25 Fields Medalists collectively slam AI companies: Don't use math problems as benchmarks.

Beating AI News Flash: Terence Tao, Deng Yu, Peter Scholze, and 25 other Fields Medalists jointly issued a statement, saying that AI companies are treating the solving of major mathematical problems as a benchmark and have already begun to harm mathematical research. The statement acknowledges that large models have made rapid progress in recent months and can already solve important unsolved problems in multiple fields, but mathematics places greater value on the new concepts, new methods, and understanding that arise from this.

What they are most dissatisfied with is that AI companies are rushing to announce results. The statement says that some solutions have been publicly announced before there has been time to organize complete proofs, distill new methods, and fill in prior citations, resulting in serious authorship and plagiarism problems. Without mathematicians continuing to digest and organize these results, the ideas found by AI will also find it difficult to truly enter the mathematical system.

The statement did not name OpenAI, but The Economist's accompanying report directly ran the headline "Top Mathematicians Are Angry at OpenAI's Practices." A few days earlier, OpenAI had just become embroiled in controversy over authorship and undisclosed research related to the Navier-Stokes proof. Terence Tao said that this statement was rapidly formed after discussions over the past week among 25 Fields Medalists, because they believed the situation was already urgent enough.