Mathematicians Dispute AI Labs' Claimed Proof Results
Among the results attributed to OpenAI's Astra and a comparable Anthropic model, both labs claimed disproofs of the 1982 Connes' Rigidity Conjecture. A circulating argument by researcher Jenny Lorraine Nielsen contends the two claims are self-contradictory: applying a published 2011 classification theorem to the papers' own constructions forces a contradiction the authors' theorems can't survive. AI critic Gary Marcus is soliciting mathematical rebuttals rather than asserting the critique himself.
Why It Matters
The dispute is playing out on the mathematics, not the marketing — a real-time stress test of whether AI-generated proofs can survive expert scrutiny as fast as labs can publish them.