post
Isn't this "Independent mathematicians who also used Lean software to check the proof that the author wrote/checked with Lean"?
It could still be valid of course, but I thought the proof was like 250000 pages and has only been looked at by the same automated framework. This is out of my wheelhouse and I don't have a dog in the fight, but I'd still like some expert human eyes on it
There are a bunch of ways that a proof written in Lean can be invalid.
The most obvious is that the problem statement is invalid or incomplete. This was not really a concern here as the proof used Mathlib's built-in definitions rather than restating the problem, but nonetheless it has been verified.
Beyond that Lean proofs can contain custom axioms/assumptions, shortcuts, tricks, exploits. And there's the possibility that the Lean compiler used in the proof had a bug.
Anyway all that has been tested and verified. The problem statement is valid. The proof doesn't contain any tricks or assumptions. It compiles on every major Lean kernel.
Basically: it is almost certainly mathematically true. The next step is humans actually understanding it.
They used 2 checkers that are maintained and developed independently of the Lean kernel. But they did use Lean formats/syntax. They've also done some manual verification, though there might be more to do. The proof did use the statement if RH directly from the well established Mathlib.
I've also seen claims that the proof has some weird things in it that might be hints that it is finding holes in the theory Lean models.
Honestly, even if it ends up not being a proof of RH, it will tell us something about computable / constructive maths.
It's 250,000 pages of explanation for 17 million (I think) lines of lean code.
It's incomprehensible, and therefore probably of little value.
It's incomprehensible as a whole but not in pieces.
It will be simplified, refactored, maybe rebuilt from the ground up. It won't be incomprehensible for long.
The last I looked the size had already been reduced by 25% based on community efforts.
Kind of fun watching the Lean Prover Zulip chat play with it.
Can someone ELI5 the implication of this proof?
The short answer is "not much."
Us assuming that the Riemann hypothesis is true is the basis of much of our mathematics. If it was proven false, it could affect a ton of the math we've built on top of it for centuries, but it being proven true is just confirmation of what everyone was sure of but couldn't quite prove.
We've confirmed something that was long suspected to be true, but learned nothing in the process. Hard to say if humans can even understand the proof. It's true... and that's it.
Usually the importance of a problem like this is that the journey to solve it leads to major breakthroughs that can touch other fields of mathematics. So it's a bit surreal to see the most famous unsolved problem finally felled, with the caveat that we learned nothing in the process.
It’s counter proof would have been monumental. This is big. Perhaps the biggest of ours and a few generations surrounding, but not practically so.
It changes little. It’s the result we’ve come to expect. Some other conjectures will use it, maybe Goldbach.
But the disproof would have hurt a lot of number theory, maybe engineering, and probably crypto.
Could primes being predictable result in a weakening of the design of encryption because it would be easier to calculate large primes?
No, not directly at least. Encryption relies on the difficult nature of factorisation, not knowing if a specific number can't be factorised (is prime).
- Possibly some of the internal components of this could be extended to help factorisation problems (but also possibly not)
- Most modern encryption algorithms moved away from primes because of other weaknesses like quantum factorisation anyway
So there are prime numbers, that are only divisible by 1 and themselves. Here's a set of all the prime numbers bellow 100:
[2, 3, 5, 7, 11, 13, 17, 19, 23, 29, 31, 37, 41, 43, 47, 53, 59, 61, 67, 71, 73, 79, 83, 89, 97]
As you start getting larger and larger numbers, primes show up less and less. The Riemann hypothesis is that there is some pattern, some formula to the distribution of prime numbers.
So if there is a pattern and you had the formula, you could predict which arbitrarily large number is going to be prime. Otherwise, you just have to find a big number and check if its divisible.
The other poster said a ton of math is built upon that idea? Like what and why?
Sorry I know asking this is alot, im just curious.
The Wikipedia page lists various consequences. Mostly things that will not mean anything at all to a layman. The other thing is that claims that "all of modern mathematics" hinges on the Riemann hypothesis are extremely overblown; it's mostly consequential to number theory, which is something many modern mathematicians work on/intersect, but there's also many who never touch it and for whom the Riemann hypothesis being true or false changes nothing.
I wish I knew for sure, but I don't.
Best guess, I can say that a lot of mathematical profs work by looking for the simplest description, then building up from there. Primes are as mathematically simple as it gets, so its kinda important to keep in mind there's "simple" numbers hiding out there in the unthinkably large digits.
We love using big primes for public key cryptography.
That's some black magic math that lets you take a pair of big primes. Keep one for yourself, give the other to the world. You can then encrypt or sign things with your number, and everyone with the public prime can decrypt or verify that you signed it. And vice versa, people can encrypt things that only you can decode.
As I recall, it all relies on the fact that it's a lot of work trying to find suitable primes for this. Knowing the pattern to how often primes occur can make it easier to break. This proof only predicts how many primes we'll find, so it's the basis for it.
On Friday Anthropic mathematician Levent Alpöge claimed to have used the company’s Claude AI model to solve the Riemann problem, perhaps the most important open quandary in mathematics.
