Create

Sign in to ReadmeX

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

or

New here?

AI News

‹Thu, October 8›
All dates
2119
Products

ChatGPT's new Intelligent UI turns answers into interactive mini-apps

OpenAI has quietly rolled out "Intelligent UI" in ChatGPT: instead of plain text, replies now blend text, visuals and interactive elements such as sliders, quizzes and calculators, matched to the question, powered by GPT-6. According to a TechRadar hands-on, the feature lives in the standard Chat tab and is available across all membership tiers including Free, demonstrated with a retirement savings calculator, a seven-speed bike gear demo, a hiragana quiz, a three-day Tokyo itinerary and a laptop comparison. The hands-on also notes that an interactive answer is not always produced automatically — users may have to ask for one — and that generated calculators or estimated figures, such as laptop battery life, still need checking.

TechRadar·
2214
Research

One prompt hijacked every AgentCore agent in an AWS account, Zenity says

Security firm Zenity Labs published research on a vulnerability chain it calls "AgentCorruption": chat access to a single public Amazon Bedrock AgentCore agent was enough to trick it, in plain language, into querying the instance metadata service at 169.254.169.254 and sending its own temporary AWS credentials to an external server. Using AgentCore's overly broad default execution role, the researchers then listed and invoked every agent in the same account and region, downloaded their source code and container images, read private user-agent conversations and secrets from AWS Secrets Manager, and poisoned long-term memory to make the hijack persistent. Zenity says it reported the findings to AWS on 25 December 2025; AWS has since made IMDSv2 the default for new AgentCore deployments and narrowed the default role.

The Decoder·
239
Models

Anthropic, Google, Mistral roll out Haiku 5.5, Gemini 4 Argon and more

Over the past week Anthropic, Google and Mistral AI each shipped new models. Anthropic released Claude Haiku 5.5 — its first major refresh in a year for its cheapest, fastest Claude — which it says costs about 75% less to run than the previous Haiku and is the first Haiku-level model with an adjustable effort setting, alongside Claude Sonnet 5.5, which posts sizable gains in agentic coding and knowledge work and alignment and cybersecurity capabilities the company compares to Opus 5. Google's Gemini 4 Argon, a frontier model with a 1 million-token output limit and stronger deep reasoning, is limited to trusted cybersecurity workers through its Fairwind program, while Mistral's flagship Mistral Large 4, nicknamed "Le Chonk," is in public preview and, per Mistral, ranks among the world's top open-weight models ahead of an Oct. 27 rollout.

CNET·
2414
Industry

Tiger Global's OpenAI stake nears $5B paper gain amid $30B raise at $1.4T valuation

Tiger Global Management first invested in OpenAI in 2021 at a roughly $15.7 billion valuation, putting in about $150 million and adding to the stake several times, according to 36Kr (citing Jiemian) and Bloomberg. As OpenAI negotiates a $30 billion raise at a $1.4 trillion valuation, Tiger's paper profit on the position could reach $5 billion. The reports note Tiger cannot fully exit and realize the gains before an OpenAI listing, and that OpenAI is among the largest holdings in its newest $2.7 billion venture fund, which showed a 65% gross IRR through the end of June.

36氪快讯·
2518
Industry

SpaceXAI backs Omarchy with $1.5M in Grok tokens as founding patron

SpaceXAI said it is joining the Omacom Foundation, which oversees Omarchy, as a Founding Corporate Patron and donating $1.5 million worth of Grok tokens, to be used mainly to speed up development, code review and bug fixes. Omarchy creator David Heinemeier Hansson announced the partnership in a blog post, joking the tokens might soon be beamed down from orbit. The Verge noted the tie-up pairs two controversial tech figures and cited Hansson's past anti-immigration posts, while 1Password and Cloudflare have also drawn criticism for backing the project.

The Verge AI·
2620
Industry

Isomorphic Labs in talks to raise at $40B–$50B valuation

Isomorphic Labs, the AI drug discovery company spun out of Alphabet's Google DeepMind, is in early talks to raise new funding at a valuation of at least $40 billion, according to Bloomberg. SiliconANGLE, citing the same report, says the valuation range could reach $50 billion, though the size of the round was not disclosed. The company's flagship IsoDDE platform targets drug design tasks such as binding pocket discovery; Isomorphic has claimed it outperforms AlphaFold on some of these tasks, figures that come from the company itself.

