As AI companies race to show their models can do research-level mathematics, a Stanford mathematician is pushing them to show their work.
Who he is
Ravi Vakil is the Robert Grimmett Professor of Mathematics at Stanford and president of the American Mathematical Society, according to his faculty page. He cofounded the Q&A site MathOverflow and San Francisco's Proof School, and wrote the algebraic geometry text “The Rising Sea.” He is one of a group of scientists in conversation with OpenAI, Anthropic and Google DeepMind, the SF Standard reports.
On what AI can do
“AI is incredibly intelligent in some ways, but it's incredibly stupid in other ways,” Vakil told the Standard. “Most of what it produces is crap, and some of it can be really useful.” He said it's useful for “sanding off the rough edges” but dangerous if people offload their creativity, learning or judgment.
On transparency
Vakil's group recently asked OpenAI to release the materials behind its new math results. “We don't want just the output. That's not science,” he said. He wants labs to share what resources they used, how they found results and what it cost. “Scientific advances are the common heritage of us all.”
He said the labs “do listen to us,” even if they don't always agree.
His worries
Vakil said he doesn't see AI replacing mathematicians, because “we're the ones doing the wondering.” His bigger fears: that young people stop learning to think mathematically, and that AI becomes “a Trojan horse for anti-intellectualism,” used to justify cutting basic math, writing and reading instruction. Ideally, he said, any student should have access to state-of-the-art tools regardless of school or family income.