Post #2644273

@jonmsterling@mathstodon.xyz (An amateur's opinion:) More generally, beyond just being validity certificates, I've always felt the value of proofs is more in bridging fields, building new frameworks (like you say), and on the more constructivist side, providing a methodology to solve a problem (I'm primarily thinking of proof = algorithm in compsci). Proofs by an LLM I would conceptually consider as a stochastic proof tree generation process, based on existing patterns in the literature. So a random collage that links axioms / lemmas / theorems to the goal statement. While random meanderings through bits of existing work might hit upon new ideas, I very much agree there is the challenge and burden to ontologically fix (and explain) that tree.