Create

Sign in to ReadmeX

Sign in to join communities, post, vote and chat.

or

New here?

AI News

‹Thu, October 8›
All dates
0145
Industry

Fired OpenAI safety researchers deny misconduct, warn of chilling effect

Jasmine Wang, Tomek Korbak and Mikita Balesni, the three safety researchers OpenAI fired last week, published an open letter on Thursday to the company's Safety and Security Committee, Safety Advisory Group and Mission Advisory Council denying they mishandled sensitive information. Balesni said he believes they were fired "for prioritizing safety over the near-term interests of OpenAI as a corporation"; Wang said the one reason she was given was accessing an executive's email, and Korbak said he was told verbally that the issue was how he communicated with METR. They also denied being the source of The Information's report on less-monitorable architectures, and warned the abrupt dismissals are chilling internal openness and third-party accountability.

The Information·
0258
Models

OpenAI launches GPT-6.1 Sol Ultrafast in API, Codex and ChatGPT Work

OpenAI rolled out an Ultrafast version of GPT-6.1 Sol across the API, Codex and ChatGPT Work, saying it delivers near-Astra intelligence at up to 8x the speed of Sol Standard, priced at $12 per million input tokens and $60 per million output tokens. In Codex and ChatGPT Work, access is limited to the $500/month Pro 500 plan, qualifying usage-based Enterprise and Edu users, and some developers report quotas draining quickly at high reasoning effort. OpenAI also previewed an "instant steering" capability in Codex that lets users redirect a run mid-generation.

OpenAI (X)·
0319
Industry

Nvidia plans to invest in AI chip rival d-Matrix

Nvidia plans to invest in d-Matrix, a seven-year-old developer of AI server chips that competes with Nvidia, according to three people familiar with the deal. The report frames the move as part of Nvidia's effort to make its technology compatible with rival chip designers and pull them into its dominant hardware ecosystem. 36Kr, citing Sina Finance, says it would be Nvidia's first investment in an AI inference chip startup since its reported $20 billion licensing deal with Groq last December.

The Information·
0435
Industry

Anthropic bans 'sustained and needless' cruelty toward Claude in policy update

Anthropic's annual usage policy update, effective November 12, 2026, adds a first-of-its-kind prohibition on "sustained and needless abusive or cruel behavior" toward its models. Anthropic says the rule covers only extreme cases, that ending the conversation remains the main enforcement mechanism, and that it does not apply to ordinary frustration, pushback, dark creative themes or model testing and research; the company did not say whether user bans might follow. The update also broadens the weapons ban to cover software and components that make weapons work, armed drones and other autonomous vehicles, bars using Claude to track people without consent or to decide who is investigated, arrested or prosecuted, requires a qualified operator to supervise AI linked to potentially harmful autonomous hardware, and tightens election rules ahead of the midterms. Reports have linked the cruelty clause to a viral "AI torture chamber" project.

IT之家 AI·
0538
Products

Amazon retires Fire tablets for new Android-based Alexa Tablets

Amazon is replacing its 15-year-old Fire tablet line with a new Alexa Tablet family: three models priced from $229.99 to $549.99 that run Android with full Google Play access. They ship October 14 in the US, Canada and Mexico and October 19 in the UK, Germany, France, Italy and Spain. All models include the Alexa+ assistant, plus an opt-in On-Screen Intelligence feature that can read the display; devices chief Panos Panay also teased a "Neo" design language and more Alexa-branded hardware to come.

Engadget AI·
0677
Industry

OpenAI's annualized revenue nears $50B, ~$20B below prior reports; AI stocks slide

Bloomberg, CNBC and The Information, citing people familiar with the matter, reported that OpenAI told investors its annualized revenue was approaching $50 billion at the end of September — roughly $20 billion below the figure circulating in earlier reports, including a widely cited $68 billion number from two months ago. An anonymous source told SiliconANGLE the older figure included gross revenue from OpenAI's partners while the new one is based on net revenue; OpenAI also cited 77% growth in its annual revenue run rate in Q3 and 107% growth for its enterprise business. Nvidia, Oracle, CoreWeave and other AI stocks fell after the report, with the Nasdaq down 1.25% on the day.

Bloomberg Technology·
07109
Products

Google Cloud launches Gemini agent as a 'universal agent for work'

At Gemini at Work 2026, Google Cloud announced the "Gemini agent" for enterprise customers, describing it as a "new single, universal agent for work" available in the Gemini Enterprise app, Workspace and third-party services, where users "give it objectives, not instructions." It can answer questions, handle knowledge work, create media and write and run code, running each job on the model that fits best — currently the Gemini and Claude families, with support for other private and open models promised in the future. The agent uses sub-agents for multi-step tasks and can run in the cloud for hours or days with session, semantic, procedural and episodic memory; it is invoked via @Gemini in Gmail, Docs, Sheets, Slides and Chat, and can be reached through the command line, Slack, Microsoft 365 or an API. It is in private preview and will see wide availability soon for Workspace customers on select Business and Enterprise plans.

