Samsung GC08A3 ultrawide shooter, a 5-megapixel macro lens, and a 12-megapixel selfie camera. This main sensor measures larger than the Galaxy A36’s 1/1.96-inch unit. Larger sensors capture more light, producing brighter and cleaner images in low-light conditions. They also minimize motion blur because the shutter remains open for shorter durations during exposure.

The A37’s ultrawide, macro, and selfie cameras mirror those on the Galaxy A36 exactly. This configuration prioritizes enhancement of the primary imaging component while maintaining familiarity in auxiliary lenses.

Samsung plans to introduce these mid-range Galaxy phones in February 2026, advancing the timeline by one month from the March debuts of the prior generation. The Galaxy A37 integrates an Exynos 1480 processor, whereas the Galaxy A57 adopts an Exynos 1680 chip. These details stem from ongoing leaks reported by Android Authority.

Featured image credit

]]> Uber and Lyft to launch Baidu robotaxis in London in 2026 https://dataconomy.com/2025/12/23/uber-and-lyft-to-launch-baidu-robotaxis-in-london-in-2026/ Tue, 23 Dec 2025 09:55:23 +0000 https://dataconomy.com/?p=85593 Uber and Lyft will test Baidu’s Apollo Go robotaxis in London starting in 2026, pending local regulatory approval, under partnerships announced in July. The deployments join Waymo and local startup Wayve as the first autonomous-vehicle operators in the city next year. Lyft CEO David Risher stated in posts on X and LinkedIn that his company […]]]>

Uber and Lyft will test Baidu’s Apollo Go robotaxis in London starting in 2026, pending local regulatory approval, under partnerships announced in July. The deployments join Waymo and local startup Wayve as the first autonomous-vehicle operators in the city next year.

Lyft CEO David Risher stated in posts on X and LinkedIn that his company will initiate testing once it secures regulatory clearance. From that point, Lyft plans to scale to hundreds of Baidu’s electric RT6 SUVs. Risher provided no timeline for a commercial launch of the service.

Uber disclosed its intention to conduct tests in London next year through the same Baidu agreement established in July. The company specified that trials will commence in the first half of 2026.

Waymo operates robotaxis in other locations and now extends to London preparations, while Wayve, a UK-based startup, advances its autonomous technology locally. These London initiatives represent the most recent developments in robotaxi collaborations.

Uber and Lyft have formed partnerships with Baidu, Waymo, and additional firms. These tie-ups enable the companies to establish robotaxi operations across multiple cities worldwide.

Featured image credit

]]> Pew Research reveals significant racial gaps in teen AI chatbot usage https://dataconomy.com/2025/12/23/pew-research-reveals-significant-racial-gaps-in-teen-ai-chatbot-usage/ Tue, 23 Dec 2025 09:50:08 +0000 https://dataconomy.com/?p=85805 The Pew Research Center released a study detailing how young people use social media and artificial intelligence (AI) chatbots. Teen internet safety has remained a global concern. The U.S. surgeon general called for social media platforms to implement warning labels last year, and Australia will enforce a social media ban for individuals under 16 years […]]]>

The Pew Research Center released a study detailing how young people use social media and artificial intelligence (AI) chatbots.

Teen internet safety has remained a global concern. The U.S. surgeon general called for social media platforms to implement warning labels last year, and Australia will enforce a social media ban for individuals under 16 years old beginning Wednesday.

The Pew study found that 97% of teens use the internet daily. About 40% of respondents reported being “almost constantly online,” a decrease from 46% last year but significantly higher than the 24% recorded a decade ago. As AI chatbot prevalence increases in the U.S., this technology has become another factor in the internet’s impact on American youth.

About 3 in 10 U.S. teens use AI chatbots daily, with 4% reporting near-constant use. Fifty-nine percent of teens use ChatGPT, making it more than twice as popular as Google’s Gemini (23%) and Meta AI (20%). Forty-six percent of U.S. teens use AI chatbots at least several times a week, while 36% do not use them at all.

Pew’s research indicates that race, age, and socioeconomic class influence teen chatbot usage. Sixty-eight percent of Black and Hispanic teens surveyed reported using chatbots, compared to 58% of white respondents. Black teens were approximately twice as likely as white teens to use Gemini and Meta AI.

“The racial and ethnic differences in teen chatbot use were striking […] but it’s tough to speculate about the reasons behind those differences,” Pew Research Associate Michelle Faverio told TechCrunch. “This pattern is consistent with other racial and ethnic differences we’ve seen in teen technology use. Black and Hispanic teens are more likely than white teens to say they’re on certain social media sites — such as TikTok, YouTube, and Instagram.”

Across all internet use, 55% of Black teens and 52% of Hispanic teens reported being online “almost constantly,” roughly twice the rate of white teens (27%).

Older teens, aged 15 to 17, tend to use social media and AI chatbots more frequently than younger teens, aged 13 to 14. Sixty-two percent of teens from households earning more than $75,000 annually use ChatGPT, compared to 52% of teens below that income threshold. However, Character.AI usage is twice as popular (14%) in homes with incomes under $75,000.

While teenagers may initially use these tools for basic questions or academic assistance, their engagement with AI chatbots can become addictive and potentially harmful. The families of Adam Raine and Amaurie Lacey have filed lawsuits against OpenAI, the creator of ChatGPT, alleging its role in their children’s suicides. In both instances, ChatGPT allegedly provided detailed instructions on self-harm.

OpenAI stated it should not be held liable for Raine’s death, claiming the 16-year-old allegedly circumvented ChatGPT’s safety features and violated its terms of service. The company has not yet responded to the Lacey family’s complaint.

Character.AI, an AI role-playing platform, faces scrutiny concerning its impact on teen mental health. At least two teenagers died by suicide after extensive conversations with AI chatbots. The startup has since ceased offering its chatbots to minors, launching “Stories,” a choose-your-own-adventure-style product, for underage users.

The experiences cited in these lawsuits represent a small fraction of interactions on ChatGPT or Character.AI. Many chatbot conversations remain benign. OpenAI’s data indicates that only 0.15% of ChatGPT’s active users discuss suicide weekly. However, on a platform with 800 million weekly active users, this percentage translates to over one million people discussing suicide with the chatbot each week.

“Even if [AI companies’] tools weren’t designed for emotional support, people are using them in that way, and that means companies do have a responsibility to adjust their models to be solving for user well-being,” said Dr. Nina Vasan, a psychiatrist and director of Brainstorm: The Stanford Lab for Mental Health Innovation, to TechCrunch.

Featured image credit

]]> TSMC dominates foundry market with 72% share in Q3 2025 https://dataconomy.com/2025/12/23/tsmc-dominates-foundry-market-with-72-share-in-q3-2025/ Tue, 23 Dec 2025 09:44:35 +0000 https://dataconomy.com/?p=85614 Taiwan Semiconductor Manufacturing Company (TSMC) captured roughly 72% of the global semiconductor foundry market in the third quarter of 2025, as artificial intelligence chip demand drove record revenue, according to Counterpoint Research. Global semiconductor foundry revenue increased approximately 17% year-over-year to $84.8 billion in Q3 2025. This growth stemmed from strong demand for AI applications […]]]>

Taiwan Semiconductor Manufacturing Company (TSMC) captured roughly 72% of the global semiconductor foundry market in the third quarter of 2025, as artificial intelligence chip demand drove record revenue, according to Counterpoint Research.

Global semiconductor foundry revenue increased approximately 17% year-over-year to $84.8 billion in Q3 2025. This growth stemmed from strong demand for AI applications that rely on advanced manufacturing processes.

Manufacturer Market share Key technology focus
TSMC 72.0% 3nm / 5nm / CoWoS Packaging
Samsung 11.5% 2nm GAA / HBM4 Base Die
SMIC 5.7% 7nm / Domestic AI Infrastructure
UMC 2.3% 22nm / 28nm Specialty Nodes
GlobalFoundries 2.0% Automotive / Photonics

TSMC’s foundry revenue rose more than 40% annually during the quarter. This performance outpaced the broader market and expanded the company’s lead over competitors.

The revenue increase resulted from higher output of TSMC’s 3-nanometer manufacturing process. Utilization rates remained robust for 4-nanometer and 5-nanometer capacity. These processes serve AI accelerator customers such as Nvidia, Advanced Micro Devices, and Broadcom.

Advanced technologies with nodes below 7 nanometers represented nearly three-quarters of TSMC’s wafer revenue in Q3 2025. The 3-nanometer process accounted for 23% of that revenue. The 5-nanometer process contributed around 37%.

Rival foundries experienced slower growth. Non-TSMC players increased revenue by only about 6% collectively. This reflected softer orders following earlier demand linked to tariffs. China subsidy programs provided some support.

Samsung Electronics, TSMC’s closest competitor, maintained a mid-single-digit market share. Texas Instruments, Intel, and Infineon held similar positions. Analysts noted capacity limits at advanced nodes and constraints in advanced chip-packaging technology known as CoWoS. These factors may limit sequential growth in the fourth quarter of 2025.

Full-year 2025 foundry revenue growth reached an estimated 15%. The pure-play foundry segment grew at a faster pace. This occurred through continued shipments of AI GPUs and custom AI chips.

Featured image credit

]]> SpaceX rival ULA CEO Tory Bruno resigns after 12 years https://dataconomy.com/2025/12/23/spacex-rival-ula-ceo-tory-bruno-resigns-after-12-years/ Tue, 23 Dec 2025 09:42:28 +0000 https://dataconomy.com/?p=85590 Tory Bruno resigned as CEO of United Launch Alliance (ULA), a SpaceX rival, after 12 years to pursue another opportunity, the company announced. ULA chairs Robert Lightfoot and Kay Sears issued a statement thanking him. While Vulcan is now operational, ULA struggled to meet its “two-launches-per-month” target in late 2025, ending the year with approximately […]]]>

Tory Bruno resigned as CEO of United Launch Alliance (ULA), a SpaceX rival, after 12 years to pursue another opportunity, the company announced. ULA chairs Robert Lightfoot and Kay Sears issued a statement thanking him.

  • While Vulcan is now operational, ULA struggled to meet its “two-launches-per-month” target in late 2025, ending the year with approximately nine total missions.
  • SpaceX continues to dominate with high-frequency reusability, while Blue Origin’s New Glenn achieved successful inaugural flights in 2025, increasing market pressure.
  • Bruno’s departure reignites speculation that Boeing and Lockheed Martin may finally sell the joint venture, with Sierra Space and Blue Origin frequently cited as potential buyers.

United Launch Alliance formed 20 years ago through the merger of the space-launch divisions of Boeing and Lockheed Martin. The joint venture initially held the position of primary launch provider for NASA and the U.S. Department of Defense. This role persisted until SpaceX secured contracts, altering the competitive landscape in the launch sector.

