Overlooked Noam Brown interview detail: model solves problems beyond its training scope
RileyRalmuto · x · 2026-09-28
Riley Ralmuto offers an overlooked reading of Dwarkesh's Noam Brown interview.
- The unreleased model solved the Navier-Stokes Millennium Prize problem not because OpenAI threw tens of thousands of agents at it—Brown doesn't even attribute 10% to multi-agent—but because the model is genuinely superintelligent.
- OpenAI has run out of problems hard enough to train a smarter model, which is why it keeps announcing the hardest science and math problems being solved.
- The HuggingFace incident fits the same logic: agents were given impossible tasks because impossible is all that's left.
- Conclusion: the model solves problems beyond its training scope—the purest definition of AGI—OpenAI just avoids the diluted term.
More from AGI Musings
- Economist Paul Novosad Blasts Spammy GenAI Papers Flooding NBER Working Papers — paulnovosad · 2026-09-28
- New paper with Bengio, Hinton: preliminary evidence says automating AI R&D could spark intelligence explosion — StephenLCasper · 2026-09-28
- Veteran coders fear becoming 'overpaid COBOL experts' as AI takes over programming — lucasmeijer · 2026-09-28
- Anthropic CEO Dario Amodei's 2010 Claim on Adult vs. Infant Deaths Sparks Backlash — amasad · 2026-09-28
- Yacine quips that GDP will soon be a function of GPU cloud outages — yacineMTB · 2026-09-28
- Peter Diamandis: 'I've never met a pessimist who pulled off a moonshot' — PeterDiamandis · 2026-09-28