25 Fields Medal Winners Warn AI Benchmark Race Is Harming Mathematics

My take: On September 11, 25 Fields Medal winners, mathematics' highest recognition, published a joint declaration at mathandai.org arguing that AI labs' race to solve mathematical problems as benchmarks is harming the discipline. Among the signatories: Terence Tao, Peter Scholze, Maryna Viazovska, Cédric Villani, and Pierre Deligne.

The problem they are raising is fundamental: AI labs optimize for speed, publishing "breakthroughs" on mathematical problems at a pace that outstrips the community's ability to verify them. That is not mathematical progress: it is a marketing race that uses mathematics as a benchmark surface. The distinction matters, because attributing the solution of a classical mathematical problem to an AI model without independent verification distorts both the AI field and mathematics itself.

For creators, researchers, and technical teams using AI for analytical work, this declaration is a reminder that benchmarks alone do not guarantee a system is capable in practice. Does your team evaluate AI tools with real tasks from your own work, or do you rely primarily on the benchmarks the vendors themselves publish?