NAll news

Tuesday581 items

  1. Guillermo Rauch
    AI score39

    Rauch: Stateless sandboxes are the wrong benchmark; agents need to carry their workspace

    The Vercel developer account states that Vercel Sandboxes with Drives start 2.4x faster, with median creation time dropping from 1.7s to 0.7s and p95 dropping from 2.2s to 1.6s, with no SDK upgrade or configuration changes required. Vercel CEO Guillermo Rauch commented that stateless sandboxes are the wrong benchmark, and that agents are only useful when they carry their own workspace (identity, skills, repositories, memory, etc.); he also praised the Vercel Sandbox team for delivering on latency and throughput SLOs, while maintaining reliability and security amid clusters deployed across 19 global regions, new features shipping weekly, and week-over-week demand growth in the double digits.

    Vercel Sandboxes with Drives start 2.4x faster. • Median creation drops from 1.7s to 0.7s • p95 creation drops from 2.2s to 1.6s • No SDK upgrade or config changes needed https://t.co/nuQLD0BZrp

    #English text is being prepared.#Agents#English text is being prepared.

  2. Swyx
    AI score25

    swyx shares MIT's Anqi Zhang discussing recursive language models

    swyx reposted with "lmaoooo" the Latent.Space podcast content: MIT's @a1zhang explains why Claude Code, Codex, and Pi are essentially the same, how RLM generalizes across different tasks with the help of code, context offloading, and recursive subagents, and what an OpenAI experiment with 10,000 agents and 130B output tokens reveals.

    Recursive Language Models: Claude Code, Agent Swarms, Big Research Bets, & the $40M AI Problem Solver https://t.co/VceYUKpcTo MIT’s @a1zhang explains why Claude Code, Codex, and Pi are basically the same, how RLMs use code, context offloading, and recursive subagents to generalize across tasks, why one expert can sometimes replace a trillion-token brute-force search, what OpenAI’s 10,000-agent, 130B-output-token experiment reveals about the future of language models, and why academia’s biggest advantage is the freedom to take weird, ambitious research bets.

    #English text is being prepared.

  3. Ethan Mollick
    AI score26

    Mollick: AI adoption rate continues to grow

    Ethan Mollick said the AI adoption rate is still continuing to grow, and recalled that last summer people once thought the adoption rate was declining. He also mentioned that Simon Willison had a good discussion of the methodological changes in a post.

    In updating a graphic showing U.S. company AI adoption over time according to a long-running U.S. Census Bureau survey, I found out that the Bureau's work has been sullied by that asinine "Super Intelligence" executive order https://t.co/YEu1tnWweT

    #English text is being prepared.

  4. Ethan MollickOpinion
    AI score62

    Ethan Mollick shares research on GPT-6 Astra directly rewriting model weights

    Ethan Mollick shared a study by Rikiya Takehi and said he is waiting for replication, but considers it a big deal. The study claims that having GPT-6 Astra directly modify the weights of Qwen3.5-0.8B to improve its math ability, changing only 0.016% of the weights, fixed the model's reasoning loops, improving math ability by up to 6x and instruction following by 3x, all without using gradients or training; the authors call this Weight-Space Programming.

    We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: https://t.co/bOWCcFhCVq

    #English text is being prepared.#English text is being prepared.#English text is being prepared.

  5. East Money
    AI score42

    China's first AI theatrical film "Sanxingdui: Future Past" flops at the box office: only 1.26 million yuan in 5 days

    China's first AI feature-length theatrical film to obtain a public screening license, "Sanxingdui: Future Past," has grossed only about 1.26 million yuan in cumulative box office over 5 days since release, with an attendance rate of less than 1%, and some platforms predict its total box office may reach only two to three million yuan. The film's production cost is in the tens of millions of yuan range, and audience feedback centers on "chaotic narrative," "too many logical holes," and a "strong AI feel." The article argues that AI can compress costs and enable small and medium-sized teams to dare to take on heavy-VFX subject matter, but audiences pay for the story, not for the tool.