"We have full confidence that Alpöge's proof of the Riemann Hypothesis is correct, and we congratulate him on this achievement," mathematicians wrote in a joint statement by the Lean and Mathlib communities
Are we calling it "Alpöge's proof" despite the fact that he didn't come up with it?
He has a writeup describing some of the process: https://alpo.ge/remark.pdf
Maybe we should just call any proof that uses calculus Newton's proof.
I think it's unclear how the LLM was involved, I doubt the LLM came up with it itself. It's a tool and assisted him but he ultimately is responsible for it. If it were garbage he's the one who will take the reputations hit.
I doubt the LLM came up with it itself
In math, these things are attributed to those who can actually complete the proof. Maybe this could be called Alpöge’s hypothesis of the Riemann problem's proof or something if he did come up with the idea.
I don't think ownership is determined by whose reputation would take a hit. I mean, at this point, the Lean and Mathlib communities would also take a reputation hit if it was proved wrong, but it doesn't make sense to name it after them.
You think Claude should be credited instead?
Maybe all mathematicians that ever published anything should be credited? After all, the plagiarism machine can only predict characters based on what it's been trained on.
Claude isn't a person, so no. I think if Claude did the important work, then it shouldn't be named after anybody.
The proof just... showed up, then? Nobody made it?
Nobody made it?
Is that so weird, though? I'm drinking soda from a can that nobody made. My desk is covered with things that nobody made. Bots make tons of stuff and we don't name those things after the bots or their operators.
There's no reason to think that a "body" has to make something just because it exists and is complicated. That's the same mistake that creationists make instead of understanding evolution.
But that's all off-topic. The question isn't where it came from, but who it is named after.
That's already long established. It's the Riemann hypothesis, it's right in the title.
I'm not sure what you're suggesting at this point. When it's published in a journal are you suggesting the byline should be left entirely blank?
You have completely forgotten the context for these comments, so rather than asking me to explain things, I think it would be more straightforward for you to just go back and read the comment chain.
Alright, so should it have been:
"We have full confidence that nobody's proof of the Riemann Hypothesis is correct, and we congratulate nobody on this achievement," mathematicians wrote in a joint statement by the Lean and Mathlib communities
Because that seems quite unhelpful to me.
If the question is how to refer to the proof using the English language so that people won't be confused, they could have said "this proof."
And they can still congratulate people involved. That's obviously not what we were talking about, since that has nothing to do with naming the proof.
Alpöge did contribute to the work, though. And if someone else comes up with another proof it'll need something to distinguish it.
I think this is really getting into meaningless quibbling. These things need labels to distinguish them, why not Alpöge? It wouldn't have happened without him directing Claude.
I think this is really getting into meaningless quibbling.
This was the topic from the start. If this is meaningless quibbling, then my original comment and your initial reply were meaningless quibbling.
These things need labels to distinguish them, why not Alpöge? It wouldn’t have happened without him directing Claude.
I think most people have decided that an "AI artist" might be able to direct the AI to create something beautiful, but it seems wrong to call them the same type of artist who creates things themselves. This is the same kind of discussion.
then my original comment and your initial reply were meaningless quibbling.
Can confirm.
At the end of the day none of us are the final arbiters of the English language, it's a collective decision and while certain organizations and principles might weigh heavily on it, even they don't really get to decide. We're certainly allowed to put our votes and our arguments out there but yeah, we're definitely just quibbling. It's going to be called whatever it gets called, and what we think it should be called is pretty much completely irrelevant.
And as long as I'm here, I'm going to say that I think it's perfectly reasonable for someone to claim credit for an AI's work if they're the ones who prompted it, and that is neither dishonest nor unreasonable as long as we know that AI was used and to what extent it was used.
As you say, Claude is not a person. I agree we certainly cannot and should not give credit to it, no matter how much of the work it did. But that doesn't have to mean nobody gets credit. Some of the work was done by someone else, no matter how dismissive we might be of the quality and comparative weight of that work.
FaceDeer seems to be raising the concern that Alpöge did not do a significant amount of work, and thus should be denied credit as well. I'd dispute that. I think if nobody did any significant amount of work except for one algorithm that can't be credited, then I would suppose credit can be claimed by whoever wants it on a first-come-first-served basis, even if they have only done an insignificant amount of work on the result, there is simply no one else to give any credit to and they can claim it by default. The person who prompted the algorithm is the most logical choice to assign credit to, given that there is really no other plausible or sensible choice and "nobody" does not seem like an acceptable or practically useful choice.
Given that AI was trained on all the collective works of humanity, I think humanity deserves credit too. But we already tend to give ourselves credit for everything we do, and like "nobody" that is also too broad and not a practically useful choice. We are going to have to start to distinguish a lot of these kind of situations in the future. Naming a particular individual person to give them credit has worked well for us in the past and there's no reason to stop using that convention now simply to encompass a different almost equally arbitrary point of view of correctness.
"Most people" is a very context-sensitive thing in this age of social media bubbles.
