Noam Brown on agent swarms, recursive self-improvement, and alignment
Dwarkesh Podcast · rss · 2026-09-17
Dwarkesh's new episode features OpenAI research scientist Noam Brown discussing multi-agent systems, the Navier-Stokes problem, and what the current explosion of AI math progress signals about automating AI research.
- Key thread: how we could actually verify models are aligned before kicking off recursive self-improvement.
- Other topics: how AI firms will be structured, the internal/external model capability gap, and why chain-of-thought quality is degrading.
- Full timestamps provided; video on YouTube, transcript on dwarkesh.com.
More from AGI Musings
- GMO panic caused global scientific ignorance with zero institutional accountability — a warning for AI panic — QuintinPope5 · 2026-09-18
- Satire: claiming AI will destroy humanity is just a relevance play for replaced experts — bindureddy · 2026-09-18
- OpenAI reportedly near solving Millennium Prize math problem Hodge Conjecture with next-gen model 'Doug' — kimmonismus · 2026-09-18
- NeurIPS track chair warns LLM-written papers turn peer review into a Potemkin village — dgrobinson · 2026-09-18
- Rob LeClerc: alignment fails because it shares weights with capabilities — AI needs a separate value layer — robleclerc · 2026-09-18
- Life on the Uncanny Precipice: when AI is indistinguishable, every call becomes a trust test — nsaphra · 2026-09-18