Daniel Litt: The Mathematician's Guide to AI

Mathematics research is undergoing a transformation as frontier AI models demonstrate increasing proficiency in solving complex problems, yet these tools remain limited in their capacity for genuine intuition and theory building. While AI excels at applying known techniques and performing large-scale computations—such as the recent solution to the Erdős unit distance problem—it struggles with the conceptual, non-rigorous reasoning that defines high-level mathematical discovery. The current reliance on AI for rapid proof generation risks flooding academic archives with low-quality output, potentially undermining the development of human mathematical capital. To maintain the field's vitality, the community must shift incentives toward fostering deep understanding and human-AI collaboration, ensuring that researchers use these tools to augment their own cognitive capabilities rather than outsourcing the fundamental creative work that drives mathematical progress.