I don’t know shit about mathematics, except that when numbers and letters collude, that’s a fucking conspiracy against me. That said, I strongly suspect that for legit mathematicians, solving thorny math problems is only half of what they do, but rather, by operating from a platform of experience, intuit the nature of numerical systems, what that means in relation to other systems, and how we use them to understand and navigate the world.
The title sounds as ridiculous as proclaiming professors of literature are worried because an LLM can parse a sentence.
There’s a fundamental knowledge, both in the specifics of systems and holistically that can only be gained by solving these quandaries through the lens of human interpretation.
LLMs will not eclipse us.
However, I do believe highly intelligent AI is possible. It’s just not LLMs.
No they aren’t.
They should be worried! Just like how the calculator replaced them!
There is already no need for mathematicians, I can perfectly well make any square root on my calculator. 🤣
Keep in mind that quite a number of those AI based “breakthroughs” in mathematics turned out to be wrong.
And also, most of the claims around LLMs solving Erdos problems have eventually been found to just be referencing obscure existing solutions in papers people had forgotten about or that were poorly labeled
Interesting.
Got any sources for this claim?
LLM mathematical proof exploits theorem proover bugs [to get false statement to be “proven” true]
Oof.
Not at hand, but there were a number of news items about this in the last few weeks.
deleted by creator
You don’t really need sources, we all know AI tends to hallucinate when presented with a problem they can’t solve.
So far I’ve only read about 1 example of a problem that an AI may have solved. All the rest are without details and are not confirmed in any way.Remember AI companies are huge on propaganda for their technology, but not so big on admitting the shortcomings.
Edit:
Changed the use of the word evidence to sources, because apparently people misunderstood it.“you don’t really need evidence” we are talking about mathematics. The evidence is the point.
This is not about the mathematical evidence, but the lack of evidence that AI has made mathematical breakthroughs.
The point here is there is rarely any evidence for the claims about AI amazing accomplishments.
The burden of proof is on the one with the claim that AI has made these accomplishments, not on the one being sceptic about it.Articles with proofs have been published, both by AI companies as well as by mathematicians who disclosed the whole development of the proof was done autonomously by LLMs.
Which is way more possible to link to. But you chose not to?
Riiiight. Just like printed books did. What kind of world would it be if anyone could read about math.
Also mainframes, calculators, personal computer, spreadsheets, …
Absolutely this. You can also add supercomputers.
AI may help the field move further, but real mathematicians are still needed to evaluate the meaning of such findings.This one will do me, as someone whose profession requires practical adaptation of mathematics in real world environments.
Seeing what I can see with the development of AI, it looks to me that any application which requires intensive calculation to solve one or more direct problems would be its strength but there still would probably need the mathematician (in my case spatial scientist) set the boundaries of what needs to be solved. Pardon the pun.
How AI dealt with variables such as pressure, temperature and heat expansion at certain times of the day within a measurement or several very long baseline measurements would all be dependent on the ‘mathematician’ behind the calculations, for a real example I could give.
Suppose we had a library filled with proofs of every theorem [in mathematics], as well as excellent guides that could, given a question, take us to the answer and explain it. What would a mathematician do in such a library?
If you ask the question this way, the answer becomes clear: they would be unbelievably excited, and immediately get to work. They would immediately start asking questions: how does one prove the Riemann hypothesis? The Hodge conjecture? Their own pet obsession (in my case, the Grothendieck-Katz p-curvature conjecture)? Then they would work until they understood the answer. The job would not be done, not even close.
This paragraph by Daniel Litt feels to me like hubris coming from someone in the position of power. Many tech-optimistic mathematician are not at the risk of being replaced by AI because they are either on the tenure track or already tenured.
We need to acknowledge that mathematics is a subject of more human importance than economical importance. In this hyper profit driven world, field without a primary economical drive will necessarily shrink significantly.
Mathematics have no doubt experienced that: in the cold war, mathematics is behind most of the technological advancement that lead to concrete economical output. Later, as many of these fields stablized, engineer and computer scientists takes the place of mathematician, and mathematics shrunk significantly.
