Measuring AI's ability to autonomously generate mathematical proofs

Mission

AI has the potential to accelerate scientific discovery, but we lack reliable ways to measure research-level reasoning because it is difficult to separate true reasoning from memorization. Mathematics is the ideal testbed: correctness is unambiguous, verification standards are exceptionally high, and there is no reproducibility crisis. The First Proof Project will provide independent assessments of the reasoning abilities of AI systems in the context of research mathematics. Progress in rigorous evaluation here can inform how AI should be developed and deployed across science.

Team and Editors

Editorial Board

Mohammed Abouzaid Stanford University

Nikhil Srivastava University of California, Berkeley

Rachel Ward University of Texas at Austin

Lauren Williams Harvard University

Board of Directors

Andrew Blumberg Columbia University

Martin Hairer EPFL and Imperial College

Tamara Kolda MathSci.ai

Daniel Spielman Yale University

Shmuel Weinberger University of Chicago

Executive Director

Mohammed Abouzaid

executivedirector@1stproof.org

Get Involved

If you are a mathematician interested in contributing to First Proof, please reach out to us at contact@1stproof.org.