    #English text is being prepared.#Video

  6. The Verge AI
    AI score50

    Anthropic Tightens Claude Code Restrictions: Prohibits Replicating Existing Games

    Anthropic has updated the system prompt for Claude Code, prohibiting the tool from being used to replicate existing video games, while still allowing users to create original games, games of a certain genre, or games based on users' own designs and assets. Anthropic's Addy Osmani confirmed that the company has updated Claude Code's prompt on Opus 5.5, but did not confirm the specific wording in screenshots posted by X users.

    Same story · 12 reports

    #Product updates#English text is being prepared.#Anthropic

  7. Claude BlogFirst-partyProduct release
    AI score52

    Anthropic Launches Message Threads in Public Beta on the Messages API

    Anthropic has launched the public beta of message threads for the Messages API on the Claude Platform; agents now only need to send newly added content each turn rather than resending the full conversation history. For requests of 500k to 1M tokens, median time to first token (TTFT) drops by about 35%, and the feature is enabled by default for most Claude Code users. To use it, add the thread parameter to the request, send the full conversation and open the thread in the first turn, then include only the new message and the previous response ID in subsequent turns; threads expire 5 minutes after the last request, are stored encrypted in memory during that period and not written to disk, and token usage remains unchanged.

    Same story · 4 reports

    #Product updates#Agents#English text is being prepared.

  8. Business Insider TechProduct release
    AI score55

    John Deere launches AI assistant JD to help farmers handle decades of field data

    John Deere unveiled its AI assistant JD in September at the Farm Progress Show in Iowa. It is integrated into the Operations Center platform, which connects farmers' field and machine data, and farmers can ask questions on their phones, for example determining the best harvest time based on historical trends. Melissa Neuendorf, principal product manager for customer AI solutions at John Deere, said farmers' data volumes are too large and manual analysis can sometimes take weeks, while JD can complete the analysis in minutes. A McKinsey survey of 5,500 farmers released in September showed that 72% expect generative AI to affect operations within five years, but only 17% are currently using it; Matt Darr, a professor at Iowa State University, mentioned that farmers are concerned about data privacy and its impact on commodity prices.

    #Product updates#Agents#English text is being prepared.

  9. TechCrunch AI
    AI score62

    New York Times and Other Media Sue Brave for Alleged Content Scraping

    The New York Times and other media outlets sued the browser Brave in Manhattan federal court, alleging that it used an unnamed crawler to build an index of more than 30 billion web pages, leaving publishers unable to block it. The complaint also targets Brave's practice of bypassing paywalls and blocking publishers' ads while selling its own ads; the New York Times is seeking an injunction and damages. On the same day, the Daily News, Chicago Tribune, and Orlando Sentinel filed separate similar lawsuits.

    #Industry news#English text is being prepared.

  10. Lauren Tan
    AI score22

    Lauren built a game with Grok Bot and Cursor, and its non-story mode has been played nearly 100,000 times

    After the game Lauren built with Grok Bot and Cursor launched, its non-story mode has been played nearly 100,000 times. Since launch, she has added 17 languages, adjusted the balance, and upgraded all visual effects, and she also plans to add a new equipment system.

    Day 28/ tibo's game has landed i spent $5.99 on this domain, but i had a blast making it with grok bot and cursor! i hope you enjoy it and find all the easter eggs 🤍 if i did it right you should be able to play it inside of X too https://t.co/diV7PyPaqv

    #Industry news

  11. Claude DevsFirst-partyProduct release
    AI score62

    Anthropic Opens Message Threads Beta in the Messages API

    Anthropic has launched the message threads beta on its Messages API. The API caches conversations server-side, so subsequent requests only need to send the new message and the previous message ID, allowing Claude to start responding faster. The company says the feature is already integrated in Claude Code, with median time to first token for 500K to 1M token requests reduced by about 35%, and median request size for long conversations dropping from about 800 KB to 1.8 KB. To use it, add the thread parameter and beta header to requests. The feature is available on Claude Platform, Claude Platform on AWS, Claude in Microsoft Foundry, and Claude on Google Cloud, and is billed at standard token prices.

