upvote
No, and this is why we're fucked. The benefit of having a proof completed is having access to the techniques which completed the proof. Without an understanding of those techniques we'll be applying the result blindly, without understanding of its consequences or context.

And I doubt the actual proof will be released for months-to-a-year at the very least. Until then we can only rely on Lean and AI's sloppy interpretation of it.

reply
> we'll be applying the result blindly, without understanding of its consequences or context.

the AI will understand these and apply them accordingly. You don't have to.

reply
I think so. My understanding is that the proof contains many re-proofs of existing, already known results. It may also contain multiple proofs of intermediate lemmas, as LLMs frequently implement things over again rather than reusing an existing implementation. It's very possible that only a small fraction of the proof contains new mathematics.
reply
People claim to understand the OpenAI 150-page quasi-Riemann proof, which is apparently full of extraneous detail. Alpoge speculates that this new proof also simplifies considerably.
reply
so this is 1,000 times longer, so hopefully it does simplify.
reply
It sounds comical, but we can just wait for a smarter AI to make the proof more digestible.
reply
maybe having a big proof allows for reduction into smaller proof? at least in code that often works
reply