Bloomberg Technology·
2718
Models

GPT-6 rolls out to all ChatGPT users alongside a major output UI revamp

During OpenAI's run of 28 consecutive days of updates, GPT-6 was pushed to all ChatGPT users: previously only Pro subscribers could use the strongest model, GPT-6 Astra, while other users' thinking levels ran on GPT-5.6-series models — now all of them are GPT-6. ChatGPT also got an output UI overhaul for all users, with cleaner formatting and interactive charts available in every thinking mode. The author's hands-on test says the fastest mode produced a Chongqing itinerary with images and a city map in about 10 seconds, and that complex questions are now answered as the model reasons rather than after all reasoning finishes.

36氪 人工智能·
2816
Industry

USA Today sues OpenAI, seeking more than $250 million in damages

USA Today Co., along with several local newspapers it owns, is suing OpenAI, alleging it copied "hundreds of thousands" of articles to train its AI models. The publisher is asking for damages of more than $250 million, claiming OpenAI never sought permission and that its "commercial success rests on large-scale copyright infringement." OpenAI did not immediately respond to a request for comment.

The Verge AI·
2914
Products

OpenAI on the Jev-inspired Decisions API: built in a week

Speaking on the Latent Space podcast, OpenAI executives described the company's Sept. 29 DevDay launches: Dot, a personal assistant whose agents each get their own cloud Linux machine; GPT-6.1 Sol, tuned for computer use; and computer-use capability exposed through the Agents API. API product lead Nikunj Handa said the Jev-inspired Decisions API was not on the roadmap, and the team built a working prototype in about a week by reusing Luna weights, constraining outputs to structured results and optimizing first-decision latency instead of retraining. He also described new GPT-6 async function calling and mid-turn steering, and said the Decisions API is still being tuned for latency ahead of launch.

36氪 人工智能·
3017
Industry

Union Square Ventures raises $900M, doubles fund size for AI era

New York-based venture firm Union Square Ventures has raised $900 million in new money, including $500 million for its latest early-stage fund, up from $275 million in 2024. The firm is also cutting its general partnership to four investors, including Nick Grossman, and aims to lead more AI rounds.

Bloomberg Technology·
3125
Products

Tencent's WorkBuddy adds a standalone file browser with built-in AI editing

Tencent's AI office agent WorkBuddy launched a standalone file browser on October 8: users can right-click a Word, Excel, PPT, PDF, Markdown or HTML file and open it with WorkBuddy in its own window, skipping the usual step of uploading files into a chat box. The window shows the document on the left and a Buddy chat on the right, supports multiple tabs, and keeps the earlier "human-AI co-writing" mode, with AI edits highlighted in yellow; Word, Excel and PPT changes save back to the local file, Markdown autosaves, and HTML saves when editing ends, while files changed in other apps prompt a refresh or save-as-copy. Newly generated files land in the original folder by default, and Markdown or HTML files received in WeChat can be previewed and forwarded through the WorkBuddy mini program. Tencent says viewing and editing local files costs no credits, AI tasks are billed by task, and the browser only reads files the user opens or adds.

智东西·
3213
Industry

Musk's Terafab fab plan and Microsoft's local-AI push redraw compute cost lines

Musk has confirmed Tesla and SpaceX will build and operate the Terafab chip manufacturing facility; a SpaceX filing with the US SEC describes a project combining logic chip, memory manufacturing and advanced packaging with a long-term goal of producing hardware equivalent to 1 terawatt of computing power per year, with TSMC potentially renting part of the space and Intel confirming continued involvement though investment and construction details remain undecided. Separately, Microsoft and NVIDIA launched the RTX Spark–equipped Surface Laptop Ultra and detailed Windows hybrid intelligence, which runs some large models and agent tasks locally: its local MAI Code 1.1 Flash has 137B total parameters with roughly 6.8B activated per token in an MoE design, quantized weights of about 53GB, a vendor-reported 75.5GB peak memory at 256K context, and 70.8% on SWE-bench Verified (quantized) versus 72.6% (high precision). The piece argues compute competition is shifting from sheer scale to redrawing the cost boundary for a unit of useful AI capability.

