AI cracks – A newly built AI model, developed by OpenAI, has solved the planar unit distance problem—an 80-year-old question linked to Paul Erdős—producing a counterexample that overturns his conjecture about the best way to place points. Mathematicians describe it as a “
On an infinite sheet of paper, the question sounds like something you could teach in a minute—then spend a lifetime chasing. How many equal-sized lines can you draw that connect points, if you’re free to place the dots however you want?
That puzzle is the planar unit distance problem. and it became Paul Erdős’s “most striking contribution to geometry” because it sits right at the edge of the simple and the impossibly deep. Erdős’s central conjecture. built around patterns of points arranged in a grid. implied the maximum number of connections would rise only slightly more than the number of points themselves. For decades, mathematicians chipped away at it. The most recent improvement to Erdős’s conjecture came more than 40 years ago—until now.
An artificial intelligence model built by OpenAI has produced what experts are calling a decisive breakthrough. stunning enough that some researchers say they did not expect to live to see it solved. “This is a problem that I didn’t expect to see solved in my lifetime. ” says Misha Rudnev at the University of Bristol. UK. “It’s absolutely a bomb.”.
Tim Gowers at the University of Cambridge. writing in a blog post accompanying the work. called the result “a milestone in AI mathematics.” He added that if a human had written the paper and submitted it to the Annals of Mathematics—then asked for a quick opinion—he would have recommended acceptance “without any hesitation.” He also said no previous AI-generated proof had come close.
What makes the achievement so jarring is the way it breaks the story Erdős set in motion. The AI found a counterexample showing Erdős was “significantly wrong.” Instead of the highly symmetric point patterns that had driven so much of the effort. it constructed arrangements “in less symmetric patterns” that can yield a far greater number of connected pairs.
Will Sawin at Princeton University described his own reaction in a way that captures the tension of the moment. “My immediate reaction was disbelief,” he says. “I thought the way that it was trying to solve it wouldn’t work. but then I looked at it more and I convinced myself that it does work. I pretty quickly became convinced this is the most significant achievement by AI in mathematics so far.”.
OpenAI has not explained exactly how the model differs from publicly available AIs or how it was trained. But the company’s researchers have publicly commented that the model is “general purpose” and wasn’t trained “with the goal of doing math research.”
Even so, the proof’s engine appears to come from a very specific mathematical source. The AI borrowed a technique from algebraic number theory to construct vast lattices in much higher dimensions than the two-dimensional world of a plane. Then. once these higher-dimensional shapes were built and identified. it collapsed them down to two dimensions—producing what one researcher described as a kind of shadow of the higher-dimensional structure.
Kevin Buzzard at Imperial College London said the counterexample is complex. “Although the ideas to produce it were already in the literature, it certainly takes some ingenuity to put them together,” he said.
Still, not everyone sees the result as a simple demonstration of brute computational power. Samuel Mansfield at the University of Manchester. UK. points to a quieter limitation inside the mathematics community itself: mathematicians may not have even considered that Erdős’s original conjecture could be false. “Even if mathematicians did experiment with disproving it. very few geometry specialists would have then been knowledgeable enough in advanced number theory to do so. ” he said. “This is something that requires you to know a lot about multiple areas.” In his view. the shock fades when you recognize what an AI would be good at doing: combining tools from different parts of mathematics.
The planar unit distance problem has long appealed to mathematicians for the “pure intellectual challenge. ” Rudnev says. and the new result may not immediately unlock direct applications for other outstanding problems. But it has already set something in motion—because once the proof is understood, techniques travel.
After seeing the work. Sawin used the technique the AI discovered to produce a slightly improved. higher number for how many points could be joined together. Buzzard said that matches a broader pattern. “Like many other AI breakthroughs. it did not take humans long at all to internalise. understand and generalise the arguments. ” he said. He contrasted that with some human breakthroughs, where validation can take “months or years” for the wider community.
The planar unit distance problem is still about dots on an infinite sheet of paper. But the feeling around it has changed. Erdős’s grid-centered expectation has been overturned. and the proof—born from a general-purpose AI model—has moved from skepticism to acceptance with a speed that surprised even the people now celebrating it.
AI mathematics planar unit distance problem Paul Erdős OpenAI algebraic number theory lattices geometry proof breakthrough