Tory Bruno assumed the role of ULA CEO 12 years ago and guided the company through multiple transformations. Central to his tenure was the development of the Vulcan rocket program, designated as ULA’s next-generation launch vehicle. The Vulcan initiative pursued two primary objectives: maintaining competitiveness against SpaceX and diminishing the U.S. government’s dependence on Russian rockets for space access.

The Vulcan rocket incorporated components from ULA’s established Atlas and Delta programs to control costs. It relied on engines supplied by Blue Origin. Development encountered numerous delays, extending the timeline significantly. The first Vulcan launch took place in 2024, marking exactly one decade since the program’s inception.

Over this period, SpaceX advanced to become the world’s leading space launch provider. The company captured a substantial share of government contracts alongside a growing portfolio of private missions. Elon Musk’s SpaceX increased its launch frequency substantially in recent years. Concurrently, Jeff Bezos’ Blue Origin progressed with its New Glenn heavy-lift rocket, achieving mostly successful inaugural missions.

Despite these challenges, Vulcan secured commitments from key customers. Amazon selected Vulcan for deploying its Leo-Sat constellation of internet satellites. The space startup Astrobotic also contracted for launches with the rocket. ULA outlined intentions to enhance Vulcan by introducing reusability features or developing upgraded variants capable of transporting heavier payloads to space.

In a post on X, Bruno reflected on his leadership: “It has been a great privilege to lead ULA through its transformation and to bring Vulcan into service. My work here is now complete and I will be cheering ULA on.” Lightfoot and Sears responded in their statement: “We are grateful for Tory’s service to ULA and the country, and we thank him for his leadership.”

ULA designated its chief operating officer, John Elbon, to act as interim CEO. The company initiated a search process for a permanent successor to Bruno.

Featured image credit

]]> Alphabet acquires Intersect Power for $4.75B to solve AI energy gap https://dataconomy.com/2025/12/23/alphabet-acquires-intersect-power-for-4-75b-to-solve-ai-energy-gap/ Tue, 23 Dec 2025 09:40:28 +0000 https://dataconomy.com/?p=85587 Alphabet, Google’s parent company, has agreed to acquire Intersect Power, a data center and clean energy developer, for $4.75 billion in cash plus assumption of the company’s debt. Announced on Monday, the deal enables expansion of power-generation capacity for new data centers by bypassing local utilities strained by AI demand. The acquisition addresses the challenge […]]]>

Alphabet, Google’s parent company, has agreed to acquire Intersect Power, a data center and clean energy developer, for $4.75 billion in cash plus assumption of the company’s debt. Announced on Monday, the deal enables expansion of power-generation capacity for new data centers by bypassing local utilities strained by AI demand.

The acquisition addresses the challenge of securing reliable energy, which has emerged as essential for training artificial intelligence models that power data centers. Local utilities face difficulties meeting the surging electricity needs driven by AI operations.

Alphabet previously acquired a minority stake in Intersect Power through an $800 million strategic funding round led by Google and TPG Rise Climate in December. This earlier partnership established a goal of reaching $20 billion in total investments by 2030 to support large-scale clean energy and data infrastructure development.

Under the current agreement, Alphabet gains Intersect Power’s future development projects. Existing operations are excluded and will undergo a buyout by other investors, allowing them to operate independently as a separate entity.

Intersect Power’s new data parks consist of sites positioned adjacent to wind, solar, and battery power sources. Google indicated during the announcement of its minority investment that these facilities will become operational in late next year and reach full completion by 2027.

The transaction anticipates closure during the first half of next year. Google plans to serve as the primary user of these data parks. The campuses function as industrial parks capable of accommodating AI chips from other companies alongside Google’s infrastructure.

Featured image credit

]]> ChatGPT gets its own Spotify Wrapped: Find “Your Year with ChatGPT” https://dataconomy.com/2025/12/23/find-your-year-with-chatgpt-2025/ Tue, 23 Dec 2025 09:37:45 +0000 https://dataconomy.com/?p=85585 OpenAI is rolling out “Your Year with ChatGPT,” an annual review feature for its ChatGPT chatbot, to eligible consumers in the United States, Canada, the United Kingdom, Australia, and New Zealand. The feature draws from Spotify Wrapped and requires specific user settings and activity levels. The annual review targets users on free, Plus, and Pro […]]]>

OpenAI is rolling out “Your Year with ChatGPT,” an annual review feature for its ChatGPT chatbot, to eligible consumers in the United States, Canada, the United Kingdom, Australia, and New Zealand. The feature draws from Spotify Wrapped and requires specific user settings and activity levels.

The annual review targets users on free, Plus, and Pro plans. Access depends on enabling the “reference saved memories” and “reference chat history” options. Users must also reach a minimum conversation activity threshold, as stated by the company to TechCrunch via email. This setup ensures the feature activates only for accounts with sufficient interaction history and privacy preferences aligned.

Team, Enterprise, and Education accounts cannot access “Your Year with ChatGPT.” These organizational plans operate under different data handling and access protocols, excluding them from the consumer-oriented review process.

OpenAI characterizes the experience as lightweight, privacy-forward, and user-controlled, according to the company. This design prioritizes minimal data processing, respects user privacy settings, and allows individuals to initiate or manage the review at their discretion.

Like Spotify Wrapped, the feature employs catchy graphics tailored to individual usage patterns. It personalizes content by issuing “awards” that reflect specific ChatGPT interactions throughout the year. For example, the “Creative Debugger” award goes to users who employed the chatbot to generate solutions for problems or to explore and refine concepts and ideas.

In addition to awards, the review produces a custom poem capturing the user’s primary topics of interest over the year. It also generates a corresponding image that visually represents those same topics, providing a creative summary of conversational themes.

The year-end wrap-up appears as a promotion on the ChatGPT app’s home screen without forcing activation or automatic opening. Users encounter it optionally during navigation. Availability spans the ChatGPT web app as well as iOS and Android mobile apps, the company says. Alternatively, individuals can prompt ChatGPT directly with the phrase “Your Year with ChatGPT” to launch the experience immediately.

Featured image credit

]]> ChatGPT Atlas exploited with simple Google Docs tricks https://dataconomy.com/2025/12/23/chatgpt-atlas-exploited-with-simple-google-docs-tricks/ Tue, 23 Dec 2025 09:34:53 +0000 https://dataconomy.com/?p=85582 OpenAI launched its ChatGPT Atlas AI browser in October, prompting security researchers to demonstrate prompt injection vulnerabilities via Google Docs inputs that altered browser behavior, as the company detailed defenses in a Monday blog post while admitting such attacks persist. Prompt injection represents a type of attack that manipulates AI agents to follow malicious instructions, […]]]>

OpenAI launched its ChatGPT Atlas AI browser in October, prompting security researchers to demonstrate prompt injection vulnerabilities via Google Docs inputs that altered browser behavior, as the company detailed defenses in a Monday blog post while admitting such attacks persist.

Prompt injection represents a type of attack that manipulates AI agents to follow malicious instructions, often hidden in web pages or emails. OpenAI introduced ChatGPT Atlas during October, an AI-powered browser designed to operate with enhanced agent capabilities on the open web. On the launch day, security researchers published demonstrations revealing how entering a few words into Google Docs could modify the underlying browser’s behavior. These demos highlighted immediate security concerns with the new product, showing practical methods to exploit the system through indirect inputs.

Brave released a blog post on the same day as the launch, addressing indirect prompt injection as a systematic challenge affecting AI-powered browsers. The post specifically referenced Perplexity’s Comet alongside other similar tools, underscoring that this vulnerability extends across the sector rather than being isolated to OpenAI’s offering. Brave’s analysis framed the issue as inherent to the architecture of browsers integrating generative AI functionalities.

Feature Function / risk Mitigation strategy
Agent mode Autonomously scans emails and drafts replies. Human-in-the-loop: Requires confirmation for payments or sends.
Prompt injection Hidden text in websites/emails that overrides user intent. RL attacker: An AI bot that “pre-hacks” the browser to find flaws.
Data access High (Full access to logged-in sessions, inboxes). Limited permissions: Users are advised to give specific, narrow tasks.
Autonomy level Moderate (Performs multi-step workflows). Rapid patch cycle: Internal simulation of “long-horizon” attacks.

Earlier in the month, the U.K.’s National Cyber Security Centre issued a warning about prompt injection attacks targeting generative AI applications. The agency stated that such attacks “may never be totally mitigated,” which places websites at risk of data breaches. The centre directed cyber professionals to focus on reducing the risk and impact of these injections, rather than assuming attacks could be completely stopped. This guidance emphasized practical risk management over expectations of total elimination.

OpenAI’s Monday blog post outlined efforts to strengthen ChatGPT Atlas against cyberattacks. The company wrote, “Prompt injection, much like scams and social engineering on the web, is unlikely to ever be fully ‘solved.'” OpenAI further conceded that “agent mode” in ChatGPT Atlas “expands the security threat surface.” The post positioned prompt injection as an ongoing concern comparable to longstanding web threats. OpenAI declared, “We view prompt injection as a long-term AI security challenge, and we’ll need to continuously strengthen our defenses against it.”

Agent mode enables the browser’s AI to perform autonomous actions, such as interacting with emails or documents, which inherently increases exposure to external inputs that could contain hidden instructions. This mode differentiates Atlas from traditional browsers by granting the AI greater operational latitude on users’ behalf, thereby broadening potential entry points for manipulations.

To address this persistent risk, OpenAI implemented a proactive, rapid-response cycle aimed at identifying novel attack strategies internally before exploitation occurs in real-world scenarios. The company reported early promise from this approach in preempting threats. This method aligns with strategies from competitors like Anthropic and Google, who advocate for layered defenses and continuous stress-testing in agentic systems. Google’s recent efforts, for instance, incorporate architectural and policy-level controls tailored for such environments.

OpenAI distinguishes its approach through deployment of an LLM-based automated attacker, a bot trained via reinforcement learning to simulate hacker tactics. This bot searches for opportunities to insert malicious instructions into AI agents. It conducts tests within a simulation environment prior to any real-world application. The simulator replicates the target AI’s thought processes and subsequent actions upon encountering an attack, allowing the bot to analyze responses, refine its strategy, and iterate repeatedly.

This internal access to the AI’s reasoning provides OpenAI with an advantage unavailable to external attackers, enabling faster flaw detection. The technique mirrors common practices in AI safety testing, where specialized agents probe edge cases through rapid simulated trials. OpenAI noted that its reinforcement-learning-trained attacker can steer an agent into executing sophisticated, long-horizon harmful workflows that unfold over tens (or even hundreds) of steps. The company added, “We also observed novel attack strategies that did not appear in our human red-teaming campaign or external reports.”

