Mathematical capabilities of LLMs have improved dramatically, solving major outstanding problems. But the push by AI companies to use math problems as benchmarks is detrimental to mathematics and its community. Goals of AI companies and mathematicians are severely misaligned.

Research mathematics deals with understanding basic structures. Famous problems serve as landmarks. Solving one has been a sign of new insights, studied by mathematicians through talks, discussions, simplifications, ideally leading to textbook presentations. Some ideas become tools used by all.

The mathematical community is a miniature humanity, joined by core values. Its most precious resources are students and ideas, nurtured with care. For students, problems develop skills. Ideas are disseminated in talks and careful writeups, taking time and human interaction.

AI's success in solving major problems has made headlines. But solving problems is only a tool for conceptual understanding. Mass production of "true/false" statements could destroy fertile ground. Solutions are announced in a rush, leaving no time for proper writeup, isolation of new methods, or citing previous work. This raises severe attribution and plagiarism questions. Without willing mathematicians, AI-conceived ideas would never become fully alive.