Skip to content

openai/math

  • URL: https://github.com/openai/math
  • Stars: 14295
  • Language: Lean
  • Category: AI & Agents
  • Topics: None

What it does

This repository contains mathematical manuscripts and proof artifacts produced by an internal OpenAI model, focusing on various mathematical problems. It includes results at different stages of verification, with some outputs building upon earlier model results.

Why it's interesting

Unlike typical mathematical repositories, this one leverages an internal AI model to generate and evaluate mathematical proofs, providing a unique intersection of AI and mathematics. The collection is organized into families of related papers, which is not a common structure in traditional mathematical literature.

How it works

The results were generated using an unreleased internal OpenAI model, with each result requiring an average of three hours of compute time. The repository includes a Lean library for formal proofs, with approximately 42% of the results formalized. The manuscripts are organized into families based on related topics, and the repository is continuously updated with new formalizations as they become available. However, not all results are formally verified, and some may contain issues.

Get started in 5 minutes

To explore the repository, start by reviewing the overview for descriptions of the families and use the manuscript map to locate specific papers. You can also access the preprints/ directory for PDFs and source files.

Watch out for

The repository contains results at varying stages of verification, and some unformalized results may have issues. The README does not specify a license, which could affect usage rights. Additionally, the internal model used for generating results is not publicly available, limiting reproducibility.