> Let’s say an AI supplies an incomprehensible, ugly proof. Then you can work with it, and you can analyze it. Suppose this proof uses 10 hypotheses to get one conclusion—if I delete one hypothesis, does the proof still work? That’s a science that doesn’t really exist yet because we don’t have so many AI-generated proofs, but I think there will be a new type of mathematician that will take AI-generated mathematics and make it more comprehensible.

My thoughts exactly about designing public APIs of code. A role which traditionally performed only by experienced developers, now it can be greatly simplified and accessible to everyone.