Fields Medalists warn AI labs are turning famous proofs into a scoreboard
Twenty-five Fields Medalists, including Terence Tao, signed a declaration posted Friday that AI companies' push to solve mathematical problems as a benchmark is "detrimental to the science of mathematics," and that lab goals and the math community's goals are "severely misaligned," Tao writes on What's new.
The Fields Medal is math's top prize, roughly Nobel-class. A benchmark here means treating famous unsolved problems as a leaderboard for models.
Tim Fernholz at TechCrunch frames the letter against this week's feud and notes that on Thursday OpenAI withdrew sponsorship of a Caltech AI math event after Caltech researchers criticized it.
Truman Dickerson at Business Insider quotes OpenAI research lead Dan Roberts saying the company is no longer sponsoring Caltech's "math hackathon" and wants to engage with the math community more on how to integrate the technology.
An open letter from current and former Caltech mathematicians called the event "likely to have destructive impacts for the mathematical community." They labeled rushed AI claim-then-verify labor "slop mathematics": mathematicians verify, disseminate, and sometimes discredit the claimed results unpaid.
Anthropic and OpenAI had agreed to support the event with $2 million in AI credits. It was unclear how many were OpenAI's or what happens to them now. First round: 40 hours plus $20,000 in tokens for teams solving an open problem.
The declaration's core claim is older than any one sponsorship. Solving is only a proxy for conceptual understanding.
Rushed announcements leave no time for proper writeups, isolating methods, or citing prior work. Without willing mathematicians to integrate ideas into the canon, the human transmission chain breaks.
Labs want medals on a scoreboard. The medalists want the craft that turns a true statement into something a student can still learn.
TechCrunch notes OpenAI's recent high-profile proof claim remains unverified, and that NYU's Tristan Buckmaster accused OpenAI of pressuring him not to credit an Anthropic-employed collaborator. That is Buckmaster's accusation via TechCrunch, not an adjudicated theft finding.
The Fields letter is not a ban on AI tools. Tao's post says AI can enhance genuine mathematical study. Whether the change helps or destroys the field, the signatories write, will be decided by the humans who control the technology.
The scoreboard is getting louder. The writeups are not.