Models, products and AI startups, near-duplicates collapsed to the most credible source
Researchers Turn 109k ML Papers Into Topic PagesA team grouped roughly 109k papers accepted at ICML, NeurIPS, ICLR, CVPR and AAAI since 2016 into about 1,260 topics and extracted results tables from the full text of 79k of them. Topic pages show best benchmark numbers with the source paper, recent ablations and common baselines. It is free to use; the authors warn extracted numbers may not be fully comparable.
Micron AI memory sold out through 2026 with $100B in contractsAmid bubble fears and a broad AI sell-off, The Motley Fool highlights Micron, Astera Labs and Vertiv. Micron's high-bandwidth memory is sold out through 2026, with roughly $100 billion in take-or-pay contracts and $22 billion in customer deposits, yet it still can't meet over half of demand; the other two posted strong growth and raised guidance. All carry risk if AI capex slows.
Preprint probes how source attribution shifts LLM argument ratingsAn unreviewed preprint tested 20 arguments under four source conditions across GPT-5.6 Sol, Claude Opus 5.5, Gemini 4 Argon and DeepSeek Flash. Credibility was most source-sensitive, several models showed an irrelevant-expertise penalty, and agreement shifted less.
Alibaba Qwen Code Repo Reportedly Leaks qwen4-max and Other Model IDsA Reddit user reports that pull requests in Alibaba's Qwen Code repo reference model IDs including qwen4, qwen4-max, qwen4-coder-plus and qwen4-omni-120b-think, with 262,144-token context. The claim is not officially confirmed.
Blackstone, Pimco chiefs debate perils of the AI boomAt the HKMA Global Financial Leaders' Investment Summit, Blackstone President Jon Gray and Pimco CEO Emmanuel Roman debated whether global credit markets can absorb the tech industry's multitrillion-dollar AI spending spree.
Google brings Dreambeans to India with support for 13 languagesGoogle has expanded its experimental AI app Dreambeans to India, its first market outside the US, adding support for 13 local languages including Hindi and Tamil. The app draws on Google services like Gmail, Calendar and Photos to auto-generate personalized illustrated stories. Sensor Tower estimates about 55,000 US downloads since launch, with October downloads down 84% month over month.
Wall Street chiefs see robust demand for massive AI dealsTop Wall Street executives said capital markets will keep financing the global AI infrastructure buildout, with Morgan Stanley projecting cumulative AI-related spending could reach $15 trillion over the next decade.
State Street CEO: hard to see AI as a bubble yetState Street Chairman and CEO Ron O'Hanley said it is "hard to see" AI as a bubble yet, with demand supporting the massive investments being poured into the technology.
DBS CEO on how Southeast Asia's biggest bank deploys AIDBS Group CEO Tan Su Shan discussed the bank's aggressive AI adoption, said the Binance hack shows the need for tighter cyber hygiene, and noted growth in wealth flows in Singapore and Hong Kong in an interview with Bloomberg.
Apple hints at earning more from third-party AI appsOn its earnings call, Apple said AI could open revenue opportunities through third-party apps and partnerships, without specifying how. Its services business is at record levels but under pressure as the App Store faces regulation, weaker mobile gaming and lower commissions, with estimated U.S. commission revenue down 18%.
User mines 16 years of city council notes with local Qwen modelA user fed 16 years of city council minutes into a locally hosted Qwen 3.8 27B model, finding the police chief obtained a $9,000 grant for Flock cameras, with no public comment recorded. The user seeks a way to scrape YouTube auto-transcripts of meetings.
YC's Garry Tan spends up to $200K a year on AI tokensY Combinator's Garry Tan said on a podcast that he works so intensively with AI agents that he jokes of "cyber psychosis" and barely sleeps, spending up to $200K a year on tokens.
OpenAI Executive Urges Caution in China-US AI Talks, Warns of DistillationAn OpenAI executive urged caution in China-US AI talks and warned about distillation, where rivals train on a model's outputs, the South China Morning Post reported. The remarks highlight technical-protection concerns in discussions between the two countries.
NCSC: AI malware likely to hit UK infrastructure by 2028The UK's National Cyber Security Centre annual review says AI-powered malware is highly likely to be used in ransomware attacks on UK critical national infrastructure by 2028, warning Russia and China will routinely use AI in cyber-attacks.
OpenAI appoints Jill Hazelbaker as Chief Corporate Affairs OfficerOpenAI announced that Jill Hazelbaker will join as Chief Corporate Affairs Officer, leading global communications and corporate positioning. The appointment signals OpenAI's push to strengthen external messaging and brand building.