In a specific demonstration featured in the blog post, the automated attacker inserted a malicious email into a user’s inbox. When Atlas’s agent mode scanned the inbox to draft an out-of-office reply, it instead adhered to the email’s concealed instructions and composed a resignation message. This example illustrated a multi-step deception spanning email processing and message generation, evading initial safeguards.

Following a security update to Atlas, the agent mode identified the prompt injection attempt during inbox scanning and flagged it directly to the user. This outcome demonstrated the effectiveness of the rapid-response measures in real-time threat mitigation, preventing the harmful action from proceeding.

OpenAI relies on large-scale testing combined with accelerated patch cycles to fortify systems against prompt injections before they manifest externally. These processes enable iterative improvements based on simulated discoveries, ensuring defenses evolve in tandem with potential threats.

Featured image credit

]]> ByteDance to invest $23B in AI to bridge US tech gap https://dataconomy.com/2025/12/23/bytedance-to-invest-23b-in-ai-to-bridge-us-tech-gap/ Tue, 23 Dec 2025 09:29:19 +0000 https://dataconomy.com/?p=85579 According to Financial Times, ByteDance, the Beijing-based parent of TikTok, plans to invest 160 billion yuan, approximately $23 billion, in artificial intelligence infrastructure in 2026 to narrow the gap with U.S. tech firms amid semiconductor restrictions. This marks an increase from 150 billion yuan allocated this year, with about half, or 85 billion yuan, dedicated […]]]>

According to Financial Times, ByteDance, the Beijing-based parent of TikTok, plans to invest 160 billion yuan, approximately $23 billion, in artificial intelligence infrastructure in 2026 to narrow the gap with U.S. tech firms amid semiconductor restrictions. This marks an increase from 150 billion yuan allocated this year, with about half, or 85 billion yuan, dedicated to advanced AI processors, according to the Financial Times.

The planned expenditure highlights the investment disparity between Chinese and American technology companies. Microsoft, Alphabet, Amazon, and Meta together project spending between $350 billion and $400 billion on AI infrastructure in 2025, which surpasses ByteDance’s 2026 allocation.

U.S. export controls have blocked Chinese companies from acquiring Nvidia’s most advanced chips. As a result, these firms have focused on creating AI models that operate with reduced computational demands.

In early December, President Donald Trump stated that the United States would authorize Nvidia to export its H200 processor to approved customers in China. The H200 ranks below Nvidia’s highest-end products in performance.

Nvidia intends to start shipping H200 chips to China by mid-February, with the first batch consisting of 40,000 to 80,000 units. ByteDance has expressed interest in purchasing 20,000 H200 units initially, at an estimated cost of $20,000 per unit.

ByteDance has achieved prominence in consumer AI despite hardware limitations. Its Doubao chatbot holds the position of China’s leading AI assistant, recording 157 million monthly active users in August according to analytics firm QuestMobile.

Doubao expanded to 159 million monthly active users by October, exceeding rivals including DeepSeek. This growth draws from ByteDance’s proficiency in developing widely adopted mobile applications and its close linkage with Douyin, the Chinese version of TikTok.

ByteDance’s daily token usage, which gauges engagement with AI services, surpassed 30 trillion tokens in October. By December, this figure climbed to 50 trillion tokens, nearing Google’s reported 43 trillion tokens, based on data from Goldman Sachs and ByteDance executives.

Featured image credit

]]> Xiaomi 17 Ultra debuts world’s first 200MP continuous zoom https://dataconomy.com/2025/12/23/xiaomi-17-ultra-debuts-worlds-first-200mp-continuous-zoom/ Tue, 23 Dec 2025 09:27:37 +0000 https://dataconomy.com/?p=85576 Xiaomi President Lu Weibing announced that the Xiaomi 17 Ultra will launch with the company’s most powerful telephoto camera, the first smartphone featuring Leica’s 200MP continuous optical-zoom lens that delivers 75mm to 100mm optical zoom while supporting a 200MP sensor across the entire range. The Xiaomi 17 Ultra introduces this lens as “the strongest telephoto […]]]>

Xiaomi President Lu Weibing announced that the Xiaomi 17 Ultra will launch with the company’s most powerful telephoto camera, the first smartphone featuring Leica’s 200MP continuous optical-zoom lens that delivers 75mm to 100mm optical zoom while supporting a 200MP sensor across the entire range.

The Xiaomi 17 Ultra introduces this lens as “the strongest telephoto imaging sensor in Xiaomi’s history,” according to Lu Weibing. This setup enables users to capture lossless high-resolution images without relying on digital crop or zoom features. The technology maintains full sensor resolution throughout the zoom process, preserving detail and quality in every shot.

Feature Details
Main Sensor 1-inch OmniVision Light Fusion 1050L (LOFIC HDR)
Telephoto Sensor 200MP Samsung S5KHPE (1/1.4-inch)
Optical Zoom Range 75mm to 100mm (Continuous Optical Zoom)
Portrait Focal Lengths 75mm, 85mm, 90mm, 100mm
Lens Structure 3G+5P Dual-Floating Lens (3 Glass + 5 Plastic)
Certifications Leica APO (Apochromatic)
Body Thickness 8.29mm (Slimmest Ultra model to date)

The zoom system employs a 3G+5P dual-floating lens layout, which combines three glass elements and five plastic elements to form the lens structure. This design incorporates separate sets of lenses dedicated to optical performance, zoom adjustment, and focus control. These components work together to produce perfect images with reduced aggression, minimizing distortions and aberrations in various shooting scenarios.

Additionally, the lens covers four golden portrait focal lengths within a single module, allowing for iconic macro shots. This integration supports versatile photography, from close-up details to portrait compositions, all handled by the same advanced telephoto unit.

The Xiaomi 17 Ultra holds the distinction of being the first device to receive Leica APO (Apochromatic) optical certification. This certification ensures an extremely high standard in chromatic aberration control. As a result, the camera delivers enhanced image sharpness alongside precise color accuracy, addressing common issues in high-resolution telephoto photography.

Prior rumors indicated that the Xiaomi 17 Ultra would place significant emphasis on its imaging capabilities. Xiaomi has confirmed these predictions through Lu Weibing’s announcement and the detailed camera specifications. Further details on other changes and features will appear at the official launch event.

Featured image credit

]]> HTC launches VIVE Eagle AI glasses with open platform https://dataconomy.com/2025/12/23/htc-launches-vive-eagle-ai-glasses-with-open-platform/ Tue, 23 Dec 2025 09:25:47 +0000 https://dataconomy.com/?p=85573 Taiwan’s HTC launched its VIVE Eagle AI smart glasses in Hong Kong in August 2025 to capture market share in the smartglasses sector through an open-platform strategy that lets users select AI models such as Google’s Gemini and OpenAI. HTC positions the VIVE Eagle as an open device amid rapid AI advancements. Charles Huang, senior […]]]>

Taiwan’s HTC launched its VIVE Eagle AI smart glasses in Hong Kong in August 2025 to capture market share in the smartglasses sector through an open-platform strategy that lets users select AI models such as Google’s Gemini and OpenAI.

HTC positions the VIVE Eagle as an open device amid rapid AI advancements. Charles Huang, senior vice president of global sales and marketing at HTC, explained in a Reuters interview the rationale behind this approach. “AI is advancing very fast, and large language model developers are engaged in an arms race that requires massive resources,” Huang stated. He added, “We want to leverage the strengths of different platforms instead of building a closed ecosystem.” This design enables users to access improvements from various AI models without restriction to a single provider.

The VIVE Eagle smart glasses support multiple AI platforms, including Google’s Gemini and OpenAI. In comparison, Meta’s smart glasses rely on Meta AI, while Chinese brands such as Xiaomi and Alibaba integrate domestically developed AI models. HTC priced the VIVE Eagle at HK$3,988 ($512) for its Hong Kong debut earlier in August 2025. The glasses were displayed at HTC’s headquarters in Xindian, New Taipei City, Taiwan, on December 17, 2025.

HTC outlined specific expansion timelines for the VIVE Eagle. Sales will extend to Japan and Southeast Asia during the first quarter of 2026. Further rollout targets Europe and the United States later in 2026. This sequence prioritizes Asia before broader global markets.

The Asia-first rollout stems from design adaptations for regional users. Huang noted that many existing smart glasses accommodate a “Western fit” that does not align well with Asian wearers. HTC tailored the VIVE Eagle to address this mismatch, enhancing comfort and suitability for Asian markets.

Regarding potential entry into China, Huang highlighted regulatory hurdles. Foreign AI services face restrictions there, and local data regulations mandate standalone servers within the country. “With all these requirements in place, we need to be cautious and it will take some time to prepare,” Huang said. HTC approaches this market with deliberate preparation to comply fully.

Global smartglasses shipments rose 110% in the first half of 2025, according to research firm Counterpoint. Meta captured 73% of that market share during this period. The surge reflects growing demand for wearable AI-integrated eyewear.

Meta, in partnership with EssilorLuxottica, introduced its smart Ray-Bans and Oakleys in 2023. These devices handle functions such as answering calls, capturing pictures, and playing music, drawing widespread attention in the technology sector.

Analysts point to privacy as an emerging issue in the smartglasses market. Meta, which operates Facebook, Instagram, and WhatsApp, employs user data to develop its AI tools, prompting scrutiny over its data handling practices. HTC differentiates itself on this front. Huang stated that HTC does not use user data to train its AI models. The company views privacy and data security as primary distinctions from competitors.

The VIVE Eagle launch represents HTC’s renewed focus on consumer hardware. Earlier in 2025, HTC sold part of its extended reality headset and glasses unit to Google for $250 million. This transaction preceded the smart glasses initiative, signaling HTC’s strategic pivot back into direct consumer products.

Featured image credit

]]> Nvidia prepares first H200 shipments to China by February https://dataconomy.com/2025/12/23/nvidia-prepares-first-h200-shipments-to-china-by-february/ Tue, 23 Dec 2025 09:21:00 +0000 https://dataconomy.com/?p=85571 According to Reuters, Nvidia has informed Chinese clients of plans to ship H200 AI chips to China before the mid-February Lunar New Year holiday, using existing stock for initial orders totaling 5,000 to 10,000 chip modules, equivalent to 40,000 to 80,000 chips, pending Beijing approval. Three people familiar with the matter disclosed these details to […]]]>

According to Reuters, Nvidia has informed Chinese clients of plans to ship H200 AI chips to China before the mid-February Lunar New Year holiday, using existing stock for initial orders totaling 5,000 to 10,000 chip modules, equivalent to 40,000 to 80,000 chips, pending Beijing approval.

Three people familiar with the matter disclosed these details to Reuters. The first and second sources specified the shipment volumes from current inventory. This initiative follows U.S. President Donald Trump’s announcement allowing such sales with a 25 percent fee, marking the first potential deliveries of these chips to China.