Now mathematics still holds importance because people believe it still encapsulates important ideas that have the potential to be the next generation of economical driver force. And mathematicians are important to preserve and disseminate such knowledge.
So if such truth oracle described above can develop and articulate any mathematical idea better than (or even close to the quality of) any mathematicians, it would be fun for established mathematician to flip through the answer sheets of their puzzle, but it would kill the financial driver and wipe out most of professional mathematics with it.
AI won’t replace mathematicians. But AI may replace their grad students, when the Computer Science department shows how much more “work” the Al can do and steals funding from the Mathematics department.
Then, when the current Math faculty retires, there won’t be any new Mathematicians to take their place.
All of the evidence presented in the article are either PR from LLM corpos or based on pre print review papers not yet peer reviewed. There probably will be some problems or class of problems these tools will help with, but I don’t see any more reason to think mathematicians will be less necessary than search engines replaced scientists or horseless carriages replaced wheel manufacturers.
LLMs are notoriously bad at doing even the simplest math problems
Your knowledge is out of date. For example, Claude runs a Linux container with Python and Sympy, so it absolutely can do a good chunk of math now. Still makes mistakes, but it’s good enough to serve as a LaTeX assistant. Similarly for ChatGPT but crappier. If you have the computing power, I think you can also give a local LLM access to Python tools with OpenWebUI.
Finding a proof is different than solving a math problem.
How so? Particularly in intuitionist logic it would seem to me that they are the same thing.
You’re right. The first is way harder and requires being able to do the second. Which it can’t.
Except this article is literally about the response to AI doing it multiple times.
https://arstechnica.com/ai/2026/06/openais-math-breakthrough-played-to-ais-strengths/
https://patmcguinness.substack.com/p/openais-astra-tackles-mathematical
The whole, “LLMs can’t do basic math,” thing is outdated.
They can’t even solve axioms, much less idioms
Absolutely no way.
Even if AI can make mathematical breakthroughs as in evidence of hard to solve problems. We still need mathematicians to clear it as valid.I imagine AI might actually sharpen the understanding among real mathematicians.
AI didn’t make Chess or Go obsolete either.We don’t need mathematicians to clear the proof as valid if it is checked by a formal proof system. Mathematicians would only need to check the theorem itself to make sure it describes what it should describe.
I think chess and go are a bad comparison. Their solving does not conclude in some societal use. They are interesting only as problems, but mathematics is interesting as a solution too.
Mathematics, and really any other subject, are not just about solving formalized problem. It is much more important to understand what question to ask.
One of my colleague once said the definitions in a good (computer science) paper should be the most interesting part, theorem statements should be the second interesting, and the proofs should be obvious.
Formal proof means nothing if it cannot give us insight in other proofs.
Same with open problems, Mathematician love open problems because given that no expert are able to solve them, their solution likely involves novel mathematical ideas. Fermat’s last theorem on its own is no where near as interesting as the mathematics that leads to its soluion.
Given that AI have yet to be able to wield the mathematical corpus effectively in solving large projects (or even fully autonomously improve large software), it would need human guidence, and by that, human experts are needed to understand the problem.
To qoute another one of my colleagues, people orchestrated AI to solve an open problem are simply the apple falling on Newton’s head. Apple “knows” about the existence of gravity, because its motion follows it, but it takes a Newton to formulate and explain gravity that leads to a number of technological advancement later. Without the question “why do apple fall”, apple will keep falling, we will keep noticing it, but we would never turn that observation into useful technologies we enjoy today.
Nobody wants to become solely a proofreader and bullshit detector for AI, in þe off-chance it stumbles on a genuine proof. Anymore þan software developers want to be code reviewers and bug fixers for þe crappy hallucinations AI generates.
They could become developers or designers Oooh wait…
Cant even become taxi driver due to autonomous cars or electrician because the field will be overrun…
Selfdriving cars: anytime soon now bro, just wait a little bit more, promise they’ll come, they’ll take over bro, bro
But I guess they will one day do that, people should know about hype-curves.
Turns out sometimes its just a guy in the Philippines.