CWE-bench v2 releases 100 security tasksCollinear AI released CWE-bench v2, a public set of 100 tasks covering 91 CWE types that tests agents on finding and fixing vulnerabilities in real open-source repos without breaking functionality. Scoring is fully programmatic and reproducible, and a held-out set tracks progress on tasks no model has seen.
Transformers adds small model fixtures to catch regressionsHugging Face describes how Transformers runs integration tests on CPU with small fixture models—the UMT5 fixture uses 66 MB versus 1,227 MB for the source checkpoint. The approach follows a real incident: after SDPA was enabled in July 2026, UMT5's decoder could attend to future tokens, and the bug stayed in main for nearly two months.
Efekta joins UNESCO AI education observatoryOn 26 October 2026, UNESCO's Santiago office signed a cooperation agreement with Efekta Education Group, part of EF, which joins the Latin America and Caribbean Observatory on AI in Education to promote responsible, equitable AI governance in education.
States urge Congress not to cut cybersecurity fundingTech officials from Florida, New York, and Tennessee urged Congress to reauthorize the State and Local Cybersecurity Grant Program, warning that AI is making attacks faster and more sophisticated. Tennessee said grant funds helped secure nearly 90,000 endpoints across local government networks.
Trajectory raises $45M Series AOne-year-old Trajectory raised a $45M Series A at a $300M valuation, led by Sequoia with Nvidia, Bessemer and Bain Capital Ventures. Its continual-learning platform retrains open source models from user feedback, letting legal and finance experts improve specialized models without research teams; Harvey and Rogo already use it.
User says Qwen 3.8 27b used Blender to do mathA LocalLLaMA user said that when given a math-heavy physics problem, Qwen 3.8 27b found it could run Python via Blender MCP and wrote and executed code to speed up its math. The user called the behavior unprecedented for a local model and warned that models need explicit limits.
Op-ed urges SEC to block Anthropic's IPOA Tech Policy Press op-ed calls on the SEC to block Anthropic's IPO, citing footnotes showing nearly $900 billion in spending commitments and unaudited 2026 results, with third-quarter revenue growth slowing sharply—risks it says would fall on retail investors and retirees.
Study questions LLMs' ability to read neural net weightsA paper says GPT-6 Astra can directly inspect and edit small-network weights, notably improving a 0.8B Qwen model's math. A critic argues human researchers could do the same tasks and that labeled input channels make the results far less superhuman than claimed.
Open-source research proposes new sparse attention for ultra-long contextAn open-source research team shows attention can be sparsified on Qwen 3.8 27B without observed accuracy loss, and that extending context to 10M tokens hurts some tasks but leaves most, like NIAH, fine. It says sparse attention can be trained natively on next-token loss, matching full-attention validation loss faster; kernels, SGLang integration and the paper are open.
Anthropic warns it is near recursive self-improvement capabilityIn its revised S-1A filing, Anthropic warned it may soon meet its internal definition of recursive self-improvement (RSI), where AI trains more capable AI and sharply accelerates progress. It said it has already used AI to develop models whose capability leaps it did not recognize at the time, and flagged the risk under its Responsible Scaling Policy.
US appeals court: law firm's AI use doesn't warrant cutting legal feesA US appeals court ruled that a law firm's use of AI tools does not by itself justify reducing its legal fees, rejecting a bid to cut the fees. The decision offers a judicial reference point for charging for AI-assisted legal work.
Isomorphic Labs president: AI will design all drugs one dayIsomorphic Labs president Max Jaderberg said AI will be used to develop all medicines in the future, with only the timing uncertain. The remark highlights ambitions in AI drug discovery but offers no concrete timeline.
OpenAI, Anthropic set up war rooms to curb election misinformationAhead of US midterm voting, OpenAI and Anthropic set up dedicated war rooms to monitor misuse, while Meta activated its Election Operations Center. A Brennan Center report found ChatGPT significantly improved election-disinformation protections since May, though several other image generators still produced realistic fake election images.
Palantir lifts full-year revenue outlook on strong commercial salesPalantir raised its full-year revenue forecast above Wall Street expectations, driven by surging commercial demand for its analytics tools that help companies use AI for operational decisions.
OpenAI Fundraising Stalls, Valued at $1.25T Below TargetOpenAI raised less than half of its $30 billion fundraising target, with an estimated valuation of $1.25 trillion, below its $1.4 trillion goal and behind Anthropic. Investor concerns over AI industry debt and profitability have made fundraising difficult.
The A.I. Daily