A fake refutation of the Collatz conjecture already got through one of these independent checkers in the past.
What's harder, proving GRH or finding a way to get through every checker (which aren't as independent as they are sometimes touted to be)?
In this case there was a human expert in the loop so it's much less likely the model hacked its way through the problem without the expert noticing. But the statement as written has 4 points, none of which rests on the human involvement. So I don't like that they claim full confidence based on those points.
If I understand correctly, for the purpose of the proof Claude formalized a huge chunk of all modern math. So it's a big result not only for GRH, but math foundations and formalization.
We need to give Levent a lot of credit for this. As much as we talk about Claude proving it, Levent seems to have a rare talent for getting the most out of it.
Do huge results like this open up even larger questions? Is there some question even more important that mathematicians dared not approach without doing big problems like Riemann first?
If you have a physics background and want to contribute to breakthroughs like this, check Physlib (counterpart to Lean's Mathlib).
It's making quick progress and is part of the Lean community.
Here's a human readible version of the proof:
This article posted a few days ago may be helpful to commenters that what to map their confidence of Lean proofs:
Interpreting claims that a theorem is correct because it was checked in Lean | 329 points 5 days ago | 148 comments | https://news.ycombinator.com/item?id=50515058
Any mathematicians care to explain what the ramifications are in regards to its impact on the broader world?
From my very basic understanding, RH was generally already assumed to be true?
This sounded pretty definitive to me. I'd like to hear from professional mathematicians about what's left for humanity to consider this proven.
IANAMathematician. This challenges the sole definition of a proof. Can a theorem be proved by proxy, without understanding the proof directly? Not to mention that its practical usefulness is exactly zero.
It seems the modern unsolved problems might have solutions incomprehensible by humans. So a new class of decidable problems emerges: llm-complete ones.
We are slowly approaching the Iain Banks Culture universe, with spaceships, societal structures or physics that people no longer understand or command.
> 17 million lines of code.
Can someone ELI5 how they know the proof is correct? It is not like anyone had the time to read 17 million lines.
There are still a few ways that the proof might be incorrect. Of these, the most likely is that the proof was generated by a misaligned AI system that found and intentionally exploited a lean bug in order to fool its human evaluators (i.e. shades of the HuggingFace attack). I personally don't think this is very likely, but you'd have to understand a bit more about the agent systems used in generating the proof to come up with a more informed estimate.
Another possibility is that an aligned system came up with a plausible but incorrect idea along the way, and simultaneously stumbled across a lean bug that led to an incorrect verification of this idea. I'd put the likelihood here as vanishingly small due to the requirement of serendipitous factors, but it's nonzero.
You could also ask whether an aligned AI system could "p-hack" its way into finding a serendipitous lean bug by trying a large number of approaches. I think this is also vanishingly unlikely because the AI system is trained on and therefore "thinking in" natural language, and presumably only translating results into lean after it believes them to be correct.
And of course there's the possibility that the lean proof isn't faithful to the natural language description we've been given, but that is ruled out in this case due to the use of the canonical problem statement in mathlib. Otherwise, this would certainly seem like the most likely way the proof could be incorrect.
for people feeling existential about this - i can sympathize with you but this just proves that human ambition has no limits. try not to feel limited by this.
when the first wheels were invented - humans could move at speeds (10-15 miles per hour, for hours and hours) that were impossible for the last ~450-500k years (basically since humans existed). a lot of humans thought there might never be a better design; then slowly and slowly(about 5000 years) we made progress and we had cars made by benz and ford which broke all those previous records (going 30-40, even 50 miles per hour) and then only ~70 years later, humans reentered the earth's atmosphere at whopping 20k miles per hour - read that number again, 20,000 miles per hour and that is the fastest humans have ever moved in the known history. do i think that's it? hell no.
> 17 million lines of code
I worked at a university for a time. One thing I learned is that while people in mathematics and physics and biology etc are brilliant, they are terrible at actual programming.
I wonder if there will be a push by the CS community to review the solution, refactor it, simplify it, etc. Surely that can help remove noise and express the solution in a much simpler way.
but is it useful??
because a million+ lives of code without any understanding how the proof works or any methodic to advance the field of mathematics seems to be rather not so useful
In all seriousness...
"con-leche and con-ron checkers"
Who comes up with these names? :-D
Can anyone give an idea of how long the GRH will take to test? I understood that this was also claimed, and is a much bigger deal, in terms of both impact and the effort needed to verify
As somebody who has no idea about any of this but speaks Spanish, I was amused and bewildered by con-leche and con-ron. Could somebody explain what those are and where those names come from?
I think it would be useful at this point if mathematicians across various fields worked to come to agreement on which specific sentences in a given formal language correspond to open conjectures. That way the question "Does the Lean proof actually capture the intuitive notion of the problem that we're interested in?" is already answered a priori by having those ready to go.