    Same story · 4 reports

    #Product updates#Agents#English text is being prepared.

  12. Claude DevsFirst-partyProduct release
    AI score60

    Claude Code integrates message threads, reducing first-token latency for long requests by about 35%

    After Claude Code integrated message threads, on requests of 500K to 1M tokens, the median first-token latency dropped by about 35%; in long conversations, the median size of a single request fell from about 800 KB to 1.8 KB. Message threads have entered public beta in the Messages API. The server caches the conversation between turns, so callers only need to send the new message and the ID of the previous message, and must add the thread parameter and beta header. It is available on Claude Platform, Claude Platform on AWS, Claude in Microsoft Foundry, and Claude on Google Cloud, and is billed at standard token prices.

    #Product updates#English text is being prepared.#Anthropic

  13. Claude DevsFirst-partyProduct release
    AI score66

    Anthropic Opens Public Beta of Message Threads in the Messages API

    Anthropic has opened the public beta of message threads on the Messages API. The API caches conversations on the server side, so developers only need to send the new message and the previous message ID, without resending the full conversation. The feature has been integrated into Claude Code; on requests of 500K to 1M tokens, median time to first token dropped by about 35%, and in long conversations the median request size fell from about 800 KB to 1.8 KB. To use it, add the thread parameter and beta header to requests. The feature is available on Claude Platform, Claude Platform on AWS, Claude in Microsoft Foundry, and Claude on Google Cloud, and is billed at standard token prices.

    Same story · 4 reports

    #Product updates#Agents#English text is being prepared.

  14. Peter Yang
    AI score20

    Peter Yang: Gemini is strong at writing but limited by Antigravity

    Peter Yang said he asked ChatGPT and Gemini to suggest revisions for an article at the same time. ChatGPT's suggestions were almost all poor, while Gemini's were quite excellent. However, Antigravity's browser-use feature seemed to be unavailable, so he could only paste Gemini's revisions back into ChatGPT to finish editing in the browser, and he hopes Antigravity will improve as soon as possible.

    #English text is being prepared.#Agents#OpenAI

  15. The Beijing News
    AI score49

    Unitree? No—Yufan Intelligence's quadruped robot "Lingmao" climbs Mount Tai, takes first place in autonomous speed ranking

    Yufan Intelligence's fully self-developed quadruped robot "Lingmao" won first place in the autonomous speed ranking at the first CMG World Robot Mount Tai Climbing Competition, and for the first time completed verification of fully autonomous navigation, autonomous obstacle avoidance, and autonomous decision-making capabilities. "Lingmao" weighs only 15 kilograms, has a payload of 20 kilograms, and a payload-to-weight ratio of 1.3:1. It is the smallest quadruped robot in this competition, competing against larger 40–65 kilogram-class robot dogs. Its best time for autonomously completing the 3.9-kilometer course was 35 minutes and 47 seconds, faster than the approximately 40 to 45 minutes achieved by professional Mount Tai climbing companions.

    #Industry news#Embodied AI

  16. Larsen Cundric
    AI score33

    Phone Harness MCP Lets Agents Control Cloud Phones

    shawn releases Phone Harness MCP: MCP-enabled agents (such as Claude, ChatGPT, Cursor, Codex) can act as custom connectors to control real iPhone or Android phones in the cloud, able to see the screen, tap, type and swipe, with no installation required—just paste the URL and sign in. It is now live in phone-harness 0.3.3. Larsen Cundric reposted it, calling it "pretty cool," and asked what use cases it has.

