AI and Human Mathematicians: A Serious Reckoning

Two threads today take the question of AI replacing expert human thinkers more seriously than usual. 'Why Do We Need Human Mathematicians Anymore?' links to a post by Po-Shen Loh arguing that AI math capability is advancing fast enough to challenge the premise of the profession. Separately, the thread on whether math needs to 'celebrate more than proof' argues that mathematics is suffering from Goodhart's Law, and that AI may soon produce superhuman mathematical exposition.

The debate in the comments is genuinely interesting. Some argue that understanding, not just proof production, is what matters, and that AI can verify proofs without understanding them. Others push back that this is just credentialism dressed up as epistemology. The RSA-896 factoring story adds a concrete data point: someone used idle GPU cluster time to crack a significant cryptographic challenge, treating compute as essentially free once the cluster is paid for.

The pattern is that AI capability in formal reasoning domains is moving fast enough that smart people are asking serious questions about what human expertise is actually for, not as a hypothetical but as an operational question.