The quote from the article affirming it is true:
Honestly the chance that anything is fishy here is approximately zero.
The odds of RH being false have been approximately zero for an eternity now!!! The whole point of a mathematical proof is that “approximately zero” is not irrefutable evidence! This is worse than worthless, it tells us nothing that wasn’t understood before, and gives a headline that allows people with no experience with what mathematics actually is to declare mathematicians obsolete, gives administrations justification to cut funding.
The fruits of mathematics have never been in knowing whether a specific statement is true or false, the fruits have been what’s fallen out of the creativity used in solving these very hard problems. The concept of a scheme, now the backbone of modern algebraic geometry and a critical concept for modern cryptography, is something that Grothendieck developed in solving the Weil conjectures. This concept is a million times more important than knowing that the Weil conjectures are true. If we had had a machine say “the chance of the Weil conjectures being false is approximately zero” and mathematicians felt discouraged from pursuing their own investigation, then we’d have lost out on the actual world-changing discovery that came out of solving the problem. The creativity that’s come in the process of proving hard results has invariably been the lasting impact of almost all research problems. People who can’t define uniform convergence will use purported “proofs” like this to kill the actual purpose of mathematics research.
It’s not saying the odds of the RH being incorrect is almost 0, it’s saying the odds that the LEAN code was intentionally written to give a false positive are basically 0.
it’s saying the odds that the LEAN code was intentionally written to give a false positive are basically 0
intentionality is not the issue here, dude

The issue is rather that the Lean Prover itself has had many soundness bugs, including some accidentally exploited by AI-generated proofs, and the odds that it does not have more are basically zero.
here are some links
- https://mathoverflow.net/questions/513540/should-we-trust-ai-generated-formal-proofs-in-lean-4 (July 2026)
- https://siddhartha-gadgil.github.io/automating-mathematics/posts/lean-kernel-bugs-and-junk-values/ (August)
- https://terrytao.wordpress.com/2026/10/09/what-mathematicians-should-know-about-the-lean-theorem-proverquestions-of-reliability-and-ai/ (October)
see also the list of soundness bugs on djb’s False computer-checked proofs page.
Thanks, those links are great! Especially the text by Hales.
I do not believe that mathematicians can be entirely satisfied with a system that claims to be a type theory but that cannot even promise that well-formed terms have a unique type, up to definitional equality.
This seems like a bit of an oversight. Also seems they missed an error in the foundational lean 3 thesis and when they found it, everybody was already speeding along on the lean train. Now lean 4 doesn’t have a similar foundation. Had the error been found immediately after publishing the thesis in 2019, it might not have picked up as much steam. Also, even if unique typing is ever proven, definitional equality will still be undecidable.
Instead of saying:“Lean is based on type theory”, we should say:“lean looks a bit like type theory, but who knows?”
Oh, I missed that phrasing when replying myself, good catch.
First, watch your tone. Second, I clearly meant that some technique was exploited by the AI generated code to give a false positive. Intentional or accidental both literally mean nothing when talking about something a machine generated.
watch your tone.
Sorry, i didn’t really intend to invoke the tone of that scene per se, but somehow felt compelled to reference the line when disagreeing with your statement about what the independent mathematicians’ statement is saying.
Obviously, but I feel comfortable conflating them because I don’t see a reason to find that distinction important in response to an article titled “The AI Riemann hypothesis proof is true, independent mathematicians confirm”.
One is human error that can be corrected, one is a fundamental fact of the universe.
could you not just point out, much more simply, that this is like a great math guy with a real good calculator? knowledge has not been advanced, a math guy just used a calculator to confirm what he already knew
the novelty is that the calculator printed a ton of pages
No, because I’m not ready to call someone with literally zero papers on the arxiv a “great math guy”, because the accuracy of the calculator is the question mark in its validity, and because a casually dismissive summary like that completely downplays how legitimately harmful these headlines are to mathematics as a whole. Someone with no mathematics background reading the phrase “knowledge has not been advanced” will likely not find it self-evident without further explanation.
I’m not ready to call someone with literally zero papers on the arxiv a “great math guy”
fyi Levent Alpöge (who you’re talking about here) has these papers on arxiv.
re: the recent effects of LLMs on mathematics, i recommend this post (a collection of reactions from mathematicians about OpenAI’s October 6 publications).
Thank you for the explanation, but I was just concerned about talking to laymen. The zero papers part is particularly damning, of course, but I do think someone with no mathematics background would read an actual explanation and fail to apprehend it. Not because they are AI fans, but because people tend to make good faith assumptions about fields they have no expertise in. They assume an universe of knowledge they have no access to, and might also assume that, hey, even with your explanation maybe this AI computer thing did something minimally useful regardless.
Edit: but hey if i am talking out of my ass then i dont wanna frustrate you, you can just ignore this
A blog post from ~7 months ago. Emphasis mine.
The fall of the theorem economy. How AI could destroy mathematics and barely touch it
If AI ever “wins” at mathematics, then someone else will have to be the “loser,” and the only other player in the room is human thought. Who cares if mathematics isn’t an adversarial game—and we are still figuring out the rules. When the news is this sensational, it becomes fit to print.
Here is what the nuclear scenario looks like. A year from now, an AI lab assembles a secret A-team, throws a few billion dollars of compute at the Riemann hypothesis, and comes up with a 2,000,000-line Lean proof. What are mathematicians going to say? That the proof is unintelligible? Right, but who cares? Isn’t the whole of math already unintelligible to most people?
That the theorem was still proven by the humans in the loop? Yes and no—and in that specific case I would probably vote no. That a few billion dollars represents decades of the entire mathematical sciences budget at the National Science Foundation? Correct, so what? And what if the conjecture happens to be much easier, accessible with only nominal resources?
That the conjecture itself was still produced by a human brain, along with the entire mathematical stack underneath, starting from the very language in which the conjecture was written, the concepts of infinite sums, complex numbers, meromorphic functions and analytic continuation, the interpretation of the conjecture as a profound statement about the distribution of prime numbers, and the prophecy that it should be true? And that, at this point, humans are the only beings who care about it?
This objection is profound and valid. But when you look like a sore loser, nobody cares what you say. I don’t expect this exact scenario to unfold. But I do expect to see, in the near future, one or more claims that AI has “solved math”, or “achieved mathematical supremacy.” And, given the current state of the conversation, I expect the actual evidence supporting these claims to be framed and/or understood as a defeat for human intelligence.
That a few billion dollars represents decades of the entire mathematical sciences budget at the National Science Foundation? Correct, so what?
I get that they’re saying “so what” as in “what does that mean for public perception”. But this is a serious point I constantly bring up that never gets mentioned.
Precisely what is the point of AI solving mathematical problems if it costs more than just training and hiring mathematicians who would also get like 1000 other things done with the same budget?
Damn that’s some prescient shit
yeah but can the AI tell you how many 'n’s are in the word riemann
I shit you not, first try:


