Home | Notifications | New Note | Local | Federated | Search | Logout

Note Detail

Reply to @johncarlosbaez@mathstodon.xyz
Thomas Kahle@tomkalei@mathe.social (2026-09-24 02:46:37)
@johncarlosbaez 

Why do you think mathematicians should engage with OpenAI at all after what happened / is still happening? 

Why do “we” mathematicians need to give them input?

Feels like talking with the schoolyard bully about “our problems and how to resolve them”. 

I think we also have the option to ignore them. Let them dump their 100 theoremoids on the Arxiv or their homepage and then we’ll see. 

This is how science works. Anybody can post anything and then peer review and so on…
---Reply---
John Carlos Baez@johncarlosbaez@mathstodon.xyz (2026-09-24 03:02:10)
@tomkalei - if OpenAI publishes 100 proofs of big theorems - verified using Lean, but not carefully written and not properly citing previous work - it'll be worse than if mathematicians have a chance to take this work, see what they think about it, and turn it into properly written mathematics before it's released.

You say "we also have the option to ignore them", but few mathematicians working on a field where the blah-di-blah conjecture is a central open question will simply "ignore" a proof or disproof, verified using Lean, of this conjecture.  Some may ignore it.  But others will will go on and use this new result, and those who ignore it will be left in the dust of history.  Publicly available facts don't usually disappear, and boycotting a fact doesn't work well.

---Replies---

Thomas Kahle@tomkalei@mathe.social (2026-09-24 04:33:12)
@johncarlosbaez 

I agree with what you write, that it would be better if those 100 proofs would come out digested, but I just see no chance for that, in particular not by somehow working with openai.

I'm also not saying we should ignore those proofs! We should just ignore openai and focus on the proofs!

My view is a bit like this: As soon as a model exists that can produce these proofs, these proofs exist. All of them. Just exposing them, sending a prompt, spending a million dollars only somehow prints them. But in my mind they are already here among us.

It's a bit more complicated than that because the model is not released but in a short time a model with these capabilities will be released. That time will be shorter than digestion time anyway.

And once that model exists openly, we will be back in the situation we have after the release of astra: random strangers on the web sending us solutions by e-mail, asking us to vet them, everybody being able to produce theoremoids, etc.

It is super sad, but I think it is better to accept reality and embrace open science. Put all the available information out there and then let's work with that, start digesting etc. Let's do that in the open and let's not overstate openai's role in this. 

The academic credit system has already broken down. We should accept the reality and build something new, as a community, between humans. BigTech should have no role in how we structure our community.