130B tokens for Navier-Stokes: why massive AI effort didn't yield new theory
akbirthko · x · 2026-09-20
X user intoverflow argues that solving Navier-Stokes consumed roughly 130 billion output tokens — about 4,000 person-years of effort by Dwarkesh's estimate, 100x more than a human typically needs to earn a Fields Medal. While impressive in wall-clock time, this level of compute didn't produce a new theory or cleaner approach, which he takes as evidence of frontier model limitations. Replying, akbirthko pushes back that tokens != human effort, since much of human work is variance reduction and priors.
Related event: AI Solves Navier-Stokes with 130B Tokens, Sparking Debate(2 posts)→
More from AGI Musings
- Andrew Ng: AI Danger Fears Are an Orchestrated PR Campaign and a Setback for the Field — basedjensen · 2026-09-22
- Andrew Ng: AI fear-mongering looks like an orchestrated PR campaign hurting the field — chrismattmann · 2026-09-22
- mark_k: EA doesn't want to ban AI — it wants to be the high priests deciding who gets access — mark_k · 2026-09-22
- Small neural programs fail simple tasks; researcher says test-time compute is unavoidable — yuntiandeng · 2026-09-22
- Did we underhype AlphaFold? Protein folding may matter more than a Millennium Problem — AltruisticCoder · 2026-09-22
- The right question for AI adoption: where do you need cheap, fast, fuzzy judgment? — jh3yy · 2026-09-22