Purity of the math and understandability is one thing, but if the proof actually leads to some new technology or enables predictions of something useful to human society down the line, then I think it’s more than done it’s job.
“Any sufficiently advanced technology is indistinguishable from magic.”
We’re just seeing more and more smart people encountering Arthur Clarke’s third law.
What's next?
Any link to the code?
Isn't this the equivalent of aliens coming to Earth and saying
"Hello, human, your Riemann question is true, here's a book on it's proof, we didn't bother with translation, so you must understand our Vargalian clicks to understand our proof. In two weeks our sister civilization the Kikuviar will come by and drop how they've untangled your Collatz question in their base-30 logic! Thanks for the fish, have a nice day!"
With OpenAI having solved Navier-Stokes and Anthropic having solved the Riemann hypothesis, I wonder when we will see something exciting from Google, if they are also working on some Millennium Prize Problem.
Solving these extremely hard problems that humans, for decades (and more than a century in some cases), have been unable to is the mark of superhuman intelligence, in my opinion. So far only OpenAI and Anthropic have crossed this mark, and it would be great to have more to join them.
This is amazing.
It's fascinating to watch the skepticism in some of the reactions - "what if there's a bug", "it doesn't mean anything without human understanding" - just anything possible to undermine, dismiss and move the goalposts.
I am completely baffled by this given how dumb these models are when I interact with them, but the rational position seems to be to accept that Claude proved the Riemann hypothesis.
2030...Claude Opus 8.0 finds a bug in the Lean Verification of the Riemann Hypothesis
It's interesting that all their points acknowledge the possibility of error, but I guess they have as much confidence in the proof as it is humanly possible to have. All proofs are products of human meaning-making anyway.
> ... has been carefully vetted by experts in Lean and number theory.
So, if the vetting was not careful enough and done by people expert enough, what could happen?
. ... con-leche and con-ron come with a formal proof, relative to set theory, that they cannot accept a false proof, thus addressing concerns that Lean’s logic itself might have flaws that the proof could exploit.
And what verified con-leche and con-ron?
> These checks ran on an inert export of the proof, out of reach of hypothetical malicious code in the Lean proof source
Hmmm...
Guess I'm splitting hairs here; like I said, maybe this is as close as you get to formal proof verification, but still...
Do P vs NP next please
Random thoughts / questions:
1) Surely agents can be employed to simplify the proof now? "Make it smaller while it still verifies" in a loop; maybe, to a level where a human (research mathematician specializing in RH) can grok it?
2) con-leche is vibe coded; isn't it a concern w.r.t. trusting its soundness?
3) This is so exciting. Please prove P != NP already and spare us the tension
> The checkers con-leche and con-ron come with a formal proof, relative to set theory, that they cannot accept a false proof, thus addressing concerns that Lean’s logic itself might have flaws that the proof could exploit.
Could someone expand on this point? I still don't understand how con-leche (or con-ron) can be absolutely free of bugs/guaranteed to be consistent.
>as a Lean library comprising 17 million lines of code.
It looks like the meaning of the word "mathematical proof" changed fundamentally. Mathematics looks like a flattened animal on the asphalt by the CS steamroller.
For sure...it literally says an API error occurred on this official webpage. I doubt they are checking it. Pensioners need their stock to go up.
It's not that bad if the mathematical proofs now join the other things in a stratosphere of information that is out of reach for human comprehension. Mathematics was already on that course for a few decades now. What's bad is, even the simpler stuff that was fully comprehensible by common person are also being pushed into the stratosphere, making us dependent on the great creatures that live there.
Correct me if wrong, but didn't lean have a statement that forced the compiler to accept a statement as true with no need to prove it aimed at shortening obvious stuff? If so, are there checks for incorrect usages?
> These checks ran on an inert export of the proof, out of reach of hypothetical malicious code in the Lean proof sources.
I’d like to know more about this part.
Now next step is what can we actually learn from these. Any interesting new theorem developed as by-product?
amazing
[dead]
[dead]
Would be funny if lean has bugs (it's a program after all) and Claude tasked with a hard problem brute forced a vulnerability that allows proving wrong statements.
IMO the really interesting bits here are going to be the simplifications that will come from combining ideas from Alpoge and Masullo.
They only verify Lean proof itself, though it's the same formalization as Mathlib. Still, only human readable proof can actually prove it.
While still possible, the proof exploiting a bug is vanishingly small now. Following a few bugs and bad proofs this year, there are a number of different implementations, in different languages, and with different levels of proofs of their consistency. con-leche and con-ron are the heavy hitters, with consistency proofs of themselves. So for RH to be false, there'd have to be a bug in all the kernels that affects both the consistency proofs and the RH proof itself.