Four stories lit up X this week. I opened each article to check what it says.
Meta ships its first frontier model. On Oct 12, Meta’s Superintelligence Lab released Muse Wonder and Muse Wonder Full. Through the API, Wonder costs $4 per million input tokens and $20 per million output tokens. Wonder Full is only for Muse Code subscribers. Meta says the scores match Fable 5.1 and GPT-6 Astra, but those numbers come from Meta’s own testing. Source: TechCrunch, shared by @TechCrunch.
Scientists want risky AI models cut off. More than 400 researchers, including Geoffrey Hinton and Yoshua Bengio, signed a letter asking labs to airgap cyber-capable models at first. They also want an international body to record containment failures and make them public. Source: the letter, shared by @Yoshua_Bengio.
Outside eyes inside AI labs. METR proposes that independent assessors review a lab’s internal data, including whether its AI agents could slip out of control. The labs stay in charge of access and redactions, and METR says this is not an audit. Source: METR, shared by @METR_Evals.
Washington wants “SI” to catch on. A US Treasury official told Reuters the US is pushing IMF members to say “super intelligence” instead of “artificial intelligence.” The IMF statement is due Saturday, and “new technology” was floated as a fallback. Source: Arise News, citing Reuters, shared by @ARISEtv.
Why it matters: the tools are getting stronger, and the fight over who checks them is getting louder.
Written by Max J. Meindl III. Questions or corrections: [email protected].