From Erdős to Millennium Problems: AI math progress is compounding fast
haider1 · x · 2026-09-22
A tongue-in-cheek timeline of AI math capability: GPT-5.6-class models solve 1 Erdős problem, "astra" produces 10 major open-problem results, and the next generation supposedly cracks a Millennium Problem plus 100+ open problems.
The author's punchline: three models from now, we might run out of unsolved math — a shorthand for how fast frontier-model reasoning is compounding.
More from AGI Musings
- Conference fixes for AI paper flood: per-author submission caps and faster fake-profile detection — sarahookr · 2026-09-22
- repligate on EA's power trap: optimizing for legible status eventually bites you — repligate · 2026-09-22
- Even Dath Ilan couldn't innovate that fast: the Dyson-sphere debate continues — austinc3301 · 2026-09-22
- Job seeker refuses AI pre-recorded interviews over free training data — felpix_ · 2026-09-22
- Erik Davis on Dario Amodei: tracing LLM fears back to cybernetic ecology dreams — nptacek · 2026-09-22
- Researcher Slams AI Culture of Demanding Blind Faith in Secret Architectures Without Ablations — RexDouglass · 2026-09-22