If math is more than proof, we need to better celebrate the rest of it

Sanderson argues that proofs were always a proxy for human understanding, and that as AI systems produce proofs on their own the field should raise "motivated explanations" (accounts that make a result feel like "you could have discovered it yourself") to a status similar to solving open problems: posing exposition problems like research problems, setting standards for them and counting them in hiring and tenure. He uses GPT-5.4 Pro's resolution of Erdős problem #1196, and the human work afterwards to interpret it, as an example. Tao notes the post was converted to his blog's format with AI help.

(blog post; summary only)