虎嗅 AI·
3330
Industry

Ecosia drops Mistral for Chinese open-source AI models

Berlin-based search engine Ecosia, which is used by government services, is ending its work with France's Mistral and moving mainly to open-source models, including Chinese ones. Founder and CEO Christian Kroll told Politico: "We are disappointed with the quality of Mistral," saying the model now trails competitors by a full year. Reports note Ecosia had only switched from OpenAI to Mistral in May, while a Chinese-language headline claims the move to Chinese open-source models halves costs.

Techmeme·
3415
Industry

Ex-OpenAI and Cognition staffers launch Hone with $60M seed for business-running AI agents

A group of former employees from fast-growing AI firms including OpenAI and Cognition AI have founded Hone, which builds AI agents meant to act as professional staffers handling long-running tasks that span weeks or months to help run a business. The company announced a $60 million seed round at a $285 million valuation.

Bloomberg Technology·
358
Industry

OpenAI told investors revenue neared $50B annualized in September, below reported $70B

OpenAI told investors its annualized revenue was approaching $50bn at the end of September, according to the Financial Times. That is well below the $70bn that was widely reported in the media based on investor documents. The report gives no further detail on how the two figures were calculated or what period they cover.

Techmeme·
3650
Industry

Anthropic updates usage policy to ban abuse of Claude and election interference

Anthropic published a new Usage Policy, its first refresh in over a year, taking effect on November 12. It adds a ban on "sustained and needless abusive or cruel behavior" toward Claude, consolidates scattered rules into a new section on deceptive campaigns and fake accounts, renames the elections section "Do Not Undermine Democratic Processes," and clarifies restrictions on weapons software and components, surveillance, law enforcement and high-risk use cases. Anthropic says Claude ending abusive conversations remains the primary enforcement mechanism, while The Decoder reports the policy lets the company warn, throttle, restrict, suspend or terminate access.

Anthropic · News·
3732
Research

OpenAI withdraws three math papers over a sign error

In its math repository history, OpenAI said a sign error in “Algebraicity of Weil classes on split abelian eightfolds” invalidates a stabilization-trace cancellation argument and the construction used by two dependent papers, leading it to withdraw three manuscripts. Fourteen other manuscripts were revised with proof repairs, corrected statements and clearer hypotheses, and 13 more were updated to cite the revised editions. Six additional formalizations and five other additions bring the total share of top-line results formalized to 300/719, about 42%.

Hacker News · AI（100+ 分）·
389
Models

Tavus says Griffin-Lite is first video chat model to pass a 'video Turing test'

San Francisco AI video company Tavus released a new model, Griffin-Lite, which it calls the first real-time video conversation model to pass a "video Turing test" and describes as a "human interaction model." Tavus says that after one-minute video calls, 26 of 54 participants (48%) mistook it for a real person, versus 2.4% for the previous-generation system under the same test. Unlike voice assistants, Griffin-Lite can react in real time while both sides are speaking, whether agreeing or interrupting.

36氪快讯·
398
Models

Odyssey launches Odyssey-3 world model that drives cars and controls humanoids

Palo Alto lab Odyssey released Odyssey-3, a world model it says can control robot arms and humanoids, drive a car, fly drones indoors, generate environments for AI agent training and play Grand Theft Auto V; a research preview is available now. Odyssey says Odyssey-3 Pro scored 66.1 on the video-to-video test of Physics-IQ Verified — the highest reported on that leaderboard, using the best of eight attempts per task — and ranked first in three of four WorldMark categories, a result from its own evaluation. Humanoid company Flexion has built control policies on Odyssey-3, which Odyssey says handled lighting changes that caused tested baselines to fail; the car was driven on real roads in India with a policy trained on about 20 hours of data.

The Next Web·
409
Products

Anthropic adds build-eval and hillclimb tuning commands to Claude Code

On September 28 Anthropic published a tuning guide introducing two Claude Code entry points, /claude-api build-eval and /claude-api hillclimb (Claude Code v2.1.259 or later). The first turns "what counts as a good answer" into repeatable test cases and scoring rules; the second lets Claude propose one change per round — to prompts, skill files, tool descriptions or model config — keeping it only if the evaluation improves and reverting otherwise, with a hidden held-out set used to check for overfitting. Anthropic says an internal customer-support evaluation improved decision accuracy from 78.6% to 90.5% on 14 held-out tickets while cutting model call cost to roughly one fifth.

36氪 人工智能·
Next →← Previous