A majority of these proofs have not been formally verified yet, I think people are overstating how important lean is to the success of LLMs in mathematics.
replyA paper and a lean proof are always going to be better than just a paper. I think mathematicians generally will not read AI math papers that haven't already been verified, especially since we're about to see a ton more AI math papers. Lean will remain important
replyAre there any AI generated proofs that are simple enough to be verified quickly by a human, that have not been lean verified? Or are they all basically incomprehensible?
replyThe approximation of edit distance result [1] seems pretty readable to me, but the learn proof is still incomplete [2]. It's certainly much less readable than a good human written proof but it's certainly better than the last generation of AI proofs.
reply[1] https://github.com/openai/math/blob/main/preprints/An-Almost... [2] https://github.com/openai/math/blob/main/lean/ComparatorChal...
If you think AI-generated Lean proofs are unreadable, imagine Opus 5 generating informal proofs.
replyI think OP is saying Lean does indeed help.
replywhoosh
replyOpus 5 is ancient history now. Move on.
replyYeah! They forgot to put a .5 after it! What an idiot! Just imagine if they would have written a 4!?!? We may have had to ban them from the website entirely.
replyThe point both are making is that 5.5 produces readable output and 5 to a significant degree did not.
reply