AI Systems Tested on Research-Level Mathematics Problems

This document presents the results of a rigorous evaluation of current AI systems' capabilities in solving complex, research-level mathematics problems. The study involved a set of ten challenging problems, which were contributed by a diverse group of mathematicians and arose naturally from their ongoing research. These problems spanned a broad spectrum of mathematical fields. The paper details the specific problems used, the methodology employed for testing the AI systems, and the comprehensive results obtained. Supplementary materials, including human-generated solutions, the AI-generated solutions, and detailed referee reports and logs for the AI outputs, are also provided for transparency and further analysis. This initiative aims to assess the frontier of AI in mathematical reasoning and problem-solving, offering valuable insights into where current AI systems excel and where they still face significant limitations when confronted with novel, research-grade mathematical challenges.