Just having the lean doesn’t prove it within a shadow of a doubt
Love when Philosophy comes in like a wrecking ball.
The Millennium Prize for it can only be claimed two years after it has been published in a peer-reviewed journal, and it isn’t clear if/when that is going to happen. Alpöge said he is publishing only Lean code for now because it would be "obviously inappropriate to attempt to post some massive unreadable travesty of ‘slop’" - Claude produced a 250,000 page book on the subject when he asked it to. The Lean code he is publishing, meanwhile, is 17 million lines. 🫠
if you’re gonna use the slop travesty generator you’re not allowed to feel shame about using the slop travesty generator
I have seen it said for other mathematics proofs that the AI just followed a given technique for much farther than any human had, in which case the machine is doing exactly what a machine should do: making long, exhausting labors faster and easier. I wonder if it is the same situation here or not.
All this AI stuff must suck for mathematicians as much as it does software developers, artists, and writers.
penrose was wrong
clanker solved this, people checked that, anthropic ipo hype scam etc, i think we’re missing the real point here and that is what the fuck they’re talking about is that made-up type math

grains of utter derangement from incurable poindexters. how does “proving” this madness help the wheat harvest. how is it helping us meet the steel quota.
can this math isolate the hitler particle?
Maybe this math was the hitler particle the whole time.
The thumb gave me a pang of nostalgia. It’s almost certainly a fractal made with the Chaoscope app. I used to use it. It was amazing. It’s a strange attractor fractal program that creates 3D wireframe fractals on the fly. Rending took ages of course. The program was abandoned in 2010. I stopped using it sometime after because it was an obscure program, there were no 3rd party goodies, and as cool as the fractals were the textures were very limited.
Ninja edit. play.google.com has it but it must be a limited toy. You can’t do much on a phone. And the app’s website no longer exists.










