I believe we're seeing a new kind of mathematics that will require completely new formats for publication, a bit similar to those used in experimental sciences. AI-powered mathematics should be fully reproducible, so it's the authors' responsibility to disclose the exact model type, inference settings/seeds and the full prompt history leading to the result. Of course that would ideally require open weights models.
It's not just about requiring to disclose AI use. AI-powered mathematics is a completely valid discipline that doesn't need to be shy, but it should develop its own publication culture.
I can see this being important if you only care about the results as evaluations of AI progress, but if what you care about is the math itself why should you care about the prompt or anything other than the proof?
Understanding the process that led to the proof helps to understand how to do further work on top of it, which is the goal of most mathematical research. It's not as though mathematicians are going to go launch a startup operationalizing their knowledge of how densely hyperspheres may be packed.
loading story #49136804
Because the math isn't solely about the proof being correct. You don't need to take my word for it, here's one of the most famous living mathematicians' take on it: https://teorth.github.io/tao-web/slides/age-of-ai-icm-2026.p...
loading story #49136959
loading story #49136894
While you may want AI results to somehow "not count" if the methods weren't disclosed, that doesn't present these results from poisoning the well for others. Once a result (with verifiable proof object) is delivered, the problem is solved, regardless of whether methods were disclosed.
Methods are only really necessary for results at a meta level, about the design amd evaluation of AI math systems.
why is reproduceability the thing?
shouldnt the paper be the math of the argument? the reproduction is reading the following the proof
I think this is a bit optimistic compared to my view (wrt portability). There's a large stack of software that is involved in training and probably less so in inference. I'm not saying it's impossible but there are definitely different levels of reproducibility and the academic incentive structure doesn't really prioritize reproducibility in my experience. I'm sure it varies quite a bit, I'd be curious to know how those in this problem space are thinking about reproducibility and at what level.
I know it sounds unrealistic and not aligned with academic incentive structures. But those are the exact structures that gave us a lot of headaches in the experimental sciences. I think it would be a good north star to aim for something that resembles how those are trying to address the reproducibility crisis. Better than to embrace the most black-box version of math that AI systems can produce (million-line proofs without context). Even if a reproducibility crisis is seemingly impossible (although agents so far have also been pretty good at finding compiler bugs).
If the proofs are formally verified by a proof assistant (Agda, Roq, Lean, ⋯), I see no reason we would need to know how these came about. All the information needed is in the proof.
Unfortunately, we seem to already have an example of an LLM producing a proof in a week known open problem (the Collatz conjecture) in which it looks like it was sneaking a flawed proof through bugs in the proof checker. https://infosec.exchange/@0xabad1dea/117002106099986943
loading story #49140639
What if the AI has discovered some new function F that allows it to generate (insanely large) proofs for a ton of theorems in a ton of different fields. Wouldn't you like to know more about this `F`? That seems to be the real innovation in this case. How much about it could be gleaned from the individual proofs themselves? What if this `F` is actually simple enough to be digestible by humans?