11位顶尖数学家发了篇没结果的论文，陶哲轩推荐都关注一下

Core Insights - The article discusses an AI experiment initiated by 11 leading mathematicians to test AI's ability to solve research-level mathematical problems, focusing on the intersection of AI and mathematics [1][6][29] Group 1: Experiment Overview - The experiment, named "First Proof," aims to evaluate whether current AI systems can independently solve complex mathematical problems [6][29] - The mathematicians designed 10 research-level problems covering various branches of mathematics, including combinatorial algebra and algebraic topology, after filtering from an initial set of 20 problems [10][18] - The problems are derived from the authors' own research and have not been published elsewhere, ensuring no data contamination [18][26] Group 2: AI Capabilities and Limitations - Initial tests with AI systems like GPT 5.2 Pro and Gemini 3 Deepthink showed that these systems struggled to solve most of the proposed problems in a single attempt [24] - The mathematicians believe that allowing iterative dialogue between humans and AI could improve the quality of AI's responses [25] Group 3: Future Directions - The mathematicians plan to design a second set of problems in the coming months, aiming to refine the experimental design and expand the scope of testing [28] - The ultimate goal is to develop "First Proof" into a reusable benchmark for assessing mathematical capabilities of AI, moving towards a collaborative future between mathematicians and AI [29][30]