ThursdAI: Cognition SWE-2 nears frontier, safety week drama and OpenAI's Navier-Stokes claim
thursdai_pod · x · 2026-09-11
ThursdAI's Sep 10 episode roundup:
- Coding models: Cognition's SWE-2 / Devin team model is catching up to frontier level
- Math claim: OpenAI says a 10k-agent swarm produced a forced Navier-Stokes blowup proof with Lean formalization; the show stresses the forced vs unforced distinction for the Clay statement
- Meta Muse: more than a chat wrapper — a personal AI agent with its own Linux VM and native Stripe Link, demoed live producing show chyrons
- Safety week: seven AI insiders ran a multi-day public extinction-risk cascade, plus the "I quit after six weeks" Anthropic thread
Also covered: Desert Ant Labs on-device stack, CoreWeave Agent Loops hackathon (Sep 12-13, SF), Google Lyria 2.5, and ChatGPT Voice adding Sol and Astra voices.
More from AGI Musings
- A 1980s Welsh poem about a money-singing machine resonates in the AI era — HanesPrydain · 2026-09-11
- Gary Marcus: AI is strong in constrained domains, weak in the open physical world — GaryMarcus · 2026-09-11
- Satya Nadella's "Reverse Information Paradox" resurfaces amid AI boom — otarU · 2026-09-11
- AI filmmaker argues hybrid production keeps gatekeepers in charge, backs 100% AI filmmaking — taherdhanera · 2026-09-11
- Drexler on the HuggingFace incident: system structure, not model alignment, drove the change — mattbeane · 2026-09-11
- CNBC examines risks of recursive self-improvement, quoting Conitzer — conitzer · 2026-09-11