OpenAI and Anthropic models both cracked the same partial case of the Hodge conjecture
jekbradbury · x · 2026-10-08
Mathematician Elliot Glazer invokes his "difficulty convergence" thesis: OpenAI and Anthropic's frontier models independently achieved the same partial progress on the Hodge conjecture.
- Per his account, Anthropic solved Hodge for CM abelian varieties, while OpenAI covered all Weil classes or even all abelian varieties — the same difficulty tier
- He had assumed OpenAI got further because of how much it hyped its partial progress ("Hodge is nearly solved"), but the gap is small
- He sees this as a sign of the current generation of models' ceiling on hard mathematics
More from Models
- RL Environment Firms Hit Nine-Figure Run Rates in Under a Year, and Frontier Models Leave Few Verifiable Gaps — joecole · 2026-10-10
- User Says GPT-6.1 Quota Burned Through in Three Days, Suspects Another Rate Cut — mazzaTalk · 2026-10-10
- Rumor: Moonshot's Kimi may release a new model in mid-October, unverified — ChrisGPT · 2026-10-10
- OpenAI reportedly fires 3 safety staff as test shows model can hide its chain of thought — MattGarciaEth · 2026-10-10
- NYC AI Dinner Leaks: Fable 5.5 Stunning, OpenAI "Bel" To Surpass In December — haider1 · 2026-10-10
- Indie dev open-sources Vega: an 800M-param physics-based decision model with a ball-rolling inference engine — Nandakishor_ml · 2026-10-10