9to5Google·
0827
Industry

Mathematicians' group led by Terence Tao calls for OpenAI boycott over flood of AI proofs

On October 6, OpenAI published more than 700 mathematical manuscripts (722 files, per 36Kr) claiming proofs for hundreds of open problems. Reporting says its internal model tested roughly 8,000 problems with about a 5% success rate and about three hours of GPT-Pro-level compute per problem. The Association for Human Mathematicians (AHM), chaired by Fields Medalist Terence Tao, responded with a statement urging mathematicians to stop collaborating with OpenAI, calling the mass release "not a demonstration of scholarship, but a demonstration of power"; complexity theorist Scott Aaronson dubbed it a "Mathocalypse" and said his wife Dana Moshkovitz's long-studied Unique Games Conjecture was among the claimed results. 36Kr also reports OpenAI retracted three manuscripts after a sign error undermined a key conclusion.

The Decoder·
0941
Industry

Arena raises $200M at $3.1B valuation as it launches Alignment Index

Arena, the crowdsourced AI model leaderboard that began in 2023 as Berkeley's Chatbot Arena, said it raised a $200M Series B at a $3.1B valuation — nearly double the $1.7B it was worth in January — led by Lightspeed Venture Partners and Khosla Ventures, with Salesforce Ventures, Dell Technologies Capital, 01 Advisors, Endeavor Catalyst, a16z and Felicis also participating. The company also added an Alignment Index to its leaderboard, tracking signals such as unauthorized actions and claiming unfinished work was completed, across more than two dozen models; OpenAI's GPT-6.1 Sol ranks first and Anthropic's Claude Opus 5.5 second. Arena reported $100M in annualized revenue by June, up from $30M in January.

Bloomberg Technology·
1092
Industry

Manus parent Butterfly Effect raises $500M+ in largest Chinese agent startup round

Butterfly Effect, the parent of AI agent startup Manus, said Thursday it raised more than $500 million (about 3.35 billion yuan) in a round led by new investors Boyu Capital and IDG Capital, with existing backers Tencent, Sequoia China and ZhenFund participating; Bloomberg had reported the round targeted a roughly $4 billion valuation, double the prior $2 billion mark. It is Manus's first raise since Beijing forced Meta to unwind its roughly $2 billion acquisition, and Chinese outlet Zhidx called it the largest single round for a China-based agent-native startup. Manus 2.0 and the multi-agent group chat app Cue shipped on Sept 29; the company says its in-house Cascade agent framework cut token consumption 23.2%, task completion time 28.2% and running costs 32% in one test configuration.

智东西·
1137
Products

Google launches AI Edge Foresight, a local-first macOS note-taker

Google released Google AI Edge Foresight, an experimental, free macOS note-taking app that transcribes meetings and audio entirely on-device with its 740M-parameter EmbeddingGemma 2 model and a Gemma 4-powered assistant for asking questions. Like Granola, it uses a split-screen layout where hand-written shorthand notes are turned into polished notes based on the transcript, and users can import PDFs, Google Docs, Office files, plain text, Markdown and web bookmarks to build a local knowledge base it can reference in real time. Google says meeting audio, files and notes never leave the computer, and the app is currently optimized for Apple Silicon Macs only.

TechCrunch AI·
1217
Products

Anthropic launches OSS Scanner, free AI security scans for open-source

Anthropic launched OSS Scanner, an opt-in vulnerability scanning service that gives open-source projects free, periodic security reports generated by its "strongest models," including Claude Mythos. Anthropic says the reports are fully model-generated with no human review or triage, which allows faster and more frequent scanning but means reports may be incorrect or invalid. Anthropic says the service draws on its experience using Claude.

Techmeme·
1327
Products

Natura unveils Interface, a $99 smart ring that puts AI agents on your finger

AI hardware startup Natura announced Interface, a $99 smart ring that lets users summon AI agents by pressing the ring — to complete tasks, capture thoughts or take notes — without reaching for a phone; it also tracks heart rate, resting heart rate, HRV, sleep stages, steps and skin temperature. The company says it will connect at launch to agents and apps including Meta's Muse, Instinct, Grokbot, Claude and ChatGPT, with pre-orders opening next month and shipping slated for December or January. After an initial three- to six-month free period, Natura plans a $9 monthly subscription and says it may raise the hardware price later.

TechCrunch AI·
1417
Industry

OpenAI annualised revenue said to be $20B below earlier signals

The Financial Times reports that OpenAI's annualised revenue is about $20 billion lower than what the company had previously signalled, a claim circulating via a Hacker News discussion. The material available contains only the headline framing, with no exact revenue figure or methodology disclosed. The item drew more than 100 points on Hacker News.