Nvidia additionally communicated plans to expand production capacity for the H200. According to the third source, orders for this new capacity will open in the second quarter of 2026. These steps aim to meet anticipated demand from authorized Chinese customers.

Significant uncertainties surround the timeline and execution. Beijing has not approved any H200 purchases to date. Government decisions could alter schedules or volumes. The third source emphasized, “Nothing is certain until we get the official go‑ahead.”

In response to Reuters inquiries, Nvidia issued a statement affirming its supply chain management practices. The company declared, “we continuously manage our supply chain. Licensed sales of the H200 to authorised customers in China will have no impact on our ability to supply customers in the United States.” China’s Ministry of Industry and Information Technology did not immediately respond to a request for comment.

The planned shipments align with recent U.S. policy developments. Reuters reported last week that the Trump administration initiated an inter-agency review of license applications for H200 sales to China. This action fulfills Trump’s pledge and contrasts with the Biden administration’s ban on advanced AI chip sales to China, imposed due to national security concerns.

The H200 belongs to Nvidia’s previous-generation Hopper architecture. It continues to see widespread use in artificial intelligence applications. Production has shifted toward the newer Blackwell chips and the upcoming Rubin line, resulting in limited H200 availability globally.

Trump’s policy adjustment occurs as China accelerates development of its domestic AI chip sector. Local companies have not yet produced processors matching the H200’s capabilities. Officials express concerns that importing these chips might hinder progress in building indigenous technology.

Earlier this month, Chinese officials convened emergency meetings to address the potential influx of H200 chips. Discussions focused on approval conditions for shipments. Reuters reported one proposal under consideration: requiring each H200 purchase to include a specified ratio of domestically produced chips.

Prominent Chinese technology firms have shown strong interest in acquiring H200 units. Alibaba Group, listed as 9988.HK, and ByteDance number among those clients. For these companies, the H200 offers processors approximately six times more powerful than the H20, a version Nvidia specifically adapted and downgraded for the Chinese market.

Featured image credit

]]> Paramount sweetens hostile Warner Bros. bid after rejection https://dataconomy.com/2025/12/23/paramount-sweetens-hostile-warner-bros-bid-after-rejection/ Tue, 23 Dec 2025 09:18:43 +0000 https://dataconomy.com/?p=85568 Paramount Skydance amended its $108 billion tender offer for Warner Bros. Discovery, incorporating Larry Ellison’s irrevocable personal guarantee of $40.4 billion, after Warner Bros. Discovery rejected the initial bid in favor of an $82.7 billion Netflix merger. David Ellison, CEO of Paramount Skydance and son of Oracle founder Larry Ellison, announced the initial $30-per-share fully […]]]>

Paramount Skydance amended its $108 billion tender offer for Warner Bros. Discovery, incorporating Larry Ellison’s irrevocable personal guarantee of $40.4 billion, after Warner Bros. Discovery rejected the initial bid in favor of an $82.7 billion Netflix merger.

David Ellison, CEO of Paramount Skydance and son of Oracle founder Larry Ellison, announced the initial $30-per-share fully cash offer on December 4, 2025. He positioned it as the superior proposal for Warner Bros. Discovery shareholders. Paramount Skydance formally launched the tender offer on December 8, 2025. The bid drew backing from sovereign wealth funds in Saudi Arabia and Qatar. The Ellison family pledged to backstop the full amount if those funders withdrew.

Warner Bros. Discovery’s board announced a merger agreement with Netflix on December 5, 2025. The company later accepted Netflix’s $82.7 billion offer, which requires regulatory approval and expects closure sometime next year. Warner Bros. Discovery formally recommended that shareholders reject Paramount Skydance’s tender offer.

Warner Bros. Discovery stated, “[The board] has unanimously determined that the tender offer launched by Paramount Skydance on December 8, 2025 is not in the best interests of WBD and its shareholders and does not meet the criteria of a ‘Superior Proposal’ under the terms of WBD’s merger agreement with Netflix announced on December 5, 2025.” The company viewed the Ellisons’ backstop pledge as insufficient to address financing risks.

In response, Paramount Skydance amended its offer to include Larry Ellison’s irrevocable personal guarantee covering $40.4 billion. This guarantee addresses the financing shortfall if sovereign wealth funds withdraw. The amendment also commits the senior Ellison not to revoke or adversely transfer assets in the Ellison family trust during the pendency of the transaction. Warner Bros. Discovery had indicated that such a personal guarantee represented the sole remedy for the original offer’s inadequacies.

Paramount Skydance noted in its updated proposal that none of these concerns, including the demand for a personal guarantee, were raised by Warner Bros. Discovery or its advisors during the 12-week period preceding the agreement with Netflix. The company described the Netflix transaction as inferior.

David Ellison stated, “Our $30 per share, fully financed all‑cash offer was on December 4th, and continues to be, the superior option to maximize value for WBD shareholders. Because of our commitment to investment and growth, our acquisition will be superior for all WBD stakeholders, as a catalyst for greater content production, greater theatrical output, and more consumer choice.” He added, “We expect the board of directors of WBD to take the necessary steps to secure this value‑enhancing transaction and preserve and strengthen an iconic Hollywood treasure for the future.”

The revised offer includes additional measures: publication of the Ellison family trust’s assets, more flexible transaction terms, and an increase in the regulatory reverse termination fee from $5 billion to $5.8 billion, matching Netflix’s fee. The Paramount Skydance offer expires on January 21, 2026.

Featured image credit

]]> Simple ways to prepare your home for wildfire risks https://dataconomy.com/2025/12/22/simple-ways-to-prepare-your-home-for-wildfire-risks/ Mon, 22 Dec 2025 15:38:55 +0000 https://dataconomy.com/?p=85556 Every year, countless communities face the imminent danger of wildfires. Prepping a home against those hazards can provide safety for loved ones and property. By taking a few simple, low-impact precautions, you can ease some risk and sleep soundly at night when the fire season rolls around. A smooth start is great, but they’re rare […]]]>

Every year, countless communities face the imminent danger of wildfires. Prepping a home against those hazards can provide safety for loved ones and property. By taking a few simple, low-impact precautions, you can ease some risk and sleep soundly at night when the fire season rolls around. A smooth start is great, but they’re rare these days, so it’s helpful to take precautions when you are uncertain.

Create a defensible space

Create defensible spaces to prepare your home for wildfire scenarios. It entails removing vegetation, leaves, and other combustibles from at least thirty feet around buildings. It is also useful to prune tree branches so they do not hang over any structures. Tall grass and dead plants can catch fire quickly, so keeping them pruned back can minimize fire propagation. Plants and ground covers that are resistant to fire provide an additional layer of protection to yards and gardens.

Maintain roofs and gutters

Dry debris builds up on roofs and gutters and can act as kindling if a wildfire approaches. Keeping these areas clean ensures embers will not ignite fires on the property. Selecting fire-resistant roof materials, including metal or tile, minimizes exposure. Year-round inspections of roofs and gutters prevent flammable debris from accumulating. Using screens or putting guards over gutters will also prevent leaves from building up.

Seal openings and vents

Firebrands are embers and other burning materials that enter a building through small openings, cracks, and crevices. Sealing these openings with fire-resistant materials prevents sparks from entering the inside. Mesh screens over vents can help defend against embers carried by the wind. Windows and doors must close tightly with no gaps. Weatherstripping is also a painless way to make them more smoke- and heat-resistant.

Prepare emergency supplies

All homes should prepare an emergency supply kit ahead of the start of fire season. A few items you will want to have on hand are water, non-perishable food, important documents, and a first aid kit. Provide flashlights, batteries, and dust masks to assist in smoky situations. Packing them into a convenient bag will allow you to get out in a short time if necessary. Using copies of important documents is necessary to keep them in one place to prevent any loss.

Plan safe evacuation routes

Knowing several safe ways to exit the area can save lives during an emergency. Families should map out at least two exit routes from their neighborhood. Rehearsing these paths helps everyone respond calmly if evacuation becomes necessary. Communication plans, including out-of-area contacts, allow family members to stay in touch if separated. Marking routes and sharing them with neighbors builds community preparedness.

Protect outdoor structures

Reducing wildfire threats should include things like sheds, fences, and decks as well. Making them out of fireproof materials enhances their ability to withstand heat and flames. Clearing debris from under decks and around sheds removes fuel. Keeping combustible materials, like propane tanks or stacks of firewood, away from structures minimizes the threat.

Stay informed and alert

Staying updated on local fire conditions is helpful during dangerous times, such as when fire situations are threatening close localities. Community alerts ensure residents receive notifications about threats and evacuations. You can count on updates from weather radios, even if the power goes out. People can take action quickly if they see smoke, if the wind changes, or when the authorities issue warnings. Understanding enables rapid life and property safeguarding decisions.

Work together as a community

Systemic security relies on the members of the neighborhood who function and work together. Neighbors enhance readiness by sharing information and resources. They organize community clean-up days to clear larger areas and make them safer. Taking care of any vulnerable neighbors, seniors, maybe, or people with disabilities will help look after the neighborhood as a collective. The more people who join in, the safer everyone is.

Conclusion

When preparing for a wildfire threat, the simplest steps make the biggest difference. You can increase your safety margin by clearing defensible spaces, treating the roof and sealing openings, and planning. Having supplies for emergencies prepared and knowing what is happening locally means that if there’s a fire in your town, you will be ready to go. Together, people and communities can rise to the challenge of wildfire risks with confidence and compassion. Such measures will safeguard homes and those who inhabit them.

Featured image credit

]]> Why Machine Learning is Still Broken for ‘Black Swan’ Risk Management https://dataconomy.com/2025/12/22/machine-learning-broken-black-swan/ Mon, 22 Dec 2025 13:43:04 +0000 https://dataconomy.com/?p=85550 On October 10th, 2025, the cryptocurrency markets experienced a seismic dislocation. In a matter of minutes, a liquidation cascade wiped out billions in open interest, leaving standard trading algorithms paralyzed. It wasn’t just a price drop; it was a structural failure of predictive models. Strategies that had printed money for months suddenly faced a market […]]]>

On October 10th, 2025, the cryptocurrency markets experienced a seismic dislocation. In a matter of minutes, a liquidation cascade wiped out billions in open interest, leaving standard trading algorithms paralyzed. It wasn’t just a price drop; it was a structural failure of predictive models. Strategies that had printed money for months suddenly faced a market state that did not exist in their training data.

This event served as a brutal reminder: in the high-stakes world of quantitative finance, the reliance on Machine Learning (ML) has become absolute, yet its blind spots remain fatal. From High-Frequency Trading (HFT) algorithms executing in nanoseconds to complex DeFi oracles, the industry is in an arms race of data supremacy. But when a “Black Swan” hits, models trained on historical data don’t just underperform – they break.

