OpenAI: AI Solves 80-Year-Old Math Puzzle

A Revolution in Mathematical Research

May 20, 2026, will be etched in the annals of science as the day OpenAI announced that one of its artificial intelligence models had solved a discrete geometry problem that had eluded mathematicians for 80 years. This is not merely a computational feat, but a genuine conceptual breakthrough that could transform the way mathematical research is conducted.

The problem in question is the unit distance conjecture, posed by Hungarian mathematician Paul Erdős in 1946. This conjecture asks how many pairs of points can be placed on a plane such that they are separated by exactly the same distance, for example, one unit. Erdős proposed that the maximum number of such pairs could not exceed a nearly linear bound, denoted as n^{1 + o(1)}.

An Innovative Approach

For decades, mathematicians believed that to maximize these pairs, points needed to be organized into regular structures such as square grids or triangular lattices. This geometric intuition seemed insurmountable, and researchers could not conceive that another configuration might exist.

However, OpenAI's model took a radically different path. Instead of exhaustively testing billions of geometric configurations, the AI transposed the problem into the abstract realm of algebraic number theory. By mobilizing complex structures such as CM fields and Golod-Shafarevich type class field towers, the AI discovered a new family of point configurations that surpass traditional lattices.

The bound established by this discovery is of the form n^{1 + δ}, where δ is a strictly positive universal constant. This advancement formally contradicts Erdős's conjecture. The minimal configuration illustrating this discovery requires a number of points on the order of 10^{1957}, a figure so colossal that it is inconceivable in our physical universe. It is precisely this abstraction that has prevented human minds from considering it.

Rigorous Validation

In a world where technological announcements are often met with skepticism, this discovery underwent rigorous verification. The proof produced by the AI was validated by Lean, a formal proof assistant that checks each logical step without room for interpretation. This validation eliminates the risk of error often associated with AI systems.

Moreover, a committee of renowned mathematicians, including Noga Alon, Timothy Gowers, Will Sawin, and Jacob Tsimerman, examined the entire reasoning. Their conclusion, published in an analysis paper, is unequivocal: the proof is rigorous, and the conceptual leap made by the machine is as unexpected as it is elegant.

Beyond Mathematics

This discovery is not an isolated achievement. Since the beginning of 2026, fifteen Erdős problems that had stagnated for generations have been solved, eleven of which are directly attributed to the new models of artificial reasoning. The implications of this advancement far exceed the realm of pure mathematics.

By demonstrating its ability to maintain long and abstract chains of reasoning without losing track, this technology opens concrete prospects in other fields. In quantum physics, it could enable the modeling of states of matter previously out of reach. In molecular biology, it could revolutionize the prediction of highly complex protein folding. In cybersecurity, it could design encryption systems whose robustness can be formally verified.

A New Paradigm for Research

The question is no longer whether a machine can "think," but rather to recognize that an artificial system can explore conceptual spaces inaccessible to human intuition. This capability is not due to superior intelligence, but to the absence of perceptual biases that often limit human researchers.

For mathematicians, this advancement does not mark the end of their discipline, but rather the beginning of a new era. An era where certain problems are solved not at the blackboard, but in collaboration with a machine capable of seeing what the human eye cannot perceive.