AI-assisted mathematical proofs: Tool effectiveness and cognitive boundaries

Deng Yu's interview isn't surprising, but it deserves serious attention. A Fields Medalist publicly admitting to using GPT to solve specific mathematical cases is itself an experimental sample. His phrasing is quite restrained—he lets AI handle "simple conclusions that are basically confirmed as true," which perfectly corresponds to the so-called "low-risk reasoning" scenarios in the NLP field. From my research perspective, the key to this experimental design is: AI's role in mathematical proofs is as a verifier, not an explorer.