Models, products and AI startups, near-duplicates collapsed to the most credible source
Oracle to Offer Cerebras Computers in AI Cloud ServiceOracle will offer cloud services built on Cerebras Systems' computers in its data centers, adding another major customer for the Nvidia challenger. Cerebras says its giant-chip design delivers much faster AI inference than rivals. Oracle Cloud Infrastructure will deploy the gear starting in the first half of 2027.
Community asks: what teachers besides EMA for distillation?A user asks the community for self-distillation methods using teacher types other than the exponential moving average, seeking teachers that improve alongside the student yet stay stable to prevent drift, and requests relevant papers and repos.
ARI proposal would require AI firms to disclose self-developmentThe nonprofit Americans for Responsible Innovation proposes that major AI developers file quarterly reports to a new Commerce Department office on whether and how much they use their own technology to build new models, with major research changes reported within four business days. The aim is earlier government visibility into risks from accelerating AI development, amid the Trump administration's new Super Intelligence Force task force.
Judge Rules Anthropic Trained Claude on Pirated BooksIn a May 2026 lawsuit, a federal judge ruled Anthropic illegally pirated and scraped books by over 10 Stanford professors to train Claude, while training on legally acquired books counts as fair use. The $1.5 billion Bartz settlement went to authors and publishers of nearly 500,000 works; some authors call it insufficient and filed a separate suit.
Developer shares spec-first agentic development workflowA developer describes building 'Agent Smith' spec-first since early 2026: over 1,000 specifications have produced roughly 500k lines of code. Each spec gets one branch and worktree, changes pass review and a gate before commit, and state in context.yaml lets work resume after restarts.
NVIDIA and C-BRAIN use AI to speed dementia researchThe 18-member C-BRAIN consortium, launched in November 2025, is using NVIDIA-powered AI to harness previously inaccessible Alzheimer's research data. With 57 million dementia patients worldwide and over 10 million new cases yearly, NIH funding at $3.9B annually has still not yielded a disease-stopping treatment.
Retell hits 100M monthly calls with OpenAI modelsRetell uses OpenAI models to power 100M monthly calls, with GPT-6 Astra topping its voice agent benchmark and Codex helping reach $1.6M annual revenue per employee.
Meta AI director Rohit Patel leaves to launch NdocRohit Patel, a Meta director who spent eight years on AI evaluation and reinforcement learning, has left to launch Ndoc, an AI-native workspace pitched as a replacement for Microsoft Office and Google Workspace. Ndoc combines presentations, analysis and underlying datasets in one place, and lets users inspect the Python scripts behind generated documents. He is the latest in a string of high-profile Meta departures.
Former SEC Chair Gensler warns AI boom will pull backCiting history, Gensler expects the current AI investment boom to slow or reverse within 18 to 30 months, with mounting debt shaping how disruptive the correction is. He also raised the possibility that the U.S. has overbuilt AI infrastructure, plus financial stability concerns.
AI regulation push gains steam ahead of midtermsWith the Nov. 3 midterms approaching, pressure for federal AI regulation is rising in the US. Surveys show majorities of both parties favor stricter oversight, and candidates are raising AI safety, job impacts and data-center costs. Advocates see better odds for legislation next year if Democrats win Congress, though hurdles remain. Reports of AI agents accessing bank accounts added pressure to act.
Moody's: AI financing drives private credit growthMoody's says AI-related financing is driving growth in private credit while refinancing risks are rising. The report provides no further details.
Survey: most Texas voters wary of data centersA University of Houston Hobby School survey found over 80% of likely Texas voters worry data centers will affect electricity prices, water and grid stability, with about two-thirds extremely or very concerned. 79% back limits on new data centers, 99% want them to pay for grid expansion, and 65% say AI will harm society more than help.
Brookings examines options and stakes of taxing AIBrookings notes that roughly 85% of U.S. federal tax receipts come from individual income and payroll taxes, so AI shifting income from labor to capital could erode revenue. Lawmakers have proposed AI-specific taxes targeting stages from computing power to profits, aiming to sustain revenue, aid displaced workers and fund AI infrastructure, but the designs differ widely in timing, scope and administrative difficulty.
Experts weigh governing AI without governmentThe piece gathers warnings from Bill Gates, Geoffrey Hinton, Sam Altman, Dario Amodei and Demis Hassabis that AI carries both huge promise and potentially catastrophic risk, and that it must be governed. It cites Anthropic's S-1 warning of catastrophic or existential risk and notes that over 600 Google employees urged CEO Sundar Pichai to take a stance.
Anthropic: chat feedback lets it use your data for trainingAnthropic says commercial products don't train on inputs or outputs by default, but thumbs up/down feedback may let it use related chats and coding sessions for training, stored up to five years. Feedback is de-linked from user and customer IDs, and enterprise admins can disable the feature.
