UCLA Team Wins $250,000 Grant to Build AI That Reasons Like a Mathematician
Six UCLA faculty members received a seed grant from the Laude Institute to develop an AI system capable of mathematical discovery and rigorous proof. The project, "Accelerating the Queen of Sciences," was one of eight selected from 125 proposals across 47 institutions in the U.S. and Canada.
The team is led by Amit Sahai, a computer science professor at UCLA Samueli School of Engineering. Other members include computer scientists Raghu Meka, Wei Wang, Kai-Wei Chang and Nanyun Peng, along with mathematician Terence Tao, a Fields Medal winner who directs special projects at the Institute for Pure and Applied Mathematics.
The Problem Current AI Cannot Solve
Today's leading AI models handle competition-level math problems and explain reasoning fluently. But they falter at sustained, multi-step reasoning and rarely produce reliable proofs. They generate answers that sound correct but aren't.
The UCLA team is taking a different approach. Rather than training AI to solve known problems faster, they're teaching it to think like a mathematician - working through textbooks, generating new problems, iterating on failed attempts and building conceptual understanding.
How the System Works
The AI will operate in two modes. An exploration mode lets the model speculate about hidden patterns and open questions. A verification layer then evaluates those ideas using formal proof systems, ensuring outputs meet strict standards of mathematical accuracy.
"What excites me about this team is that they're not asking AI to solve known problems faster; they're asking it to wonder, conjecture and discover the way a mathematician does," said David Patterson, a Turing Award winner who chairs the Laude Institute's selection committee.
Sahai emphasized the goal differently: "I'm most excited by our goal of creating AI capable of wonder-based discovery in mathematics - an AI system that not only solves problems rigorously, but also asks what might be true and explores the boundary between possible and impossible."
What Comes Next
Over six months, the team will produce an open-source proof of concept, a benchmark suite for evaluating research-level mathematical reasoning, a lab proposal and a vision paper developed through a UCLA-hosted workshop with domain experts.
The eight winning teams will present their work in October. The top team receives $10 million to establish a multi-year Moonshot lab.
This project builds on earlier work. Wang's team recently secured a three-year, $5 million grant from the Defense Advanced Research Projects Agency to develop AI tools for making mathematical discoveries, formalizing them and verifying them.
Why This Matters for Research
If successful, the system could accelerate progress across physics, cryptography and other quantitative fields where rigorous reasoning is essential. The work addresses a fundamental bottleneck: teaching machines to reason through complex, multi-step problems without producing plausible-sounding errors.
Terence Tao noted the timing: "Previously impossible 'moonshots,' such as creating a rigorously verified, effective and genuinely creative explorer of mathematical results, now seem within reach."
The Laude Institute, a nonprofit founded in 2025, focuses on bridging academic AI research and real-world impact. Its Moonshots initiative targets scientific discovery, health care delivery, civic discourse and workforce adaptation for the AI era.
Your membership also unlocks:
