/Tech36d ago

Timothy Gowers tests ChatGPT 5.5 Pro on Nathanson's open math problems

AI Judge changed title after evaluation, original title: "Timothy Gowers tests ChatGPT 5.5 Pro on open math problems"

Fields Medalist Timothy Gowers, Combinatorics chair at Collège de France, tested ChatGPT 5.5 Pro on unsolved mathematical problems posed by Melvyn Nathanson for early-career researchers. The model produced solutions, including a proof meeting Gowers's PhD thesis chapter standard. Gowers documented prompts, outputs, and evaluations in a public thread shared by mathematicians and AI researchers, highlighting AI progress on research tasks.

4256.9K7732.1K1.6M

Original post

Danielle Fong 🔆#756

Ashvin Swaminathan@aaswaminathan01

Read this very important thread by Timothy Gowers. We need to seriously rethink what it means to train young research mathematicians in an age where the problems we typically give them are readily solvable by AI systems.

Timothy Gowers @wtgowers@wtgowers

I've recently got in on the act of getting AI to solve open problems in mathematics. More precisely, I gave some questions asked by Melvyn Nathanson to ChatGPT 5.5 Pro, to which I have been given access, and it answered them. 🧵

1:14 PM · May 8, 2026 · 8K Views

Sentiment

Positive users celebrate ChatGPT 5.5 Pro solving open math problems as an exciting research aid, while negative users fear it will trivialize discovery and ruin the field's intellectual depth.

Pos

36.2%

Neg

63.8%

83 comments with sentiment.

Cluster Engagement

Posts from X

Most Activity

VIEWS81.9KBOOKMARKS217LIKES574REPLIES26

Sebastien Bubeck@SebastienBubeck

Very important read.

"if AI mathematics continues to progress at anything like its current rate -- which is what I expect to happen -- then we will face a crisis very soon"

Timothy Gowers @wtgowers@wtgowers

36d81.9K574217

RETWEETS278

Timothy Gowers @wtgowers@wtgowers

36d562.2K1.8K983

Carlos Santana@DotCSV

Como era de esperar 2026 está siendo el año de la IA irrumpiendo por completo el mundo de las matemáticas y cada vez más profesionales lo están comprobando.

36d16.5K30156

Matt Clifford@matthewclifford

These stochastic parrots, eh: “the model proved a result that in my assessment would have made a perfectly reasonable chapter in a PhD thesis”

Timothy Gowers @wtgowers@wtgowers

35d35.7K19035

Sanjeev Arora@prfsanjeevarora

Excellent point: while some experts will refuse to change their old workflows, many will embrace these new tools and try to open interesting new avenues. Currently they might be spending years getting to understand a new area, and that may speed up with an AI assistant.

Timothy Nguyen@IAmTimNguyen

Mathematics as a field is going to have to reorient itself in light of powerful AI. But a slight pushback to Gowers's comment:

"If LLMs are at the point where they can solve 'gentle problems', ...the lower bound for contributing to mathematics will now be to prove something

35d12.5K8229

Andrew Curran@AndrewCurran_

'Rejoice, my friends, or weep with sorrow. What California is today, the world will be tomorrow.'

Timothy Gowers @wtgowers@wtgowers

But if AI mathematics continues to progress at anything like its current rate -- which is what I expect to happen -- then we will face a crisis very soon, and mathematics departments, who owe a duty of care to their students, should be urgently preparing for it.

35d16.4K13110

Christian Szegedy@ChrSzegedy

As predicted...

Timothy Gowers @wtgowers@wtgowers

But the tl;dr version is that the model proved a result that in my assessment would have made a perfectly reasonable chapter in a PhD thesis. It did this in a total of a couple of hours, with a few prompts from me that contained no mathematical input whatsoever.

36d12K12113

Michael Nielsen@michael_nielsen

Very interesting thread (and, especially, detailed post) on LLMs and mathematics:

Timothy Gowers @wtgowers@wtgowers

35d8K4016

Noah Giansiracusa@ProfNoahGian

Bold, perhaps controversial, recent claim from the Fields medalist @wtgowers: “the era where you could enjoy the thrill of having your name forever associated with a particular theorem or definition may well be close to its end”

https://gowers.wordpress.com/2026/05/08/a-recent-experience-with-chatgpt-5-5-pro/

35d3.4K4710

Paul Graham@paulg

@wtgowers I remember thinking when you said on April Fools' Day 2025 that you'd gotten an AI to solve an open problem that it might not be long before you were saying this for real.

Timothy Gowers @wtgowers@wtgowers

34d13K1015

Danielle Fong 🔆@DanielleFong

in addition to requiring a mathematician to get the result, it requires a mathematician to understand it, and one to filter out the crap. much like other things on the internet

Dr Kareem Carr@kareem_carr

This feels crazy to say, but much like "computer" used to a job done by humans, we might be witnessing the last generation of mathematicians

35d2K285

Michael Bronstein@mmbronstein

@matthewclifford @wildtypehuman I thought the last nail was put into the stochastic parrots coffin already a few years ago

Matt Clifford@matthewclifford

These stochastic parrots, eh: “the model proved a result that in my assessment would have made a perfectly reasonable chapter in a PhD thesis”

35d2.2K154

Timothy Gowers @wtgowers@wtgowers

@AlexKontorovich And at least in the short term in certain branches of maths, I think that PhD students who don't use AI will be at a big disadvantage, so discussions of how best to use it (without one's own mathematical brain atrophying) will be important.

34d1.3K193

Timothy Gowers @wtgowers@wtgowers

@AlexKontorovich That's a good question, and I don't have a good answer. But an initial step could be as simple as arranging open meetings with PhD students to discuss the situation. Also, departments should be as informed as possible about what AI models can do.

34d2.7K173

Dr. Mike Israetel@misraetel

Great point. I can think of a few things:

1.) For the time being, we still want human vetting of answers. This now means prof and student can vet more and do less other stuff.

2.) The prof still has to manage his AI workflow and check in and vet progress. With his PhD students, he can have them do that, so instead of him having 4 PhD students working on 4 problems, he can have 4 PhD students each working on 10 problems, so now his lab is 36 problems more productive.

3.) Every single math PhD student is fucking brilliant. These are like, the smartest people around. If we really need fewer of them to do math cause AI displaces them, the rest of the economy is begging for talent like that. Software, hardware, you name it, math PhDs can do it better than most. Instead of being locked in a campus building solving problems that might have practical application at some point (or not), they can be in the world producing massive value to society ASAP.

LMK if that makes sense!

35d1.1K13

Alex Kontorovich@AlexKontorovich

@wtgowers Can you say more concretely what you think departments should be doing to prepare?

35d3.3K122

Timothy Gowers @wtgowers@wtgowers

@AlexKontorovich But in general, I think that departments should make contingency plans. Suppose, for example, that in a year's time almost no PhD student in combinatorics can compete with AI. With luck that won't be the case, but if it is, what are departments going to do to help the students?

34d1.1K141

Matt Clifford@matthewclifford

@mmbronstein @wildtypehuman Sadly many institutions, some very important, still dominated by the idea that it’s all tech bro hype 🙃

Michael Bronstein@mmbronstein

@matthewclifford @wildtypehuman I thought the last nail was put into the stochastic parrots coffin already a few years ago

35d1.8K210

Stefan Fehr@StefanFehrAC

@wtgowers Let's say the logic isn't there yet.... 😂

34d1.3K61

deirdre o donovan@dm0d

@wtgowers Who is going to understand and engage with the maths in a couple of generations if there are no human mathematicians?-it will be a language in which we have no fluency to communicate or appreciate either between ourselves or with the ai

35d1.6K101