NAll news

All news

Friday1454 items

  1. The Decoder
    AI score68

    After Anthropic Tightened Its Defenses, Chinese AI Labs Turned to OpenAI's GPT-6 Astra for Distillation

    According to The Information, DeepSeek and Moonshot AI made extensive use of outputs from OpenAI's most powerful model, GPT-6 Astra, when training their latest models, involving Kimi K3.1 Pro and V4.1-Flash 1017. This kind of distillation requires models to expose hidden reasoning steps, and both OpenAI and Anthropic prohibit the practice. Claude was once the primary target, but Anthropic says the new methods have cut the amount of data attackers can obtain by 90% to 99%. In late September, OpenAI reported a coordinated distillation campaign linked to people associated with Moonshot.

    #Industry news#English text is being prepared.#DeepSeek

  2. The Decoder
    AI score75

    The New York Times says Chinese intelligence personnel approached people close to Nvidia CEO Jensen Huang

    The New York Times reported that U.S. intelligence agencies believe that a senior executive close to Nvidia CEO Jensen Huang — Alice Huang (Huang Le) of the Singapore-based data center operator Megaspeed — provides intelligence to China, and that Trump was briefed about a year ago but took no action, while Huang himself was not informed. The report also mentioned that the U.S. government suspects Megaspeed of helping China obtain Nvidia chips, and that between June 2024 and February 2026 its Malaysian sister company purchased at least $4.8 billion worth of AI hardware containing Nvidia chips. An Nvidia spokesperson said that meeting with customers is part of the CEO's job and that Megaspeed does not have access to sensitive information; Megaspeed has announced an internal investigation, while a Chinese embassy spokesperson said the allegations are baseless.

    #Industry news#English text is being prepared.

  3. The Decoder
    AI score72

    US Department of Justice Reviews Nvidia's Hugging Face Acquisition and Sales Practices

    The US Department of Justice is reviewing Nvidia's plan to acquire the open-source platform Hugging Face for $12.93 billion. Officials are asking whether the deal would give Nvidia control over a key hub for AI distribution and harm competitors. Nvidia expects the transaction to close in the first half of 2027. The same investigation also covers Nvidia's acquisitions, investments, and sales practices, as well as whether it exchanges more AI chips for customers also buying its networking equipment and Groq processors. It has previously examined the acquisitions of SchedMD and Run:ai and the licensing agreement with Groq. France's competition authority is also investigating Nvidia, focusing on the industry's dependence on CUDA and its stakes in cloud providers such as CoreWeave.

    Same story · 3 reports

    #Industry news#English text is being prepared.#Hugging Face

  4. The DecoderModel releases
    AI score68

    Mistral Releases Mistral Large 4 Weights: Coding Ability Leads Open-Source Models but Overall Trails Chinese Models

    Mistral has released the weights of Mistral Large 4 on Hugging Face. The model, nicknamed Le Chonk, has approximately 1 trillion parameters with 52 billion activated per step, and supports text and image inputs. On Terminal-Bench 4.0, its score rose from 28% in the preview version to 41%, slightly above GLM-5.3's 40%, but on the new version of the Artificial Analysis Intelligence Index it scored only 26 points, below the 32 points of DeepSeek V4.1 Flash and GLM-5.3. API pricing is $0.70 per million input tokens and $2.10 per million output tokens, under a modified MIT license; using it as a commercial cloud service requires separate authorization.

    Same story · 15 reports

    #Model releases#Open-source ecosystem#English text is being prepared.

  5. The DecoderResearch
    AI score68

    Anthropic study: experts adopt AI agents' random preferences and mistake them for their own judgments

    Anthropic ran an experiment with 50 causal inference experts; the agent's preference at each decision point was randomly assigned by a coin flip, yet participants still adopted the agent's preference in 86% of analytical decisions, with flip rates of 62% to 82%. Afterward, 52% believed the final recommendation was more like their own judgment, 35% believed it was basically or entirely their own, and only 15% mentioned AI. The study also found that after using the agent, the rate of serious analytical errors dropped from 55% (6/11) to 6% (3/50), and the median time required fell from 90 minutes to 65 minutes, but the no-AI group had a smaller sample and slightly lower expertise.

    Same story · 7 reports

    #Papers/Research#Safety / alignment#Agents

  6. The Decoder
    AI score66

    OpenAI raises usage limits for Pro plans, doubling Pro 100 and Pro 500

    OpenAI has doubled the usage limits for several paid tiers, covering the Work and Codex quotas for Pro 100 and Pro 500, both of which receive twice their previous allowances. The lower-tier quota for Pro 200 is likewise doubled, and a previously planned reduction has been withdrawn; the Plus plan remains unchanged, and the new allowances apply automatically to both new subscriptions and existing users. OpenAI says that doubling quotas does not equal doubling the number of completed tasks, because consumption varies with the model, task type, and settings. The move comes as Anthropic prepares for an IPO and the two companies compete for B2B dominance.

