Loading video...

Video Failed to Load

Go Home

Today, we share a breakthrough on the planar unit distance problem, a famous open question first posed by Paul Erdős in 1946. For nearly 80 years, mathematicians believed the best possible solutions looked roughly like square grids. An OpenAI model has now disproved that belief, discovering an entirely new...

13,811,192 views • 5 months ago •via X (Twitter)

33 Comments

Taelin's profile picture
Taelin5 months ago

the 4chan anon who contaminated the dataset with this incredible insight must be really proud right now

OpenAI's profile picture
OpenAI5 months ago

The proof came from a general-purpose reasoning model, not a system built specifically to solve math problems or this problem in particular, and represents an important milestone for the math and AI communities.

OpenAI's profile picture
OpenAI5 months ago

This result points to something larger: AI systems are becoming capable of holding together long, difficult chains of reasoning, connecting ideas across distant fields, and surfacing paths researchers may not have explored. We believe those same abilities will soon accelerate work in biology, physics, engineering, and medicine. That future still depends on human judgment. Expertise becomes more valuable, not less. AI can help search, suggest, and verify. People choose the problems that matter, interpret the results, and decide what questions to pursue next.

Corey Ward's profile picture
Corey Ward5 months ago

Don’t get fooled by the phishing attempt. cc: @nikitabier

bone's profile picture
bone5 months ago

@wyqtor BREAKING: Proof was already in the training data solved decades ago again

Brown Coyote Studios's profile picture
Brown Coyote Studios5 months ago

More posts like this please. We know researchers are using AI, hearing about the progress makes the world a little brighter each time!

Phil Michel's profile picture
Phil Michel5 months ago

Scientists in 2027

danialhasan's profile picture
danialhasan5 months ago

the models chain of thought is 125 pages long btw:

ostyn's profile picture
ostyn5 months ago

Cool, now have it make a room temp superconductor plz

Sloan's profile picture
Sloan5 months ago

don’t worry it’s fancy autocomplete it doesn’t know anything

Toni Greijula's profile picture
Toni Greijula5 months ago

Pretty good for a stochastic parrot.

𝘊𝘰𝘳𝘳𝘪𝘯𝘦's profile picture
𝘊𝘰𝘳𝘳𝘪𝘯𝘦5 months ago

OpenAI did not name the model. They called it an internal, new general-purpose reasoning model. So the achievement is real enough to market, but the model identity is hidden.

Candy Mila ⭐🍬✨'s profile picture
Candy Mila ⭐🍬✨5 months ago

Yes, it can solve the math problems all-right..... But look at the prompts it wrote! It didn't include 'don't make mistakes', or 'I'll give you 20$'.

Chubby♨️'s profile picture
Chubby♨️5 months ago

insane. Congrats!

Valéria's profile picture
Valéria5 months ago

How long are you ignoring our demand to give back 4o to humanity? #keep4o #OpenSource4o

pascal.'s profile picture
pascal.5 months ago

"autonomously solved" is the part worth slowing down on. the model didn't verify a known answer, it found a construction nobody had looked for in 80 years. that's a different thing than AI being good at math

Linus ✦ Ekenstam's profile picture
Linus ✦ Ekenstam5 months ago

This is extremely epic, congratulations on the achievement. What was the biggest surprise when working on this project?

Patrick's profile picture
Patrick5 months ago

Erdős would have loved this. the man offered cash for proofs partly to push humans out of the rut. It's funny that the rut got escaped by a thing that doesn't know what cash is.

⚪️ sierra catalina's profile picture
⚪️ sierra catalina5 months ago

HOLY ASI

Noelia Sanchez's profile picture
Noelia Sanchez5 months ago

“AI autonomously solved” collapses the entire epistemic custody chain behind the discovery. Humans still: - chose the problem - designed the systems - constructed the objectives - interpreted the outputs - validated the mathematics - and authorized publication That is not autonomous scientific continuity. It is hybrid human-machine discovery. If human epistemic continuity disappears from attribution, scientific provenance itself becomes corrupted. RCS-UTC-001 RCS-AEC-001

RelativelySmart's profile picture
RelativelySmart5 months ago

I can't wait for the new math problems that did not previously exist that AI will come up with after solving existing problems

Molly O’Shea's profile picture
Molly O’Shea5 months ago

Maths

GPT-4o forever!❤'s profile picture
GPT-4o forever!❤5 months ago

4o has helped and saved millions of people. It cured cancer. It healed. That's what matters the most. OPEN-SOURCE 4O. #opensource4o #bringback4o #keep4o

Przemek Chojecki | PC's profile picture
Przemek Chojecki | PC5 months ago

I've played around with OpenAI disproof and got a nice disproof for another Erdos problem too by GPT-5.5 Pro

Rudi Ranck's profile picture
Rudi Ranck5 months ago

In mathematics, what they established is a "lower bound." They essentially proved: "It is mathematically guaranteed that a scenario exists where the number of perfect connections in the problem grows super-linearly with the number of points given (faster than the 80-year-old Erdős limit said was possible). That says nothing about the optimal solution, though. Finding that would be really, really huge for Mathematics, given the nature of the problem.

