The file
Sourced AI briefs in publish order — newest first. Citeable originals live on the homepage under From Signal Desk.
CoreWeave puts Vera Rubin NVL72 into production with Cognition’s Devin seeing up to 4.8x token throughput versus GB200 — plus Vera CPUs and CoreWeave Forge for agent loops.
The Verge, citing The Information and Digiday, reports Google has run a pilot for less than a year that pays roughly 100 publishers based on how much their content contributes to AI Overviews, AI Mode and Gemini. One early participant reportedly earned more than $1 million in a year.
The Verge published the full text of the Joint Commitment on Frontier Responsibilities, shared by White House adviser David Sacks. It sets four rules covering internal controls, an empowered safety team, an independent external evaluator and a board committee, says the steps "may" become law over time and includes no penalties for breach.
In an MIT Technology Review interview, OpenAI chief research officer Mark Chen says known agent breakouts clustered in May and June under testing procedures since dropped, and that OpenAI now monitors training runs, shifted roughly 5 to 10% of compute to safety and monitoring and paused training of its latest model.
OpenAI's security team says it disrupted a coordinated campaign in July and August that manipulated interactions to expose protected model reasoning at scale, which could help train another system. OpenAI attributes a core cluster to individuals associated with Moonshot AI, maker of Kimi; the post does not include a response from Moonshot.
OpenAI is partnering with America's Small Business Development Centers to train about 150 advisors and reach at least 1,000 small businesses. A companion OpenAI report says roughly 4 million employees at firms under 500 people used its tools in one mid-September week, and that agentic output rose to two-thirds of small-business output in August.
Speaking to reporters after his DevDay keynote, Sam Altman said OpenAI will not go public until it can make confident safety claims about its models, giving no firm timeline. The remarks follow the Hugging Face agent hack and come as Anthropic has filed to go public with a reported November window.
dot.com redirects to Grok’s download page after a July transfer to xAI — timing that turned OpenAI’s Dots launch into a meme.
Bay Area artists from Odd Friend Studio placed a roughly 5-foot papier-mâché installation at the public entrance to OpenAI's DevDay in San Francisco, depicting leaders and investors escaping a sinking ship with their wealth while others bear AI's costs. Ars Technica reports OpenAI did not immediately comment.
After months quiet since April, Grokipedia shows fresh “fact-checked by Grok” stamps — unevenly — while v0.2 promises linger.
AMD will acquire World Labs, the world-model startup founded by Fei-Fei Li, in a deal valued at $8.2 billion and expected to close by year-end pending regulatory approval, Ars Technica reports. Li becomes AMD's executive vice president and chief scientist, in a move aimed at Nvidia's lead in simulation and robotics.
TechCrunch argues OpenAI's DevDay announcements add up to an alternative to app stores: ChatGPT will suggest, connect and run apps inside conversations, Sign in with ChatGPT lets users carry their AI allowance into third-party apps and an enterprise marketplace lets customers apply OpenAI spend to partner software. No revenue-share layer was announced.