Two Millennium Prize problems solved in a year, but AI still can't build theories
patience_cave · x · 2026-11-22
The author reflects: he used to think an AI solving a Millennium Prize math problem would mean AGI. Two have been solved within a year — and life goes on, since agents can grind out solutions. His revised bar: theory-building and defining new problems, which even the strongest upcoming models can't do.
More from AGI Musings
- Math group slams Anthropic's AI-generated Riemann proof: no human can vouch for it — appenz · 2026-11-22
- Databricks retrieval lead on leaving SWE for a PhD — and when research careers make sense — yacinelearning · 2026-11-22
- Testing grand AI narratives with country data: Norway vs US comparison doesn't support the claim — edelwax · 2026-11-22
- Lean community certifies the Riemann Hypothesis proof: 17M lines fully verified — rbhar90 · 2026-11-22
- Studies suggest smarter AI systems become less predictable — skdh · 2026-11-22
- C players in tech have 3-5 years: treat your skills like AI models that ship upgrades — annetgriffin · 2026-11-22