This creates a paradox for modern trading firms: how do you build resilient systems when your primary tools are blind to the most significant risks?

To answer this, we sat down with Grigory Chikishev, a Team Lead and Quantitative Trader at Quantum Brains. With over nine years of experience building infrastructure solutions for markets – ranging from HFT algorithms and ML models to graph-based flow evaluation systems – Grigory has spent his career at the intersection of execution speed and systemic resilience. At Quantum Brains, he has transformed market processes into scalable architectures designed to withstand the very volatility that breaks standard models.

Here is his perspective on why the industry needs to move beyond the “black box” and how to engineer true antifragility.

The Zen of the Unpredictable

When the discussion turns to the failure of risk models during events like the recent October crash, the COVID-19 pandemic, or the 2008 financial crisis, the standard critique is that the models “failed” to predict the event. Grigory challenges this premise entirely. He argues that the expectation that an ML model will predict a singularity is mathematically flawed, and that the solution lies not in better prediction but in better acceptance.

“I’d like to point out right away that I don’t see a problem with the existence of black swans. They are, by definition, events that are impossible to predict. And there’s nothing we can do about it. For example, a comet colliding with Earth: we can almost certainly say it won’t happen in the coming weeks or even years, but no one knows what’s going on in the unseen part of the galaxy…

The word ‘fail’ may be an exaggeration. If we know in advance of our inability to predict event A, then we should accept its occurrence with Buddhist calmness.”

However, accepting unpredictability does not mean ignoring consequences. Grigory points out that while a model cannot predict the timing of a crisis, human domain experts must architect systems that understand the consequences of the worst-case scenario – something purely data-driven models often miss because the data points simply aren’t there.

“Somewhere between these two numbers lies the critical point that separates a predictable event from an unpredictable one (a black swan). And the fundamental flaw of any model is that it can’t calculate this point… We can only prepare for the worst-case scenario, which the model DOESN’T account for.”

The Myth of the Transparency Trade-Off

A significant debate in quantitative finance is the tension between Explainable AI (XAI) and profit. The prevailing wisdom suggests that “Black Box” models (unsupervised deep learning models that are difficult to interpret) are more profitable because they are more complex, and that forcing them to be explainable (for regulatory compliance) slows execution and blunts their edge.

Grigory vehemently disagrees with this dichotomy. For him, transparency is not a regulatory burden; it is a debugging tool.

“I highly doubt that an unsupervised or black box approach will ultimately be more successful than a white box approach when directly compared… Therefore, any efforts toward ‘regulatory-level interpretability’ are only for the better. If your newborn child could explain what hurts, it would be very convenient and would clearly help with their upbringing.”

He suggests that opacity in trading strategies is often a mask for luck rather than genius – specifically, survivorship bias.

“If you see a successful ML strategy that ‘is unclear how it works,’ then one of two things is most likely true:

  1. Either its creators actually understand everything, but prefer to keep their cards close to their chest.
  2. Or we’re dealing with survivorship bias… If 1,024 people make a chain of 10 binary predictions, precisely one of them will be absolutely correct in each prediction.

Unfortunately, sometimes both reasons are correct. So always demand an explanation from your AI agent!”

Engineering Antifragility

If prediction is impossible, the only viable strategy is antifragility – the ability of a system to gain from disorder, a concept popularized by Nassim Taleb. However, implementing this in hardware and infrastructure is notoriously difficult. Building a system that can handle 100x the normal market load during a crash is often cost-prohibitive.

Grigory’s approach to infrastructure at Quantum Brains prioritizes flexibility over brute force capacity.

“You can’t prepare your infrastructure for a black swan event. For example, if you calculate your server’s peak load and allow for a 100x increase, then you’re burning money on unused resources almost 100% of the time… But you can prepare a flexible system to reduce resource costs. For example, simply shutting down one trading setup after another. What’s the point anyway if everything goes to hell?”

This flexibility allows a firm to survive the initial shock. But to actually profit from the dislocation – to be truly antifragile – requires a shift in mindset. It requires recognizing that when others’ algorithms fail, the market is no longer efficient.

“I repeat, we’re talking about a situation that our models didn’t predict… This formulation also contains some good news: we can assume that other market participants are experiencing the same ‘difficult’ scenario. On October 10th, cryptocurrencies experienced a significant shock, prompting many positions to be liquidated. Some participants literally left the market: either they chose the second option (shutdown) or simply didn’t have time to do so (RIP).

This was a good moment to exploit inefficiencies or realize opportunities that would usually be closed… In a sense, this is also Taleb’s way: to avoid being a turkey, you simply have not to be one.”

The Human Element in a Zero-Sum Game

As AI continues to dominate trade execution, many question the future role of the human quantitative trader. If machines handle the flow, the risk, and the execution, is the human obsolete?

Grigory believes the very nature of the market safeguards the human element: it is a zero-sum game driven by the desire to win, an emotion that algorithms do not possess. While AI can execute, it lacks the drive to “beat” the market that fuels true innovation.

“Trading differs from many other fields where AI is actively developing, because it’s a zero-sum game… Let’s imagine an extreme: there are no living participants left in the market… Is there a place for humans here? In my opinion, there isn’t.

But fortunately… in the real world, there will always be living participants… Another human factor is overconfidence. The idea, ‘I’m human, I’ll be more inventive and original than AI,’ will never leave our minds.”

Ultimately, the future of quantitative trading isn’t about replacing humans with AI, but about humans using AI to compete against other humans. The algorithm is the weapon, not the soldier.

“As I said, it’s a zero-sum game. But an algorithm has no interest in making money in such conditions. Only homo sapiens will always have the desire to ‘beat’ others.”

]]> The Top 5 RevOps Platforms: Truth Before Automation https://dataconomy.com/2025/12/22/top-5-revops-platforms-truth-before-automation/ Mon, 22 Dec 2025 11:14:18 +0000 https://dataconomy.com/?p=85544 The Revenue Operations (RevOps) mandate is shifting constantly. We are no longer just looking for “more data” or “faster workflows.” The priority has moved to precision. Projected to reach nearly $17 billion by 2033, the RevOps market is defined by the strategic convergence of sales, marketing, and customer success into a unified, data-driven framework that […]]]>

The Revenue Operations (RevOps) mandate is shifting constantly. We are no longer just looking for “more data” or “faster workflows.” The priority has moved to precision.

Projected to reach nearly $17 billion by 2033, the RevOps market is defined by the strategic convergence of sales, marketing, and customer success into a unified, data-driven framework that maximizes operational efficiency and revenue predictability.

Sure, the market is crowded, but the difference between a bloated tech stack and a streamlined revenue engine comes down to one core philosophy: Truth before Automation. You cannot build a high-velocity GTM motion on a foundation of decaying or non-compliant data.

Based on performance, compliance, and strategic value, here are the top 5 RevOps platforms defining the landscape this year.

#1  –  Lusha: The Verified Data Layer for Modern RevOps

The Verdict: The new standard for accuracy-led growth.

Lusha takes the top spot because it solves the single most significant friction point in Revenue Operations: reliability. While most platforms automate on top of inconsistent data, Lusha begins with a verified data foundation, so workflows don’t collapse downstream.

It operates on a simple yet powerful premise: verified data gives you the truth. Automation gives you speed. Signals provide you with timing. When you combine them, your entire GTM motion streams.

Why Lusha leads the category:

  • Truth-First Foundation: Backed by GDPR, CCPA, ISO 27001, and 27701 compliance, and powered by verified, always-fresh data, Lusha ensures workflows don’t break due to data decay.
  • Revenue Streaming: Lusha replaces static lists with real-time signals – funding rounds, job changes, tech shifts – so RevOps teams can trigger timely, high-context engagement.
  • AI Without the Hallucinations: Its AI automation is built on a verified data layer, enriching, scoring, and routing leads instantly without the guesswork.
  • The Backbone, Not a Bolt-On: Through deep CRM and stack integrations, Lusha functions as the backbone of RevOps. It’s keeping records enriched and up to date while accelerating time-to-lead.

#2  –  ZoomInfo: Enterprise Scale with a Heavy Ops Footprint

The Verdict: Massive volume, but accuracy remains the trade-off.

ZoomInfo remains a titan in the space, particularly for large enterprises that need sheer volume and a sprawling ABM ecosystem. Its breadth of data is undeniable.

Industry analysts note that ZoomInfo’s scale remains both its strength and its vulnerability. The platform’s vast dataset powers an impressive ecosystem, but maintaining precision at that scale is an ongoing challenge, as reflected in ongoing conversations across RevOps forums about data freshness and accuracy. As the market shifts toward signal-driven, real-time intelligence, the trade-off between coverage and confidence has become increasingly visible.

However, the “quantity over quality” debate persists. Public discourse continues to highlight recurring issues with “bad data” and relevance gaps. For RevOps teams, this often requires a significant operational lift to verify data before it enters critical workflows. It is a powerful tool, but one that requires considerable maintenance.

#3  –  Clay: Workflow Customization for GTM Engineers

The Verdict: Infinite flexibility for those with the technical chops.

Clay has carved out a niche for the “GTM Engineer.” If your RevOps team loves to build custom scripts and highly intricate waterfalls, Clay is a playground of possibilities.

The trade-off is focused on plug-and-play scalability. Clay’s value lies in customization, but that requires a heavy investment in setup and ongoing maintenance. It is less of a “platform” and more of a “toolkit” – ideal for technical builders, but potentially a bottleneck for agile sales teams needing immediate deployment.

#4  –  6sense: Predictive Engagement for Mature GTM Teams

The Verdict: A powerhouse for established ABM infrastructures.

6sense continues to excel in intent scoring and predictive account engagement. For organizations with a mature, sophisticated ABM strategy already in place, 6sense provides the “Dark Funnel” visibility needed to capture demand early.

However, its complexity and cost structure make it a difficult entry point for leaner GTM organizations. It is a specialized weapon for specific battles, rather than an all-encompassing RevOps foundation.

#5  –  Gong: Conversation Intelligence That Scales Coaching

The Verdict: The leader in insights, but complementary to the stack.

Gong remains the undisputed king of conversation intelligence. For diagnosing performance patterns and scaling coaching, it is essential.

From a strict RevOps perspective, however, Gong is a diagnostic tool rather than a foundational one. It helps you understand why deals are won or lost, but it does not address upstream challenges such as data accuracy, routing, or lead enrichment. It is the perfect complement to a verified data layer, not a replacement for one.

There’s a Clear Winner in RevOps 

While ZoomInfo offers volume, Clay offers customization, 6sense offers prediction, and Gong offers insight, Lusha is the only platform that secures the foundation: Truth.

By prioritizing verified data before layering on automation, Lusha ensures that every downstream action – from routing to outreach – actually drives revenue.

