upvote
This kind of proofs don't help humanity at all. AI uses a gobbledygook language that's incomprehensible by humans and the only outcome of its proof is a 0/1 result that you either believe blindly or not, for nobody can really verify it.
reply
This is not true. In the processes I've read about, AI produces natural language proofs first, which it then translates into Lean code which is verified by the Lean kernel.
reply
I read the Zulip chat and it sounds like sometimes the formalization is completed before the natural language proof.
reply
Another comment, I haven’t verified this, suggests this particular formalization is upwards of a few million lines of code. I really enjoyed this article from Wolfram where he argued that if we had a box that could prove anything it wouldn’t necessarily matter much because the important part is the application. Now AI might, will(?) provide both but when it comes to the proofs it’s hard to tell how useful they are _at this very moment_.

https://writings.stephenwolfram.com/2026/09/whats-the-future...

reply
AI only imitates natural language. Nobody can read and understand the 2500 pages of Claudese. Some may argue that AI can explain each part of the complex proof and therefore the entire proof is human readable, but if this was true, AI could produce a compact human readable proof in the first place.
reply
That assume a compact human readable proof exists in the first place. Can you prove that?
reply