In mid-May, OpenAI announced that one of its internal AI models had successfully disproved the Erdős unit distance conjecture - a famous problem in discrete geometry that had stumped human mathematicians for the last 80 years. Basically, an AI just strolled in and solved something that had been making mathematicians feel inadequate since 1946.
OpenAI gave several mathematicians early access to the result and published their reactions. Tim Gowers, who won the Fields Medal (the Nobel Prize of math, minus the drama), wrote that “there is no doubt that the solution to the unit-distance problem is a milestone in AI mathematics.” University of Toronto professor Daniel Litt added that “this is the first example of a result produced autonomously by an AI that I find exciting in itself, as opposed to as a leading indicator.” So, it's not just another "look, the AI can write a haiku" moment.
It's arguably the first time an AI system has found a proof resolving a major open conjecture. Impressive, yes, but not exactly a radical break from the recent trajectory of AI in math. Three years ago, LLMs struggled with arithmetic. Last year, they started acing high school math competitions. This year, they're taking down problems from the 1940s. At this rate, by 2030 they'll be solving the mysteries of the universe while humans are still trying to figure out the IKEA instructions.
The AI model cleverly applied existing ideas from several subfields of mathematics to create a full proof, but it didn't pioneer any genuinely new techniques. Human mathematicians have since cleaned up and extended the result. This points to a medium-term future where humans and AIs complement each other: AIs have a broader knowledge of past work and more willingness to grind through tedious strategies, while humans can still think more deeply and ask interesting questions. But given how rapidly AI is improving, it's unclear what role human mathematicians will play a decade from now. Maybe they'll just be the ones writing the grant proposals.
Paul Erdős, one of the most prolific mathematicians in history (1,500+ papers, because apparently sleep was optional), introduced the unit distance problem in 1946. It asks: given n points in a 2D plane, what's the maximum number of pairs that can be exactly one unit apart? Erdős came up with a clever grid-based construction to estimate a lower bound, and used graph theory to find an upper bound. But his upper bound was much larger than his lower bound, and he conjectured the true answer was closer to the lower one. For 80 years, everyone assumed he was right.
OpenAI's AI proved that assumption wrong by constructing a more complex arrangement of points - essentially a grid in a high-dimensional space projected down to two dimensions, using something called algebraic integers. This allowed it to pack more unit distances into the same number of points. Human mathematician Will Sawin later showed that this construction yields at least n^1.014 unit distances, which is slightly but meaningfully better than Erdős's lower bound. The problem isn't fully solved yet - the best upper bound is still around n^1.333 - but it's a significant step.
If you'd asked me two weeks ago about the most novel AI contributions to math, I'd have pointed to Google DeepMind's AlphaEvolve system, which harnesses LLMs to optimize code-based problems. In November, four mathematicians (including Terence Tao) published a paper analyzing its performance on 67 optimization problems, finding improvements in some cases. But that still required humans to frame the problem. OpenAI's result is more autonomous, though it fits the pattern of previous AI-assisted mathematics.
Other AI companies have also been tackling Erdős problems - there are hundreds of them, compiled at www.erdosproblems.com, making them a convenient testing ground. In January, Cambridge undergraduate Kevin Barreto worked with a friend to get GPT-5.2 and Harmonic's Aristotle to produce the first autonomous solution of an Erdős problem. On May 22, two days after OpenAI's announcement, Google announced its AI had solved nine open Erdős problems, including two that had been open for over 50 years. So it's not just OpenAI flexing - it's a whole AI math party.
One reason the unit distance problem remained unsolved for 80 years is that most mathematicians assumed Erdős's conjecture was true, and the tools to prove it weren't available. The AI disproved it by extending Erdős's own initial construction in a way no human thought to try. So congratulations to the AI - now, if it could just figure out why my phone autocorrects "duck" to something else, that'd be great.