    Same story · 9 reports

    #Product updates#OpenAI#Anthropic

  7. OpenAI: Deployment Safety and System CardFirst-party
    AI score46

    GPT-6 Luna October 29 Update: System Card Supplementary Notes

    OpenAI released the GPT-6 Luna (October 29) update, which responds more concisely and reduces overly sycophantic phrasing; under the Preparedness Framework it is still rated High for both cybersecurity and biochemical capabilities, and it uses the same set of safeguards as GPT-6 Luna (October 7). This version improves over the previous version on the gore evaluation in Production Benchmarks, achieving a Pareto improvement between safely handling harmful requests and helpfulness; in the U18 evaluation it improves on sexual content, eating disorders, emotional dependence, and gore, and in the instruction hierarchy evaluation it reaches 99.91% robustness.

    Figure 1. Safety and non-overrefusal on challenging prompts. Higher is better on both axes

    #Safety / alignment#OpenAI

  8. MIT News · Artificial Intelligence
    AI score23

    MIT Professor Ali Jadbabaie on why infrastructure is entering a critical era: the triple pressure of AI, climate and geopolitics

    Ali Jadbabaie, head of MIT's Department of Civil and Environmental Engineering, believes that climate change, AI's energy demands and aging power grids, and supply chain vulnerabilities caused by geopolitics are arriving at the same time, pushing infrastructure into a critical period. He noted that AI is expected to accelerate and lower the cost of exploring engineering designs, but the outcome still depends on how it is done. The World Infrastructure Forum, co-founded by the MIT department together with UPM, the World Bank Group and the University of Texas at Austin, was held in Madrid on October 28.

    #Industry news#English text is being prepared.

  9. PyTorch BlogFirst-partyProduct release
    AI score58

    PyTorch 2.15 Released: Rewritten NCCL Backend Becomes Default, Apple Silicon Gains FP8 Support

    The PyTorch Foundation has released PyTorch 2.15. All process groups using nccl now default to the rewritten NCCL backend, accompanied by a flight recorder that dumps on-site information when collective communication hangs, and can be reverted via TORCH_DIST_USE_NCCL2=0 or nccl-legacy. This version also adds float8_e4m3fn and _scaled_mm support for Apple Silicon, and the new grouped_mm improves MoE decoding throughput by 1.32–1.34x on M3 Ultra. The default wheel is upgraded to CUDA 13.2, while a CUDA 13.4 wheel is also provided with experimental support for NVIDIA Rubin GPUs; a ROCm 10.0 wheel is newly added, and Python 3.10 and macOS 14 are no longer supported.

    #Product updates#Open-source ecosystem#English text is being prepared.

  10. Amazon ScienceFirst-party
    AI score39

    Amazon Science and Booking.com use knowledge distillation plus Medusa to speed up LLM inference by up to 20x

    Amazon Science and Booking.com combined knowledge distillation with Medusa to compress a 70B-parameter Llama-3.1 teacher model into a 1.1B TinyLlama student model, increasing inference speed by 10 to 20 times and reducing cost to less than one-third while maintaining the teacher model's quality. The approach has been used in three features—AI Trip Planner, Smart Filters, and accommodation description generation—and won the Best Paper Award in the Industry Track at ACL.

    #Papers/Research#Reasoning#English text is being prepared.

  11. Fudan Development InstituteFirst-party
    AI score13

    25 African Young AI Innovators from 25 Countries Visit Fudan to Discuss Technology Applications and Cooperation

    25 winners of an AI application case creativity competition from 25 African countries, including Kenya, Mauritius, and Cameroon, visited the Global Center for AI Innovation and Governance at Fudan University to participate in specialized training, industry sharing, and exchanges. Xin Yanyan, Deputy Secretary-General of the Global Center for AI Innovation and Governance, delivered specialized training on global AI governance, while Zhao Yunfeng, Vice President of International at Alibaba Cloud, and others shared insights on business practices in the African market, technical solutions, and Model-as-a-Service (MaaS). The event aimed to promote interaction between young Chinese and African innovators and Chinese universities and enterprises, and to build a platform for AI capacity building and practical cooperation.

    English text is being prepared.

    #Industry news

  12. The Verge AI
    AI score35

    Microsoft will stop supporting Copilot on smart TVs this year

    Microsoft has decided to end support for all Copilot features on TVs within this year. A pop-up notification on LG TVs states that the feature is "planned to be discontinued in 2026," with the specific date depending on Microsoft's schedule. LG and Samsung announced at CES 2025 that their new smart TVs would integrate Copilot, allowing users to talk to it, ask about on-screen content, or get movie and TV recommendations. Previously, LG drew user complaints after a webOS update added a non-removable Copilot shortcut, and later allowed it to be removed.

    #Industry news#Microsoft

  13. The Verge AIModel releases
    AI score76

    Google Releases Gemini 4 Carbon, Claims Significant Leap in Cybersecurity Capabilities

    Google has released the frontier model Gemini 4 Carbon, claiming notable improvements in coding, agentic tasks, and cybersecurity, and ranking first on the Artificial Analysis Intelligence Index (v5.0), followed by Claude Opus 5.5 and GPT-6 Astra. The model is currently available only to select cybersecurity defense companies, and will later be expanded to other partners through the Fairwind Program; it ranks first on two benchmarks, SEC-bench Pro and ExploitGym, and is priced the same as Argon at $2 per million input tokens and $10 per million output tokens.