I suspect most people don't care one way or the other.
You don't HAVE to name mathematical objects after a person. We could just call it Riemann's Theorem (a proved hypothesis is a theorem) and include the date and by-line if we need to reference a specific one in (e.g.) a bibliography.
Decolonize maths!
Related reading: https://www.theguardian.com/world/2016/jan/06/monkey-selfie-case-animal-photo-copyright
As a non-mathematics person, this still depresses me so much. It's like like a kid finding out there is no santa. Or a religious person realizing there is no god. I can't really quite describe what about this exactly depresses me, but I feel like ground has been taken beneath my feet. Not just this, the whole llm thing. Not because I'm afraid of losing my job or status. Just that the magic is gone makes me feel sad.
Probably because human thought and intellect are the things that we, as a species, see as differentiating ourselves from any other animals on this planet. And so these intellectual and creative domains feel (and, in the past, were) exclusively the domain of humanity.
Automation is now taking that away.
What do we do when it seems like we, as a species, are on track to render ourselves useless?
I don't know.
I wonder if this is what old world explorers felt like when the globe was fully mapped and there was no true frontiers anymore. What do you do when the thing that gave you a unique sense of purpose disappears in the face of technology and automation?
What do we as a society do when everyone experiences that at the same time?
Honestly, I think CS research has hinted we were going in this direction anyway. Dependently-typed languages like Idris and Agda have as term-inference for a while, it was based on exhaustive type-directed heuristically-prioritized search but it could still auto-complete/fill some non-trivial code, and Roq tactics could also make things that were disastrously wordy read much more simply and they are effectively a form of compile-time function (Elaborator Reflection is probably a better approach, but Roq's two-language model is much more well-established).
Even without quite so interesting types, there were academic exercises like "Djinn" that could exhaustively enumerate lambda-calculus function bodies given the parametrically polymorphic type signature and a library of standard functions, and that's been available for 14+ years? In an entirely different direction we've had SMT solvers can could be plugged in to a few languages to verify a LOT of properties that can be expressed in first-order logic.
Still, I think there's plenty of magic, it's just not in the places we can lay down (enough) axioms around symbolic structure / grammars.
I guess I never thought human intellect was very special, and when I learned about neural networks, this sort of AI progress felt inevitable. It's all still pretty mindblowing, but not completely inevitable
Is it even possible to distill this down to the type of proof a human would produce that is orders of magnitude more concise and has coherent meaning, or is it just largely slop that would be easier to start from scratch with the monumental task of a human-made proof than it would be to painstakingly unravel it?
Probably some of both. Though, given the short history on these proofs, it will likely be more "slop" (or theft) than anything else.
While this is orders of magnitude larger, the first 4-color theorem theorem had a lot of cases that were quick for a computer to exhaustively check. But, it turns out a lot of those cases are better explained without exhaustive checking, so the currently taught / best proof of the 4CT is shorter and has less long, mind-numbing checks.
It's entirely possible that there's actually no more "insight" from the LLMs, and they were just better able to generate the massive type signatures and terms that Lean needs to begin mechanically checking, because they are able to generate and discard good syntax much faster. But, there might also be some interesting bits that people (or LLMs) will be able to extract/abstract/generalize/shrink into something humans will find insightful. I think the Navier-Stokes related proof did have some stuff in it that humans found interesting beyond the final result.
Does this mean I can conjure up a crazy prime number with a little algebra now?
Riemann isn't a prime generator, it's a prediction of how many primes exist in any contiguous sequence of integers.
No more so than you could do before - we've been assuming that Reimann is true since forever, a lot of modern mathematics depends on it. And we already have ways to conjure up extremely large numbers that are almost certainly prime, such as Fermat's primality test. RSA encryption depends on us being able to do that, in fact.
Largely unrelated. Generating primes has always been relatively easy. All you need is a random number generator, a primality checker and the ability to keep adding one until the primality checker says it's prime.
Now if you want to count the number of primes below a certain number, you're moving closer to Riemann Hypothesis territory, but I don't think this result will help much with that either.
I mean, we're still at 1 here, there's a reason it's outputting Lean code rather than actually doing any computation by itself (Because it can't do that, it's a text generator)
The nerds have had their run. Time to give the jocks another turn and see how things go.
People who don't need to put their dick everywhere because they already did... in high school.
Or maybe the artists. Let's have them run things.
I really do not know what kind of person would comment such an ugly piece of text, but you should realize that ALL of this is the work of nerds.
And not all of them have dicks.
And some might not want to fuck everything that moves.
And most of them have a completely normal social life the moment you remove the jocks.
It's an extremely bizarre comment
If someone figures out how to set up a fediverse account, I hate to break the news, but this is 110% nerd activity. We're all nerds here.
What the fuck qualifies as a jock these days anyway? Does anyone still use that term unironically outside of a Hollywood writer's office?
You're good at sportball? You learned all those sportball rules? You keep a catalogue of all the best sportball players? Good job, nerd.
I seriously can't even tell what point they're trying to make


all 70 comments