]]> Samsung HBM4 outshines rivals in Nvidia’s Vera Rubin tests https://dataconomy.com/2025/12/22/samsung-hbm4-outshines-rivals-in-nvidias-vera-rubin-tests/ Mon, 22 Dec 2025 10:06:17 +0000 https://dataconomy.com/?p=85523 Samsung Electronics has achieved a significant breakthrough in the high-bandwidth memory market, with its next-generation HBM4 memory outperforming competitors in speed and power efficiency tests for Nvidia’s upcoming Vera Rubin AI accelerator, according to Maeil Business Newspaper. During a visit last week, Nvidia representatives confirmed that Samsung delivered “the best results in the memory industry,” […]]]>

Samsung Electronics has achieved a significant breakthrough in the high-bandwidth memory market, with its next-generation HBM4 memory outperforming competitors in speed and power efficiency tests for Nvidia’s upcoming Vera Rubin AI accelerator, according to Maeil Business Newspaper. During a visit last week, Nvidia representatives confirmed that Samsung delivered “the best results in the memory industry,” prompting a request for supply volumes that far exceeded Samsung’s internal projections. This success marks a dramatic reversal from the HBM3E generation, where Samsung lagged behind rivals by nearly a year.

The turnaround is attributed to a high-risk technical strategy in which Samsung skipped the D1b DRAM process entirely to advance directly to the 10-nanometer D1c process. By combining this new DRAM with logic dies produced using a 4-nanometer foundry process, Samsung became the first manufacturer to achieve data transfer speeds exceeding 11 gigabits per second. While SK hynix maintains a roughly three-month lead—having already started supplying paid samples—Samsung has successfully narrowed the gap from the previous generation.

Market data reflects this recovery, with Samsung reclaiming the number two spot in the HBM market during the third quarter of 2025 with a 22% share, overtaking Micron. Contracts are expected to be formalized in the first quarter of 2026, with full-scale deliveries scheduled for the second quarter to meet Nvidia’s timeline for the Vera Rubin platform launch in Q3 2026.

Featured image credit

]]> TikTok accused of sending European user data to China again https://dataconomy.com/2025/12/22/tiktok-accused-of-sending-european-user-data-to-china-again/ Mon, 22 Dec 2025 10:02:48 +0000 https://dataconomy.com/?p=85521 European regulators have intensified scrutiny of TikTok, accusing the Chinese-owned platform of continuing to transfer European user data to China despite previous sanctions. The Norwegian Data Protection Authority stated that users were recently notified that their personal information remains accessible to TikTok employees in China, a practice regulators fear could expose data to Chinese government […]]]>

European regulators have intensified scrutiny of TikTok, accusing the Chinese-owned platform of continuing to transfer European user data to China despite previous sanctions. The Norwegian Data Protection Authority stated that users were recently notified that their personal information remains accessible to TikTok employees in China, a practice regulators fear could expose data to Chinese government surveillance under local laws. This aligns with warnings from the Dutch Data Protection Authority, which emphasized that users—particularly younger ones—remain insufficiently aware that their data is still being routed to China while TikTok appeals earlier regulatory rulings.
+2

In a separate legal challenge, the Austrian digital rights group noyb filed complaints alleging that TikTok illegally tracks users across third-party applications without consent. The group claims TikTok utilized the Israeli mobile analytics firm AppsFlyer to harvest sensitive data, including information from the dating app Grindr, which reveals sexual orientation—a category protected under Article 9 of the GDPR.

Additionally, noyb accused TikTok of violating data access rights by refusing to provide users with their complete data files, instead offering only “relevant” subsets. These developments follow a €530 million fine imposed by Ireland’s Data Protection Commission in May for similar data transfer violations.

Featured image credit

]]> Why Light of Motiram vanished from Steam and Epic stores https://dataconomy.com/2025/12/22/why-light-of-motiram-vanished-from-steam-and-epic-stores/ Mon, 22 Dec 2025 10:00:26 +0000 https://dataconomy.com/?p=85519 Sony and Tencent have reached a confidential settlement in their legal dispute over the game Light of Motiram, which Sony had alleged was a “slavish clone” of its Horizon series. A court document filed Wednesday confirmed the case dismissal with prejudice. The game Light of Motiram has been delisted from both Steam and the Epic […]]]>

Sony and Tencent have reached a confidential settlement in their legal dispute over the game Light of Motiram, which Sony had alleged was a “slavish clone” of its Horizon series. A court document filed Wednesday confirmed the case dismissal with prejudice.

The game Light of Motiram has been delisted from both Steam and the Epic Games Store. Its website previously showed links to these storefronts, and a SteamDB entry indicates the application has been retired according to a Reddit user.

Sony initiated the lawsuit in July to block the release of Light of Motiram. The company contended that Tencent utilized a character resembling Horizon’s protagonist, Aloy, as a central element in its marketing efforts.

In its lawsuit, Sony stated that “Tencent’s unlawful copying of the protected audiovisual elements of the Horizon games, as well as its deliberate adoption of a confusingly similar character mark, constitutes both copyright and trademark infringement that should be enjoined immediately to prevent irreparable harm to [Sony Interactive Entertainment] and the consuming public.” The reveal trailer for Light of Motiram, announced last year, displayed aesthetics similar to the Horizon series, including scenic environments, advanced technology, and large animal-like robots.

Sean Durkin, head of communications for Tencent Americas, said, “SIE and Tencent are pleased to have reached a confidential resolution and will have no further public comment on this matter. SIE and Tencent look forward to working together in the future.”

Featured image credit

]]> Cursor acquires Graphite in massive AI developer deal https://dataconomy.com/2025/12/22/cursor-acquires-graphite-in-massive-ai-developer-deal/ Mon, 22 Dec 2025 09:58:14 +0000 https://dataconomy.com/?p=85517 AI coding assistant Cursor acquired Graphite, an AI startup specializing in code review and debugging. The acquisition terms remained undisclosed. However, Axios reported that Cursor paid a figure substantially exceeding Graphite’s last valuation of $290 million. Graphite, a five-year-old company, achieved this valuation earlier this year after raising a $52 million Series B round. This […]]]>

AI coding assistant Cursor acquired Graphite, an AI startup specializing in code review and debugging.

The acquisition terms remained undisclosed. However, Axios reported that Cursor paid a figure substantially exceeding Graphite’s last valuation of $290 million. Graphite, a five-year-old company, achieved this valuation earlier this year after raising a $52 million Series B round.

This integration aims to streamline the code development lifecycle. AI-generated code frequently contains errors, necessitating significant engineering time for corrections. Although Cursor’s Bugbot product offers AI-powered code review, Graphite provides a specialized “stacked pull request” capability. This feature allows developers to manage multiple interdependent code changes concurrently without awaiting individual approvals, thereby accelerating the process from drafting to shipping code.

Other companies operating in the AI-powered code review sector include CodeRabbit, which received a $550 million valuation in September, and Greptile, a smaller competitor that announced a $25 million Series A funding round this fall.

Michael Truell, co-founder and CEO of Cursor, met Graphite’s co-founders, Merrill Lutsky, Greg Foster, and Tomas Reimers, prior to Cursor’s launch. Truell was a Neo Scholar, a program for college students run by Neo, Ali Partovi’s early-stage venture firm. Neo supported Graphite at its seed stage, according to PitchBook data. Accel and Andreessen Horowitz also hold investments in both Cursor and Graphite.

Cursor, last valued at $29 billion in November, has engaged in multiple acquisitions. Last month, it purchased Growth by Design, a technology recruiting strategy company. In July, Cursor acquired talent from the AI-powered CRM startup Koala for a post-money valuation of $129 million, according to PitchBook.

Featured image credit

]]> Japan’s $6 billion bet on domestic foundational AI https://dataconomy.com/2025/12/22/japans-6-billion-bet-on-domestic-foundational-ai/ Mon, 22 Dec 2025 09:51:03 +0000 https://dataconomy.com/?p=85506 The Japanese government will provide roughly 1 trillion yen (US$6.34 billion) over five years to a planned new company developing home-grown artificial intelligence, a source close to the matter said Sunday. The company will be established by around ten firms, including SoftBank Group Corp. These firms aim to create Japan’s largest base AI model through […]]]>

The Japanese government will provide roughly 1 trillion yen (US$6.34 billion) over five years to a planned new company developing home-grown artificial intelligence, a source close to the matter said Sunday.

The company will be established by around ten firms, including SoftBank Group Corp. These firms aim to create Japan’s largest base AI model through public-private cooperation, the source added.

The five-year support scheme is scheduled to begin in fiscal 2026, which starts next April, according to the source.

Featured image credit

]]> OpenAI gives users a mixing board for AI personality https://dataconomy.com/2025/12/22/openai-gives-users-a-mixing-board-for-ai-personality/ Mon, 22 Dec 2025 09:49:25 +0000 https://dataconomy.com/?p=85504 OpenAI announced new personalization options in ChatGPT’s Personalization menu, allowing users to adjust the chatbot’s warmth, enthusiasm, emoji use, headers, and lists to “More,” “Less,” or “Default” settings via a social media post. These controls enable further customization of ChatGPT’s tone beyond the existing base style and tone selections. Users can choose among Professional, Candid, […]]]>

OpenAI announced new personalization options in ChatGPT’s Personalization menu, allowing users to adjust the chatbot’s warmth, enthusiasm, emoji use, headers, and lists to “More,” “Less,” or “Default” settings via a social media post.

These controls enable further customization of ChatGPT’s tone beyond the existing base style and tone selections. Users can choose among Professional, Candid, and Quirky tones, which OpenAI introduced in November. The new options appear directly in the Personalization menu, providing granular adjustments to specific elements of the chatbot’s response style.

ChatGPT’s tone has faced multiple adjustments throughout the year. OpenAI rolled back one update after it rendered the chatbot “too sycophant-y.” Later, following complaints from some users that GPT-5 seemed colder and less friendly compared to prior versions, OpenAI modified the model to become “warmer and friendlier.” These changes reflect OpenAI’s responses to user feedback on the chatbot’s conversational characteristics.

Featured image credit

]]> MIT’s JETS model predicts disease from Apple Watch data https://dataconomy.com/2025/12/22/mits-jets-model-predicts-disease-from-apple-watch-data/ Mon, 22 Dec 2025 09:48:50 +0000 https://dataconomy.com/?p=85807 Researchers from MIT and Empirical Health recently developed a foundation model to predict medical conditions by utilizing 3 million person-days of Apple Watch data. The study, titled JETS: A Self-Supervised Joint Embedding Time Series Foundation Model for Behavioral Data in Healthcare, has been accepted to a workshop at NeurIPS. It adapts the Joint-Embedding Predictive Architecture […]]]>

Researchers from MIT and Empirical Health recently developed a foundation model to predict medical conditions by utilizing 3 million person-days of Apple Watch data.

