Post

It's kind of like being told that the meaning of life is 42

Probably true. Kind of like the value of learning math isn’t to get the right answer but to learn the skill of solving problems.

The arXiv: uh, guys, we're overloaded with submissions blog.arxiv.org/2026/10/01/u... OpenAI: no worries, we got ya fam

My sense overall is that it has been good for doublechecking that proofs are solid, *but* the increasing reliance on computer program A telling you the output of computer program B is correct has called attention to the peculiarities and shortcomings of program A.

Not an expert but these are both features, not issues... when you divide you just specify that denominator cannot be 0, and - is just monus in lean, these definitions are morecommon and make theorem proving simpler, they're not quirks en.wikipedia.org/wiki/Monus

Some parts of Lean may appear odd at first, but it is extremely useful in math. Experts use it, including famous ones like Terence Tao. A proof in Lean is considered a valid logical proof by mathematicians, uncontroversially. (That does not mean Lean does not have bugs, like any software! It does)

On one hand, here’s a defense of some of these choices xenaproject.wordpress.com/2020/07/05/d... On the other, I strongly agree that this makes it much harder to find issues in proofs generated by AIs

We all trust Lean. It's a mechanist proof system. Therefore, the only thing you need to check is that the thing it proves is actually the same as the thing you care about. That does not leave much room for funny business.

arxiv.org/pdf/2610.08144

Yes, but it's also true that we (as in “the mathematical community”) were unable to come up with an answer as to how OpenAI should release these results so as to mitigate the harm:

Terry Tao has a great explainer here. youtu.be/svl_1upFpQo?...

Well, in the simplest terms, is it useful if Kuncklebunkle's Theorem is proved true, regardless of whether humans understand the proof? So, no understanding, but now we have a new tool in the toolbelt.

That’s the price one pays for selling their soul to the AI devil in exchange for the tokens needed to solve their favorite math problems.

Get on it Slop Jockeys! Very the literal shit from our butts! You're not against progress are you??

it’s just a flex–funneling vast compute to signal what precisely?

These people at OpenAI must be so dead inside.

Hah, guess that whole community letter wasn't in the training data!

some basic considerations and general rules for AI and automation... almost everything these companies have done and everything they are trying to do needs to be banned. general strike is only path

As always, there’s a relevant Futurama clip

Well, strictly speaking, yeah… so what else do we call it? "Ruthless vigor" perhaps? 🤗

Software Engineers:

Those automated solution finders for problems where a human mathematician can pre-formalise it: fine. But asking the LLM to do the formalisation as well is vibe mathing. Analogous to when vibe coders think the LLM understands the codebase.