Terence Tao, a Fields Medal recipient and UCLA professor, warned that AI is exhausting the supply of significant open mathematical problems. He argues that AI models solve these challenges faster than researchers can identify new ones, potentially hindering long-term scientific progress.
The issue stems from AI labs, including OpenAI and Anthropic, applying massive computational power to long-standing mathematical puzzles. In May, an OpenAI model disproved the 80-year-old Erdős unit-distance conjecture, a result later verified by independent mathematicians.
Anthropic’s unreleased model, Claude Mythos, also solved the unit-distance problem independently. Furthermore, Anthropic recently formalized a centuries-old proof of Fermat's last theorem, while OpenAI solved a 90-year-old problem shortly after a researcher published his own work.
Tao suggests labeling specific tasks as analysis-required to ensure that AI-generated answers include detailed reasoning. He argues that raw solutions without context provide little value to the mathematical ecosystem, though he admits that banning AI is currently infeasible.