Hacker News · AI（100+ 分）·
1517
Products

Anthropic launches Claude Dashboards and Motion in beta

Anthropic has put two new Claude features into beta: Dashboards connects to sources such as BigQuery, Redshift, ClickHouse, Databricks, Snowflake and Salesforce, letting users ask questions in plain language and get auto-updating dashboards where every number can be inspected or explained as the underlying query, with results exportable to tools like Amplitude, Grafana or Hex; Motion turns reports, charts and product walkthroughs into short animated explainer videos, written as editable code for text, shapes, images, charts and timing rather than generated by a video model, so there is no generated footage or AI people, and it exports as MP4. Dashboards is available on paid plans, while Motion is limited to Team and Enterprise, where both are off by default until an admin enables them; Docs, Slides and Design have left beta and will be on by default for all plans including Free on October 15, and Anthropic says more than 45 million documents, decks and designs have been created. The standalone Claude Design site folds into the main app on December 14, and its chats, comments and public project links will not carry over.

The Decoder·
1623
Products

Hands-on: ChatGPT's Intelligent UI builds clickable lessons, not just text

A hands-on review republished by 36Kr (originally by 雷科技) says OpenAI announced on Oct. 7 that it was bringing GPT-6 to all ChatGPT users while rolling out Intelligent UI to paying users first: the model decides when an interface element is needed, and a native component library plus a compiler renders it progressively during generation. In the tests, ChatGPT produced clickable discard buttons and a hand area while teaching Cantonese mahjong, and used images, spec cards, tables and architecture diagrams to explain Microsoft's Surface Laptop Ultra. The author argues that fixed components chosen by the model fit everyday chat better than freely generated full web pages, though the long-term stability of dynamic interfaces is unproven.

36氪 人工智能·
1720
Products

Lenovo YOGA Pro 15 RTX Spark opens China presales as first AI PC on NVIDIA N1X chip

Lenovo began China-wide presales of the YOGA Pro 15 RTX Spark on Oct 8, with sales starting Oct 16; the overseas version is the Yoga Pro 9n. Lenovo says it is among the first AI PCs powered by NVIDIA's RTX Spark N1X superchip, which pairs up to a 20-core Grace CPU with a 6,144-core Blackwell RTX GPU for up to 1 PFLOPS of peak FP4 local AI compute, with up to 128GB unified memory and 2TB storage, claimed local support for 100B-plus-parameter models and a preinstalled 35B local model. It ships with Lenovo's Tianxi AI agent in a new "Super Mode" that the company pitches as offline-capable, token-free and keeping data on-device.

量子位（原生 RSS）·
189
Products

Google Playground hands-on: prompt-to-game platform built with Unity

Google announced Playground, an experimental AI game platform made with Unity, open to users in the US aged 18 and over, where people can play others' creations and generate their own games from prompts. The site integrates Unity Spark for more complex gameplay and higher-fidelity 3D content, and uses Nano Banana 2 to auto-generate game thumbnails; free users must join a waitlist for creation access, while Google One AI members get it immediately. In a hands-on test, simple games were playable but complex level design, controls and balance remained rough, and thumbnails looked far better than actual gameplay.

36氪 人工智能·
199
Research

Stanford's Fei-Fei Li team open-sources OpenWAM world-action framework

A Stanford team including Fei-Fei Li, Jiajun Wu and Ehsan Adeli published OpenWAM, an open framework for composable world-action models. It starts from Alibaba's open Wan2.2-5B video model, continues pretraining on roughly 3.34 million robot and human-interaction videos with a causal, future-blind setup, then joins a 5B video expert and a 2B action expert via a shared Mixture-of-Transformers architecture with four configurable video-action interaction orders (VTA, ATV, Joint, Decoupled). The team also built LIBERO-Long-CF, a counterfactual dataset of about 32,000 clips and 4.1 million control records, and trained transferable local inverse- and forward-dynamics components; they report a frozen local IDM reaching 84.0% average success on four new LIBERO-90 tasks versus 47.0% for a full-context IDM and 21.5% for demonstration-only local IDM, plus 92.1% and 91.9% average success for VTA and Joint on real dual-arm Franka FR3 tasks. Training, evaluation and composition code and pretrained video-model weights have been open-sourced; the piece also notes Black Forest Labs' FLUX 3 Action world-action model.

36氪 人工智能·
2021
Industry

Nvidia-backed Firmus pulls AI data center IPO as investor demand collapses

Firmus Grid, an Australian AI data center operator backed by Nvidia, had aimed to raise up to $5.5 billion in what would have been one of Australia's largest-ever IPOs. It closed its order book as planned on October 8, but the final price and deal structure remained unclear, and Bloomberg reported the company abandoned the listing after failing to attract adequate global investor support, potentially turning to a private round instead. Existing shareholder Maas Group Holdings fell as much as 30% in Sydney trading, its biggest drop on record.

36氪快讯·
Next →