upvote
It was rushed out before being fully checked, not a huge surprise. They even say in their blog post that there are likely gaps but they hope they are fixable.
reply
I guess they leaned too much into "purely human developed" and didn't even show their proof to any LLMs before publishing it.
reply
And why would they do that when there's already precedent for AI companies stealing a mathematical breakthrough that's been put in their system?
reply
It’s a 1,000 pages dense document. I’m not even sure the PDF conversions are accurate enough to feed it ti an LLM.

Also - if that’s the future of math (larger and larger proofs and algorithms, to the point they are far beyond a human’s capability to comprehend, then I wonder how mathematicians work would look like.

reply
I recall being told that there are certain axioms, the axiom of choice for instance, which we take most of the time not because its the best tool for the job in any intrinsic sense. Its the best tool for the job in a historical sense. We use it because our parents did and it's convenient to stand on the shoulders of giants.

But if proofs are cheap, you don't have to stand on the shoulders of giants. You can instead venture where they did not.

I'm told that the axiom of determinacy--used as an alternative to the axiom of choice--represents such an underexplored alternate dimension. Presumably there are many of these: inconsistent with the math we know but internally consistent and potentially superior to our use cases than what we now use... Our current habits being more driven by historical accident than careful decision.

So a future mathematician, rather than being a Platonist, might be a pragmatist and a curator. Given the many possible formulations for mathematics, which is most fruitfully paired with economics? Which goes best with chemistry? Which parts appeal to the human sense of beauty and how can those be introduced to specialists in disparate fields such that those specialists can now, through a bit of shared structure in the formalism that underpins their work, better understand each other's domains and better collaborate about them.

I think if we tailor a mathematical theory to a job, rather than assuming that the math is primordial and immutable, that the humans doing that job will benefit. It seems like it could be pretty rewarding work to me. Maybe AI will make it practical.

reply
You can cut off the appendix so it's roughly 500 pages long and Gemini 4 can digest it in one go (context window is around a million, this takes up around 300K tokens so it's inside the effective window).
reply
>I’m not even sure the PDF conversions are accurate enough to feed it ti an LLM.

It's best practice to feed in images to the LLM. It works really really well.

reply
deleted
reply
deleted
reply
[flagged]
reply