OpenAI’s 722 Math Papers: Three Retracted in 24 Hours

OpenAI published 722 AI-generated math manuscripts on October 6, crediting an internal model unavailable outside the company, and within 24 hours three papers were retracted and 14 more were revised. A sign error in the algebraic geometry proof "Algebraicity of Weil classes on split abelian eightfolds" invalidated its stabilization-trace cancellation argument and cascaded into two dependent manuscripts, according to the report. The retractions mark the first documented cracks in the release before the math community had spent a full day with it.

OpenAI published 722 AI-generated math manuscripts on October 6, crediting an internal model nobody outside the company can access. By the next morning, three papers were pulled. A sign error in one algebraic geometry proof had quietly cascaded into two dependent manuscripts, taking them down with it. Fourteen more papers were revised the same day. The math community had spent less than 24 hours with the release before cracks appeared. What Broke and Why The error originated in a paper titled “Algebraicity of Weil classes on split abelian eightfolds.” A flaw in the stabilization-trace cancellation argument invalidated the proof — … The post