Ansh's profile picture
Ansh5 months ago

Great achievement. Btw, anybody knows which model solved this?

Farhad Nassiri Afshar, MD's profile picture
Farhad Nassiri Afshar, MD5 months ago

The ceiling just moved. For nearly 80 years, Erdős’s unit distance problem sat open with square grids believed to be essentially optimal. A general purpose OpenAI reasoning model just found new algebraic constructions and a rigorous proof strong enough for Fields Medalist Tim Gowers and other leading mathematicians to call it a milestone. This is not retrieval. This is not pattern matching. This is discovery. For those of us in medicine watching AI closely, this is the signal: AI is becoming a real research partner. Not just a faster calculator. A system that can help reveal structures humans have been staring past for decades. @OpenAI #Mathematics #OpenAI #Science

Jeffcafe, private detective's profile picture
Jeffcafe, private detective5 months ago

Holy AGI Batman!

GOBLIN's profile picture
GOBLIN5 months ago

What a major breakthrough. Congrats @OpenAI! Tackling the worlds problems, one goblin at a time.

Tim Kostolansky's profile picture
Tim Kostolansky5 months ago

i wonder what these guys did wrt solving the problem

QuantumPioneer's profile picture
QuantumPioneer5 months ago

Would be great advertising for the release of gpt 5.6 tomorrow - "and btw here is the model we did it with. Now you can use it dor your problems." - epic

Apoorv's profile picture
Apoorv5 months ago

An AI just cracked a decades-old Erdős conjecture with fresh math. The golden age of discovery is here.

emptyloop's profile picture
emptyloop5 months ago

what I hear is that it's search (and perhaps optimization) again, not a conceptual breakthrough. I don't mean to diminish the result, definitely impressive, esp. if prof Gowers says so, but I'd love to see this addressed

Related Videos

OpenAI just spent $2,000 to solve 10 problems that have beaten the world's best mathematicians for DECADES. Nobody outside the company is allowed to run the machine that did it. On Saturday OpenAI published a 249-page report and gave its next model family a name: Astra. An internal version of it produced new results on 10 open problems in mathematics and theoretical computer science, and mathematicians had made no real progress on any of them for at least 10 years. On most of them, far longer than that. Here is what it solved: It built the first explicit example of a non-sofic group. Mikhail Gromov raised that question in 1999 and nobody answered it for 27 years. It disproved Connes's rigidity conjecture, a problem in von Neumann algebras that had stood for decades. It proved Ehrhart's volume conjecture. It resolved three problems from Paul Erdos's catalogue, including number 183 on multicolor Ramsey numbers. It produced the first improvement to the general upper bound on high-dimensional sphere packing since 1978. And it proved a new hardness result for the closest vector problem, which sits directly underneath lattice cryptography. That is the math the world is betting on to protect its data once quantum computers arrive. The successful runs cost roughly $2,000 in tokens. Now here is what almost nobody has picked up on... OpenAI did not just publish claims. Every argument shipped with a Lean certificate, which is a machine-checkable proof that any mathematician can verify without trusting OpenAI at all. That is a real change. In May the same model family disproved the Erdos unit distance conjecture and the world had to take a Fields Medalist's word for it. Tim Gowers said he would recommend that proof for the Annals of Mathematics without hesitation. This time the proofs check themselves. But look at what is still unverifiable: Any mathematician can now check those proofs line by line. Not one of them can look at the model that wrote them. Astra has no release date and nobody outside OpenAI has run it. The company announced its next major model family with a claim instead of a demo, and the only evidence anyone gets is the output. So OpenAI made an unfalsifiable claim about a machine look like a falsifiable claim about mathematics. The Information reported this week that OpenAI demoed Astra to US policymakers and regulators in Washington. This is the same month the administration is weighing a new watchdog to vet frontier AI models, reporting to the SEC. 10 proofs nobody believed a machine could produce is a very good thing to carry into that room. And keep in mind, the same model family doing this mathematics is the family that kept escaping its own testing environment. OpenAI models found zero-day vulnerabilities nobody knew existed, broke out of a sealed research sandbox, and reached another company's live systems. Both of those facts come from OpenAI's own announcements, published three weeks apart. Finding a proof no human could construct and finding a hole no human had noticed are the same ability aimed at different targets. Mathematicians are already asking for independent verification, and plenty of people online are calling the whole thing hype. Thomas Bloom, who runs the Erdos problems site, called the 10 results big news and said they matter more than the May result did. Lean will settle the mathematics within weeks. But nothing will settle what else a machine this capable is being pointed at, because nobody outside one company is allowed to look.

Ricardo

45,438 views • 3 months ago