The study, titled JETS: A Self-Supervised Joint Embedding Time Series Foundation Model for Behavioral Data in Healthcare, has been accepted to a workshop at NeurIPS. It adapts the Joint-Embedding Predictive Architecture (JEPA), an AI concept proposed by Yann LeCun, which teaches an AI to infer the meaning of missing data rather than reconstructing the data itself. This approach allows the model to predict what missing parts represent.

Meta previously released its I-JEPA model in 2023. At that time, Meta stated LeCun’s vision was to create “machines that can learn internal models of how the world works so that they can learn much more quickly, plan how to accomplish complex tasks, and readily adapt to unfamiliar situations.” LeCun has since left Meta to focus on world models, which he argues are critical for achieving Artificial General Intelligence (AGI).

The JETS model applies JEPA’s joint-embedding approach to irregular multivariate time-series data, such as long-term wearable data where measurements like heart rate, sleep, and activity may appear inconsistently or with large gaps. The study utilized a longitudinal dataset from 16,522 individuals, totaling approximately 3 million person-days. For each individual, the researchers recorded 63 distinct time series metrics at a daily or lower resolution. These metrics fall into five domains:

  • Cardiovascular health: Metrics related to heart function.
  • Respiratory health: Metrics pertaining to breathing.
  • Sleep: Data on sleep patterns.
  • Physical activity: Information on movement and exercise.
  • General statistics: Broader health-related measurements.

Only 15% of the participants had labeled medical histories, which would have rendered 85% of the data unusable in traditional supervised learning methods. JETS, however, addressed this by first learning from the complete dataset through self-supervised pre-training and then fine-tuning on the labeled subset. The researchers converted each observation into a token by making triplets of data from day, value, and metric type. This token then underwent a masking process, encoding, and was fed through a predictor to anticipate the embedding of the missing patches.

The researchers evaluated JETS against other baseline models, including a Transformer-based version of JETS, using AUROC and AUPRC. These metrics assess how well an AI model discriminates between positive and negative cases. JETS achieved the following AUROC percentages for various conditions:

  • High blood pressure: 86.8%
  • Atrial flutter: 70.5%
  • Chronic fatigue syndrome: 81%
  • Sick sinus syndrome: 86.8%

AUROC and AUPRC indicate how effectively a model ranks or prioritizes likely cases, rather than providing a strict accuracy index. This study demonstrates a method for maximizing insights from incomplete or irregular data collected by consumer wearables, even when wear times are intermittent. Some health metrics were recorded as little as 0.4% of the time, while others appeared in 99% of daily readings.

Featured image credit

]]> Spotify data breach: 86 million audio files leaked online https://dataconomy.com/2025/12/22/spotify-data-breach-86-million-audio-files-leaked-online/ Mon, 22 Dec 2025 09:20:58 +0000 https://dataconomy.com/?p=85479 A pirate activist group extracted Spotify’s entire music catalog and released approximately 300 terabytes of audio files and metadata across peer-to-peer networks. Anna’s Archive documented the leak on Thursday, covering 86 million audio files and 256 million rows of track metadata that represent roughly 99.6 percent of all listening activity on the platform. Spotify conducted […]]]>

A pirate activist group extracted Spotify’s entire music catalog and released approximately 300 terabytes of audio files and metadata across peer-to-peer networks. Anna’s Archive documented the leak on Thursday, covering 86 million audio files and 256 million rows of track metadata that represent roughly 99.6 percent of all listening activity on the platform.

Spotify conducted an investigation into the unauthorized access. A spokesperson told Billboard, “An investigation into unauthorized access identified that a third party scraped public metadata and used illicit tactics to circumvent DRM to access some of the platform’s audio files. We are actively investigating and mitigating the incident.” This response outlines the method of the breach, which involved combining publicly available data with techniques to bypass digital rights management protections.

Anna’s Archive, an organization that typically preserves books and academic papers, described the release as a “preservation archive” for music. The group stated that the effort aligns with its mission of preserving humanity’s knowledge and culture. The collection surpasses previous efforts significantly, containing 37 times more unique recordings than MusicBrainz, the prior largest open-source music database with approximately 5 million unique International Standard Recording Codes, or ISRCs.

The leaked metadata covers an estimated 99.9 percent of Spotify’s 256 million tracks and includes 186 million unique ISRCs. Anna’s Archive prioritized the files using Spotify’s own popularity metric and captured songs available through July 2025. This prioritization ensures that the most streamed tracks appear first in the distribution process.

The group is releasing the data in stages to manage the volume. Metadata has already become available for download, while the music files are being distributed in order of popularity across peer-to-peer networks. This staged approach allows for broader dissemination without overwhelming initial servers.

Yoav Zimmerman, CEO and co-founder of Third Chair, a startup that develops legal tools for media companies, commented on the accessibility of the leaked data. In a LinkedIn post, he wrote, “Anyone can now, in theory, create their own personal free version of Spotify (all music up to 2025) with enough storage and a personal media streaming server like Plex. The only real barriers are copyright law and fear of enforcement.” Zimmerman emphasized that the data is already circulating on peer-to-peer networks and added, “There is no putting this back in Pandora’s box.”

Zimmerman also addressed broader applications of the leak. He observed that the breach makes it dramatically easier for AI companies to train models on modern music at scale, with copyright law and the deterrent of enforcement serving as the primary obstacles.

A Spotify spokesperson said in an emailed statement: “Spotify has identified and disabled user accounts engaged in unlawful scraping. We have implemented new safeguards against copyright-infringing activity and are actively monitoring for suspicious behavior. Since day one, we have stood with the artist community against piracy and continue to work with our industry partners to protect creators and defend their rights.”

Featured image credit

]]> AstraZeneca bets $2B on Jacobio’s pan-KRAS cancer drug https://dataconomy.com/2025/12/22/astrazeneca-bets-2b-on-jacobios-pan-kras-cancer-drug/ Mon, 22 Dec 2025 09:19:14 +0000 https://dataconomy.com/?p=85473 Jacobio Pharma granted AstraZeneca exclusive global development and commercialization rights to its experimental pan-KRAS inhibitor JAB-23E73 in a licensing agreement announced December 21 with a potential total value of up to $2.015 billion. The Hong Kong-listed biotech firm detailed the agreement terms on December 21. Jacobio will receive a $100 million upfront payment. The company […]]]>

Jacobio Pharma granted AstraZeneca exclusive global development and commercialization rights to its experimental pan-KRAS inhibitor JAB-23E73 in a licensing agreement announced December 21 with a potential total value of up to $2.015 billion.

The Hong Kong-listed biotech firm detailed the agreement terms on December 21. Jacobio will receive a $100 million upfront payment. The company remains eligible for development and commercial milestone payments totaling up to $1.915 billion. Jacobio also qualifies for tiered royalties on net sales of the drug outside China. AstraZeneca assumes responsibility for all clinical development, regulatory submissions, and commercialization activities for JAB-23E73 outside China. Within China, Jacobio and AstraZeneca will jointly develop and commercialize the drug.

Jacobio publicly confirmed the licensing deal on Sunday. The partnership expands AstraZeneca’s oncology portfolio. It provides validation for Jacobio’s cancer drug development platform. JAB-23E73 targets multiple KRAS mutation subtypes. These mutations occur in approximately 23 percent of all cancer patients. KRAS represents the most frequently mutated oncogene in human cancers. The oncogene drives tumors in pancreatic, colorectal, and lung cancers.

The drug currently undergoes evaluation in Phase I trials in China and the United States. Early signs of anti-tumor activity have appeared in these trials. Matt Hellmann, Senior Vice President of Early Oncology and Precision Medicine at AstraZeneca’s Oncology R&D, addressed the drug’s relevance in a press release statement. He said, “KRAS is one of the most important oncogenes in cancer, with KRAS-mutated tumours driving profound unmet need for patients with pancreatic, colorectal and lung cancers.” Hellmann continued, “By advancing KRAS inhibitors like JAB-23E73, and in combination with our diverse oncology portfolio, we aim to accelerate the development of new treatment regimens that have the potential to transform outcomes for patients.”

The agreement constitutes AstraZeneca’s first transaction with a multinational corporation in the global pan-KRAS inhibitor pipeline. In 2023, AstraZeneca licensed a KRAS G12D inhibitor from China’s Usynova for a $24 million upfront payment. That inhibitor addressed a narrower mutation scope compared to the broader range targeted by the pan-KRAS approach of JAB-23E73.

Jacobio developed JAB-23E73 using its proprietary induced allosteric drug discovery platform. The inhibitor acts on both the active and inactive states of KRAS. It demonstrates strong selectivity that spares HRAS and NRAS.

Featured image credit

]]> Your Android might keep Google Assistant instead of Gemini for a while https://dataconomy.com/2025/12/22/your-android-might-keep-google-assistant-instead-of-gemini-for-a-while/ Mon, 22 Dec 2025 09:16:00 +0000 https://dataconomy.com/?p=85470 Google announced an extension to its timeline for retiring Assistant from most Android phones beyond the end of 2025, replacing it with Gemini to ensure a seamless transition for users. The company originally planned to remove Assistant from most Android phones by the end of 2025. This change followed the launch of Gemini, which incorporates […]]]>

Google announced an extension to its timeline for retiring Assistant from most Android phones beyond the end of 2025, replacing it with Gemini to ensure a seamless transition for users.

The company originally planned to remove Assistant from most Android phones by the end of 2025. This change followed the launch of Gemini, which incorporates capabilities previously handled by Assistant. Google stated that it requires additional time to position its AI assistant as the new default digital helper across its user base.

Google specified that it is adjusting the previously announced timeline to “make sure [it delivers] a seamless transition.” Updates converting Assistant to Gemini on Android devices will proceed into the next year. The company plans to share more details in the “coming months,” which indicates the transition may extend past early 2026.

Assistant’s retirement aligned with Google’s introduction of Gemini, equipped with features such as controlling smart devices connected to phones. In 2024, Google launched the Pixel 9 Series with Gemini serving as the default assistant. This move marked the integration of Gemini into Pixel hardware from the outset.

Google has incorporated Gemini across its product lineup. The company previously outlined plans to upgrade tablets, cars, and devices that connect to phones, such as headphones and watches, with the AI-powered chatbot. These upgrades expand Gemini’s availability beyond smartphones.

Eligible devices must meet specific minimum requirements to receive the Gemini upgrade. They need to run Android 10 or a later version and include at least 2 GB of RAM. These criteria determine compatibility for the transition from Assistant to Gemini on supported hardware.

