This morning, I woke up to news of a breakthrough on Erdős’ unit distance problem. Disproving Erdős’ Unit Distance Conjecture would be remarkable in its own right. What makes this moment extraordinary, however, is that the breakthrough was achieved by an OpenAI model. You can watch this fairly recent video if you are interested more in the history of the problem:

This is the kind of event that forces one to pause for a moment.

The unit distance problem is the kind of problem that we like to present to students as evidence that simplicity of statement has almost nothing to do with simplicity of solution. It is a classical problem in combinatorial geometry, posed by Erdős in 1946, and it has resisted generations of very strong mathematicians. I first came across this question in Tom Bohman’s probabilistic combinatorics class at CMU. The question is deceptively simple: how many pairs of points among points in the plane can be exactly one unit apart? Erdős constructed examples showing that one can achieve slightly more than linear many such pairs, and conjectured that his construction was essentially optimal, up to lower-order factors. This conjecture has been disproved by OpenAI. For decades, the problem has occupied a strange place in mathematics: elementary to state, resistant to attack, and connected to deep ideas in geometry, number theory, graph theory, and incidence combinatorics. It is exactly

There are many things I plan to check since the details matter enormously. Was the model merely searching over a well-designed space of constructions? Did it propose a configuration that humans then recognized and proved correct? Did it generate a proof independently? How much scaffolding was provided? How much human judgment was required to formulate the search, filter the candidates, and certify the result? The proof and the chain of thought are available and can be found via links in this post by OpenAI.

Nonetheless, this breakthrough feels like the closing of a long chapter in human history, one that began with the discovery of mathematics itself. This was not a marginal problem, nor merely one of those older Erdős questions that may have become easier over time because of progress in adjacent areas, or because they no longer attract sufficient attention. The unit distance problem was a major and central challenge. That is why this moment feels different. It is compelling evidence that a new era of mathematical discovery has begun.

The consequences are far reaching in so many ways. Here are few that come to mind from an academic perspective: hiring and teaching.

Academic hiring has always used papers as imperfect proxies for something deeper: taste, originality, technical strength, independence, and the ability to open new directions. If AI systems can produce conjectures, constructions, or even proofs, then the meaning of a publication begins to change.This will force departments to rethink what they are actually selecting for. In the past, the central question was often: “Can this person prove difficult theorems?” In the near future, the better question may be: “Can this person identify the right mathematical frontier, use all available tools intelligently, and turn raw discovery into understanding?” The premium may shift from isolated technical virtuosity toward taste, synthesis, problem formulation, and conceptual explanation.

This does not mean that students no longer need to learn proofs, algebra, geometry, or algorithms. Quite the opposite. A student without mathematical foundations will be unable to judge whether an AI-generated argument is meaningful or nonsense. The danger is not that students will know too much because of AI; the danger is that they will accept outputs without understanding them.

Perhaps the most important educational lesson is humility. If an AI system can make progress on a problem that resisted human attention for decades, then we should be less confident about the boundaries we draw between routine and creative work. But humility should not mean despair. It should mean that we prepare students for a world in which mathematical intelligence is no longer located only inside individual human minds, but also in the interaction between humans, machines, and the problems themselves. Someone once supposedly wished, “May you live in interesting times.” We are no longer merely discussing whether AI will assist mathematics, education, and research. We are watching it begin to reshape the standards by which human intellectual work itself is judged. These are interesting times indeed—perhaps too interesting, but undeniably ours.