    Introducing: Phone Harness MCP 🔌 Any agent that supports MCP can now control a real iPhone or Android in the cloud: > works in Claude, ChatGPT, Cursor, Codex, and more as a custom connector > nothing to install, just paste a URL and sign in > sees the screen, taps, types, and swipes like you do Available in phone-harness 0.3.3. https://t.co/88u1gSgvtS

    #English text is being prepared.

  17. Testing CatalogModel releases
    AI score56

    Nace AI Releases Financial Decision Model Drex for Finance

    Nace AI has released Drex for Finance, a multimodal decision model for financial scenarios, used for sub-second financial decisions such as fraud detection; it can combine documents or transaction logs with a fixed set of candidate answers to output a probability score for each answer. On 3,361 decisions across 48 held-out financial benchmarks not used in training, its accuracy was 85.7%, compared with 70.9% for GPT-6 Luna. The model is available only in Nace Console.

    Announcing Drex for Finance - the multimodal decision model, post-trained for sub-second financial decisions. Financial operations require precise, regulatory-grade decisions with zero hallucinations across filings, support, credit risk, live fraud prevention, quant signals, and compliance. Evaluated on 48 held-out finance benchmarks: - 85.7% accuracy across 3,361 decisions (+14.8% vs. GPT-6 Luna) - 77.3 Finance Index (+13.7 vs. Jev 1.14, +11.5 vs. Drex-Max v1.1) Trained for sub-second financial intelligence across the entire pipeline: - Financial Documents & Filings: Extraction decisions, classification, and clustering - Credit & Lending: Automated underwriting, PD modeling, and EWS - Fraud & AML: Sub-second transaction scoring, ATO defense, and SAR clustering - Markets & Trading: Regime detection, meta-labeling, and LOB execution signals - Insurtech & RegTech: Claims fraud detection, sanctions screening, and audit reconciliation Available exclusively on Nace Console: https://t.co/vdEavE1LhL

    #Model releases#Reasoning

  18. Swyx
    AI score28

    swyx: AI Engineer YouTube Reaches 2 Million Engineers Monthly

    swyx said that after 3 years of building from scratch, the AI Engineer (aie) YouTube channel now reaches 2 million unique engineers each month, 76% of whom have not yet subscribed. He called for subscriptions to maintain momentum and push for a breakthrough around the ~1M subscriber milestone.

    In Jan this year I called my content strategy shot: "Scaling without Slop". It's finally starting to work. It took us 3 years to reach our first 100k on youtube. It only took 1.2 months for the next 100k. Similar other metrics on AEO/SEO/subscriber traction and have a lot of New Media ideas that I'm excited to pursue. officially giving notice of the next phase of Latent Space, AINews, and what the rest of swyx inc has been cooking below

    #Industry news

  19. Rohan PaulProduct release
    AI score67

    Skyfall AI releases engineering world model SKY-001, which can design physical products and output construction steps

    Skyfall AI has released the engineering world model SKY-001, which can design physical products by planning parts and construction steps in 3D. It is claimed that its CAD precision is comparable to GPT-6 Astra, while the cost per design is about 150 times lower. SKY-001 outputs designs and ordered construction sequences, including a part tree, a sketch of each step's operation, and each intermediate solid. The model works in two stages: the Hierarchical Composer generates the overall object and decomposes it into assemblies and parts, and then the Construction World Model, using the parent level as context, predicts the visual steps for constructing each leaf node.

    Very proud to showcase our latest demo of how our Engineering World Model SKY-001 can design a complex Jaguar car with 1000s of complex parts. In order to get us to Autonomous Manufacturing, we need to design systems that can think like a ‘AI General Engineer’: planning complex physical structures, long term sequences, and feasibility directly in the physical domain rather than approximating it through next-token prediction. Engineering design operates in the spatial reasoning domain through hierarchical decomposition, constraint satisfaction, and the sequential construction of complex physical objects. World models are going to be the next frontier in AI after LLMs and SKY-001 is just the beginning.

    #Model releases#Multimodal#Reasoning