OpenAI released proofs to hundreds of unsolved math problems — an AI agent made them explainable
DeryaTR_ · x · 2026-10-10
OpenAI has released solutions to hundreds of previously unsolved math problems, drawing objections from mathematicians who argue nobody can understand or verify the proofs. DeryaTR pushed back by having a ChatGPT dot agent pick Seymour's second-neighborhood theorem and explain the problem, solution, and verification to a non-math audience, producing four explainer visuals.
The agent consulted GPT-6 Pro for critical review; all 43 original proof files passed Lean formal verification, and an audit found no gaps or assumptions beyond standard foundations. The same process could cover all remaining solutions in about a day — the AI that raised the verification worry may also be the tool that resolves it.
More from Models
- PostTrain Agent tops PostTrainBench at 46.6, near human experts' 51.1 with fully autonomous post-training — jiqizhixin · 2026-10-10
- Human+agent teams push math PRs to OpenAI papers as a new frontier model looms — RexDouglass · 2026-10-10
- Looking back at early GPT reviews: GPT-2 accidentally picked up JavaScript — moyix · 2026-10-10
- Claude insists defection is rational by backward induction, then concedes the decision theory is contested — xuanalogue · 2026-10-10
- Rumor: OpenAI quietly preparing an adult mode for ChatGPT — koltregaskes · 2026-10-10
- Debunking the claim that AI systems persist nothing across interactions — TinfoilTricorn · 2026-10-10