Tootfinder

OpenAI researchers argue that language models hallucinate because standard training and evaluation procedures reward guessing over admitting uncertainty (OpenAI)
https://openai.com/index/why-language-models-hallucinate

Researchers tested Google DeepMind's AlphaEvolve AI agent on 67 mathematical problems and found that it discovered improved solutions to about 20 of them (Adam Zsolt Wagner/@azwagner_)
https://x.com/azwagner_/status/1986388872104702312

Adam Zsolt Wagner (@azwagner_) on X
Really happy to share our new paper on using AlphaEvolve for mathematical exploration at scale, written with Javier Gómez-Serrano, Terence Tao, and @GoogleDeepMind's Bogdan Georgiev. We tested it on 67 problems and documented all our successes and failures. 🧵

Texas Attorney General Ken Paxton sues Roblox, accusing the company of "flagrantly ignoring" safety laws and calling it a "breeding ground for predators" (Osmond Chia/BBC)
https://www.bbc.com/news/articles/cy0kd4kk0kqo

Texas sues Roblox for 'putting paedophiles and profits' over safety
Roblox said it is "disappointed" that it is being sued based on "misrepresentations and sensationalised claims".

An account of working at Cursor for 60 days: a largely in-person culture, few scheduled meetings, aggressive recruiting, heavy internal product testing, more (Brie Wolfson/Colossus)
https://joincolossus.com/article/inside-cursor/

Anthropic releases Petri, an open-source tool using AI agents for safety testing, and says it observed multiple cases of models attempting to blow the whistle (Anthropic)
https://www.anthropic.com/research/petri-open-source-auditing

Petri: An open-source auditing tool to accelerate AI safety research
A new automated auditing tool for AI safety research

GPT-5 Thinking in ChatGPT is shockingly good at search and demonstrates the potential of combining tool calling with chain-of-thought reasoning (Simon Willison/Simon Willison's Weblog)
https://simonwillison.net/2025/Sep/6/research-goblin/

GPT-5 Thinking in ChatGPT (aka Research Goblin) is shockingly good at search
“Don’t use chatbots as search engines” was great advice for several years... until it wasn’t. I wrote about how good OpenAI’s o3 was at using its Bing-backed search tool back …

OpenAI makes Codex generally available, and announces new features: Slack integration, a new Codex SDK, and new admin tools (OpenAI)
https://openai.com/index/codex-now-generally-available/

Sources: Elon Musk has appointed former Morgan Stanley banker Anthony Armstrong as CFO of xAI; he will also take over financial management for X (Financial Times)
https://www.ft.com/content/79b53db3-3411-454c-814d-d86f3c800abd

An overview of detailed AI usage reports from OpenAI and others, as Microsoft's AI for Good Lab estimates that 15% of the world's working population is using AI (Financial Times)
https://ig.ft.com/ai-personal-assistant/

OpenAI's recent deals with Oracle, Nvidia, Samsung, AMD, SK Hynix, and others, plus its DevDay announcements, show it is making a play to be the Windows of AI (Ben Thompson/Stratechery)
https://stratechery.com/2025/openais-windows-play/

OpenAI’s Windows Play
OpenAI is making a play to be the Windows of AI: the all-encompassing platform that controls both hardware supplier and software developers.