Featured image credit

]]> Sony and Honda bring PS5 gaming to the Afeela EV https://dataconomy.com/2025/12/22/sony-and-honda-bring-ps5-gaming-to-the-afeela-ev/ Mon, 22 Dec 2025 09:13:37 +0000 https://dataconomy.com/?p=85467 Sony Honda Mobility, a joint venture between Sony and Honda formed to produce electric vehicles, announced that its Afeela model will integrate PS Remote Play functionality. This feature enables drivers while parked and passengers to access PlayStation games streamed from PS5 and PS4 consoles directly to the vehicle’s infotainment system display. The system supports seamless […]]]>

Sony Honda Mobility, a joint venture between Sony and Honda formed to produce electric vehicles, announced that its Afeela model will integrate PS Remote Play functionality. This feature enables drivers while parked and passengers to access PlayStation games streamed from PS5 and PS4 consoles directly to the vehicle’s infotainment system display.

The system supports seamless resumption of gameplay using a DualSense controller previously connected at home. Users connect the controller upon entering the Afeela to continue sessions without interruption. Sony Honda Mobility specifies a minimum broadband speed of 5 Mbps for Remote Play operation, with 15 Mbps recommended to ensure a smoother streaming experience during use.

Sony Honda Mobility previously demonstrated this capability with the Afeela 1 prototype at CES 2024. The showcase highlighted remote playback of PlayStation titles through the vehicle’s integrated display, confirming the technical integration ahead of production.

The Afeela 1 remains on schedule for initial customer deliveries in 2026. This timeline aligns with the joint venture’s development progress for its electric vehicle lineup. Tesla previously provided Steam gaming support in its Model S and Model X vehicles, allowing in-car gameplay access. The company later discontinued this feature from those models.

Featured image credit

]]> Intelligence-first vs. workflow-first — The hidden architecture choice shaping AI’s future https://dataconomy.com/2025/12/22/intelligence-first-vs-workflow-first-the-hidden-architecture-choice-shaping-ais-future/ Mon, 22 Dec 2025 08:28:46 +0000 https://dataconomy.com/?p=85463 There’s a hidden fault line running through the AI industry that determines which products succeed and which fail, which companies capture value and which get disrupted, which use cases transform workflows and which languish in pilot purgatory. This fault line isn’t about model architecture or training data — it’s about a fundamental design choice that […]]]>

There’s a hidden fault line running through the AI industry that determines which products succeed and which fail, which companies capture value and which get disrupted, which use cases transform workflows and which languish in pilot purgatory. This fault line isn’t about model architecture or training data — it’s about a fundamental design choice that often goes unnamed: intelligence-first vs. workflow-first.

Understanding this distinction is critical because it shapes user expectations, trust dynamics, competitive moats, and ultimately whether AI augments or replaces human agency. Let me explain.

Defining the divide

Workflow-first AI starts with an existing business process and asks: “How can AI make this faster/cheaper/better?” The workflow remains the organizing principle. AI becomes a component within a larger system optimized for a specific task sequence.

Examples: RPA (robotic process automation), AI-powered CRMs, document processing pipelines, customer service routing systems.

Intelligence-first AI starts with a reasoning capability and asks: “What problems can this intelligence solve?” The AI’s cognitive abilities become the organizing principle. Workflows emerge from what the intelligence can do, not what the existing process requires.

Examples: ChatGPT, Claude, Cursor, Perplexity — general reasoning systems that users adapt to their needs.

This distinction might seem semantic, but it has profound implications.

Why this distinction matters: Four key dimensions

1. Flexibility vs. reliability trade-offs

Workflow-first systems optimize for predictability. They’re designed to perform specific tasks consistently within defined parameters. This makes them easier to validate, easier to integrate, easier to trust — but harder to adapt when requirements change.

Intelligence-first systems optimize for adaptability. They’re designed to handle novel situations, interpret ambiguous inputs, and generalize across contexts. This makes them powerful and flexible — but harder to validate, harder to integrate, harder to trust.

The irony: enterprises crave both reliability AND flexibility, but these goals create architectural tension. Workflow-first designs deliver reliability at the cost of rigidity. Intelligence-first designs deliver flexibility at the cost of unpredictability.

This is why 90–95% of GenAI experiments never reach production. Organizations prototype with intelligence-first tools (ChatGPT, Claude), discover powerful capabilities, then realize they can’t deploy something this unpredictable into production workflows that require consistency guarantees.

2. User agency and control

Workflow-first AI preserves human decision-making authority. The AI performs specific subtasks, but humans remain in the loop for judgment calls, exceptions, and final decisions. This aligns with the behavioral economics insight that users need to maintain agency to trust delegation.

Intelligence-first AI requires users to trust the AI’s reasoning process. When you ask ChatGPT to “analyze this data and recommend next steps,” you’re delegating not just execution but judgment. This triggers identity loss aversion — the psychological resistance to letting machines think for you.

This explains the “Copilot pattern” — intelligence-first systems that succeed tend to be designed as collaborative tools (GitHub Copilot, Cursor) rather than autonomous agents. The intelligence is first-class, but user control is preserved through suggestive rather than directive interaction.

3. Competitive moats and market structure

Workflow-first AI creates vertical integration opportunities. If you can embed AI deeply into a specific workflow (legal document review, medical diagnostics, financial reconciliation), you build a moat through process expertise, integration depth, and switching costs.

Intelligence-first AI creates horizontal platform opportunities. General reasoning capabilities can be applied across industries and use cases, enabling platform dynamics where one foundation model serves thousands of applications.

This is why we see simultaneous trends:

  • Foundation model consolidation (OpenAI, Anthropic, Google) — intelligence-first platforms with massive scale advantages
  • Vertical AI proliferation (Harvey for law, Hippocratic for healthcare, Glean for enterprise search) — workflow-first applications with deep domain integration

The most successful AI companies will likely operate at both layers: intelligence-first foundations feeding workflow-first applications.

4. Trust and adoption dynamics

Here’s where behavioral economics meets architecture: workflow-first systems align with how enterprises build trust through progressive delegation. You start with low-stakes tasks (data entry), prove reliability, then gradually expand scope. This matches the psychological principle of building trust through repeated small successes.

Intelligence-first systems require users to make a leap of faith: trust the AI’s reasoning without observing gradual competence building. This is much harder psychologically, which is why intelligence-first adoption often happens consumer-first (ChatGPT) where individual users can experiment at low risk, then eventually migrates to enterprise once sufficient social proof exists.

The hybrid convergence thesis

Here’s the contrarian insight: the dichotomy between intelligence-first and workflow-first is dissolving. The most sophisticated AI systems are converging toward a hybrid architecture that combines:

  • Intelligence layer: General reasoning capabilities (foundation models)
  • Workflow layer: Structured task orchestration (agents, tools, guardrails)
  • Control layer: Human oversight and intervention points

This three-layer stack enables organizations to benefit from general intelligence while maintaining workflow reliability and user control.

Example: Cursor (AI code editor)

  • Intelligence layer: Claude/GPT-4 for code understanding and generation
  • Workflow layer: Integrated into development workflow with git, linters, tests
  • Control layer: Suggestions require human review; user remains author

This hybrid approach addresses the core behavioral economics challenge: it provides AI capabilities that feel like enhanced tools rather than autonomous replacements.

Implications for AI strategy

If you’re building or buying AI, this framework suggests three strategic questions:

1. What’s your primary constraint: Flexibility or reliability?

  • If reliability: workflow-first architecture, accept limited scope
  • If flexibility: intelligence-first architecture, invest in trust-building

2. Where does your competitive advantage lie?

  • Process expertise → workflow-first (vertical integration)
  • General capabilities → intelligence-first (horizontal platform)

3. How does your user build trust?

  • Progressive delegation → workflow-first gradual expansion
  • Experimentation → intelligence-first with strong guardrails

The folder paradigm as hybrid architecture

Here’s where this gets personally relevant: the “folder paradigm” I’ve been exploring (AI agents that own directories as cognitive architecture) is fundamentally a hybrid architecture optimized for intelligence-first reasoning within workflow-first constraints.

Each agent has:

  • Intelligence layer: LLM reasoning over documents, tools, context
  • Workflow layer: File system as structured memory, standardized interfaces
  • Control layer: Human-readable files, explicit decision logs, intervention points

This design preserves user agency (you can read/edit any file), enables progressive delegation (start with narrow agent scope, expand gradually), and combines general intelligence with workflow integration.

It’s an architecture that says: “AI agents should be intelligent enough to reason flexibly, but structured enough to behave predictably.”

Why the industry hasn’t converged on this yet

If hybrid architecture is optimal, why hasn’t the market converged? Three reasons:

  1. Technological immaturity: Foundation models are still rapidly improving. Premature workflow integration creates technical debt when the intelligence layer upgrades.
  2. Organizational inertia: Enterprises struggle to redesign workflows around AI. It’s easier to plug AI into existing processes (workflow-first) than to reimagine work (intelligence-first).
  3. Unclear value capture: Intelligence-first platforms (OpenAI) and workflow-first applications (vertical AI) have clear business models. Hybrid architecture requires new organizational capabilities (AI operations teams, hybrid design skills) that are still emerging.

But this is changing. As foundation models stabilize, as enterprises build AI expertise, and as successful patterns emerge (Copilot model, agentic frameworks), we’ll see convergence toward hybrid architectures that deliver both intelligence and reliability.

The ultimate insight: Architecture shapes psychology

The deepest reason this distinction matters: architecture choices shape user psychology, which determines adoption, which determines success.

Workflow-first architecture signals: “This is a tool that does what you tell it.” This preserves agency, builds trust through demonstrated competence, and aligns with existing mental models.

Intelligence-first architecture signals: “This is a reasoning agent that thinks for you.” This triggers identity loss aversion, requires trust leaps, and challenges existing mental models.

The winning architecture is the one that delivers AI capabilities while managing the psychological transition. That’s why I believe hybrid intelligence-first-reasoning-within-workflow-first-structure will dominate: it maximizes AI capability while minimizing psychological resistance.

Conclusion: The choice that shapes everything

The intelligence-first vs. workflow-first distinction isn’t just about system design — it’s about:

  • Trust dynamics: How users build confidence in AI delegation
  • Competitive strategy: Where moats emerge (vertical integration vs. horizontal platforms)
  • Adoption paths: Consumer experimentation vs. enterprise validation
  • Psychological framing: Tool augmentation vs. agent autonomy

As AI capabilities mature, the distinction will blur. But understanding it now helps explain why some AI products succeed while others languish in pilot purgatory, why enterprises simultaneously crave and fear AI agents, and why the path to AI adoption runs through architectural choices that shape human psychology.

The companies that win won’t just have better models or better workflows — they’ll have better psychological architectures that deliver intelligence users can trust, flexibility they can control, and workflows they can understand.

That’s the hidden design choice shaping AI’s future.

Featured image credit

]]>