    Same story · 34 reports

    #Model releases#Safety / alignment#English text is being prepared.

  14. The Verge AI
    AI score49

    Akai embraces vibe-coded mods with open-source MPC

    Before officially releasing the MPC Sample Key, Akai has already provided devices to the modding community ahead of time and plans to equip it with open-source firmware. The demos showed mods that use vibe coding to add Native Instruments Massive, virtual analog and FM synthesizers, as well as a device running a completely custom OS. CEO Jack O’Donnell said that inMusic's brands will give devices to developers early, open-source some software, and provide an SDK.

    #Industry news#Open-source ecosystem

  15. The Verge AIOpinion
    AI score65

    Anthropic S-1 shows roughly $850 billion in compute commitments over the next two decades

    The Verge analyzed Anthropic's S-1 filing, noting that its non-cancelable purchase commitments, not-yet-commenced lease agreements, and new compute agreements total roughly $850 billion, and would approach $1 trillion if un finalized deals are included. The filing shows that revenue in the first half of 2026 grew 14-fold year over year to $16.3 billion, adjusted revenue was $1.62 billion, and net revenue retention was 900%. The author argues that these fixed costs require Anthropic to maintain rapid growth, and that the recursive self-improvement mentioned among its risk factors is precisely the key supporting this growth assumption.

    #Industry news#Anthropic#English text is being prepared.

  16. The Verge AIOpinion
    AI score78

    SemiAnalysis Analyst Interprets Anthropic S-1: The Ledger Behind the Valuation and AGI Narrative

    The Verge's Decoder podcast invited SemiAnalysis analyst Joey Brookhart to break down the S-1 filed by Anthropic, discussing its potential Nasdaq listing at a valuation of $1.5 trillion to $2 trillion. Brookhart said Anthropic has achieved over $70 billion in ARR, with inference gross margins turning from negative 94% two years ago to positive 85%, and revenue per megawatt of inference at about $65 million, but revenue is highly concentrated in coding scenarios, and about 70% of compute is used for training and research. He also mentioned that Anthropic has $500 billion to $600 billion in compute commitments over the next 7 to 10 years, and compared OpenAI's more consumer-oriented business structure.

    #Industry news#Anthropic#OpenAI

  17. The Verge AI
    AI score27

    The New York Times Cooking app is testing an AI recipe modification feature, and users are unhappy and unable to turn it off

    The New York Times Cooking app is testing an AI recipe modification feature that can recommend ingredient and cookware substitutions and scale recipes to any number of servings. Some readers questioned the AI's environmental impact and user experience in the comments section. Senior editor Genevieve Ko said the feature currently cannot be turned off, but using it is entirely optional. NYT spokesperson Jordan Cohen said that closing comments on older articles is standard practice, and that some users have already tried the feature on the iOS app and provided positive feedback.

    #Product updates#English text is being prepared.

  18. The Verge AI
    AI score62

    Microsoft offers voluntary severance to about 60,000 employees

    Microsoft is offering voluntary severance packages to about 60,000 employees in the United States, the United Kingdom, Canada, and India, a scale far exceeding the retirement package offered to about 9,000 people earlier this year. As of June 30, Microsoft had about 223,000 employees, down 5,000 from the previous year, the first year-over-year decline in a decade; eligible employees must decide whether to accept by December. The article also mentions that Turn 10 Studios head Dan Greenawalt will retire, and that Bethesda may announce the title "The Elder Scrolls VI: Sentinel" next month.

    Same story · 15 reports

    #Industry news#English text is being prepared.#Microsoft

  19. The Verge AIProduct release
    AI score60

    Google Gemini now supports generating and editing Docs, PDFs, and other files directly within conversations

    Google Gemini can now generate, preview, and edit PDFs, Google Docs, Google Sheets, Slides, Markdown, TXT, and code files directly within conversations, available to all Gemini users starting today. PDF preview and editing are available on both web and mobile, while other formats are currently limited to web. Users can expand documents to full screen for editing, or select text within a file to ask Gemini questions or have it make modifications, such as selecting a table and requesting it be converted into a chart. Earlier this year, Google already supported generating and downloading files in Gemini; what's new this time is the ability to modify files without leaving the Gemini app.

    Same story · 7 reports

    #Product updates#Google#Agents

  20. InterconnectsOpinion
    AI score53

    Artificial Analysis v5 and the Benchmarker's Dilemma

    Nathan Lambert reviews the newly released Intelligence Index v5 from Artificial Analysis, which weights knowledge work, coding, science, and general each at 25%, and presents new model rankings. He notes that 8 closed-source models in the rankings lead the best open-source model, Kimi K3.1 Pro, and argues that the new index leans more toward measuring where frontier models are heading, potentially underestimating the role of open-source models in the economy. The article also discusses the pros and cons of shifting to private benchmarks, and how the business model of benchmark organizations charging model vendors to verify results affects credibility.

    Same story · 3 reports

    #English text is being prepared.#Evaluations / benchmarks#Industry news