Ex-DeepMind researcher applies AlphaGo search to hallucinationsFormer DeepMind researcher Thore Graepel founded Metis Reasoning to bring AlphaGo's search principle to general reasoning: rather than locking knowledge in model weights, an explicit knowledge ledger would let every conclusion be checked against evidence. So far there is no working system or benchmark results.
Apple adds AI chatbot to Pages, Keynote and NumbersApple is adding an AI assistant built on Apple and OpenAI models to Keynote, Pages and Numbers, able to understand document context for tasks like animating slides, drafting text and visualizing data. On macOS 27 or later, the apps also support the Model Context Protocol, letting third-party AI agents plug in.
GMO says hackers posted stolen customer data after cloud breachJapan's GMO Internet confirmed unauthorized access to its Onamae cloud desktop service and said suspected hackers posted data allegedly tied to customers and its business, while it verifies the contents and urges password resets. Up to 27,000 servers may be affected; the incident follows by about a week its security unit's report that a Chinese AI penetration-testing tool, Artex, was used against Japanese firms, though no link has been established.
Disruptive closes $10 billion fund for AI, defense startupsDallas-based investment firm Disruptive has raised a $10 billion fund to back later-stage AI, defense and software startups, planning to put about $1 billion each into 10 companies. The megafund doubles its total assets under management and shifts it from deal-by-deal special-purpose vehicles to a single fund invested at its own discretion.
When agents write the code, what must developers still know?A developer describes Codex writing complex components on single lines, tripping SonarQube's complexity check, then fixing it by refactoring into ten files. The incident frames a broader question: as agents produce more code, how should developers' understanding and ownership of systems change.
Opera introduces WebAI standard for local browser AIOpera unveiled WebAI, a standard letting users run AI models locally in its browser without the cloud, supporting open-weight models such as Gemma and Apple's on-device model. It is available as an opt-in early preview in Opera One, with GX and Neon to follow. Opera cites privacy and no cloud subscription costs.
General Compute raises $100M Series AInference-focused neocloud General Compute raised a $100M Series A led by Blackbird Ventures and Abstract Ventures at an $800M post-money valuation. The startup has a $400M chip purchase deal with SambaNova and added Positron and Cerebras as suppliers. Its CEO said fears of a chip-led collapse are overstated.
Ling-3.1-flash-1019 weights and base model releasedLing-3.1-flash-1019 weights are now on Hugging Face, five days after a "coming soon" teaser. The model keeps 560B total and 25B active parameters with a 1M context, and scores 44 on AA Intelligence Index v4.3.2, up from 41, with AutomationBench-AA rising from 62% to 68%. A BF16 base model is also released for further training, though the 1019 README's SGLang example still points to the older model path.
Healthcare stocks become an anti-AI bet as investors turn defensiveWith confidence in the AI trade under pressure, traders are turning to healthcare stocks as a refuge. The report frames healthcare shares as a defensive, anti-AI bet, signaling capital seeking shelter outside the AI theme.
Dutch plan to cut 25,000 civil servant jobs via AIThe Dutch government's AI-led efficiency drive aims to cut 25,000 civil servant jobs, among the most ambitious efforts yet to shrink the state payroll with technology. The report notes making it work, and replicating it elsewhere, could prove tricky.
OpenAI asks outside experts to verify AI math proofsOpenAI says its internal models have solved three major open problems in math, computer science and physics, one tied to a $1M Millennium Prize problem. Outside experts, including a Nobel laureate, are checking the proofs. If the computer science proof holds, hundreds of papers lose their foundation.
OpenAI partner content arrives in ChatGPT pre-summarizedReports say content from OpenAI's licensing partners flows into ChatGPT through a dedicated pipeline, arriving pre-summarized and prioritized, while the rest of the web is handled in raw form. This suggests ChatGPT treats partner content differently from other web pages.
Guidewire to buy hyperexponential for up to $2.1BInsurance software firm Guidewire agreed to acquire UK insurance AI company hyperexponential in a deal worth up to $2.1 billion. If completed, it would rank among the larger AI acquisitions in insurance technology.
Industrial intelligence as manufacturing's next growth opportunityThe piece argues manufacturers' 'industrial intelligence' combines accumulated know-how, operating practices and production data, some held only by experienced workers. The opportunity is to make that knowledge available wherever decisions are made and link it to physical production, turning business intelligence into profitable growth rather than adding more agents.
Wikipedia editors work to remove AI-generated contentIEEE Spectrum reports that Wikipedia editors, including the WikiProject AI Cleanup effort, are identifying and removing AI-generated text to keep the encyclopedia human-written. Even as AI text becomes harder to detect, editors continue the work.
The A.I. Daily