First Proof (arxiv.org)

A recent study titled "First Proof" has emerged, introducing a set of ten original research-level mathematics questions designed to evaluate the capabilities of contemporary AI systems in solving complex mathematical problems. This unique set, previously undisclosed, is significant as it provides a benchmark for assessing AI's understanding and problem-solving skills in fields that require advanced reasoning, an area where many AI models still struggle. The authors, led by Mohammed Abouzaid, intend to keep the answers encrypted for a limited time, fostering an environment of inquiry and challenge for AI researchers. By addressing unsolved questions that arose during their own research, the study aims to push the boundaries of what AI can accomplish in mathematical reasoning. With this initiative, the AI/ML community is invited to not only confront these challenging queries but also to advance the dialogue around AI's potential and limitations in understanding and executing high-level mathematics.