An event that could overturn perceptions of artificial intelligence's capabilities in fundamental mathematics: HSE graduate and mathematician Dmitry Rybin presented a counterexample to the Dinits—Garg—Goemans conjecture. This conjecture from network flow theory had remained unproven for nearly 30 years, since 1999. And it was disproven not by a human, but by the GPT-5.6 Pro model, in just four prompts and a few hours of work.
A pure "meme" with a breakthrough result
Notably, Rybin gave the neural network no narrow instructions or hints. Instead, he used general, almost philosophical formulations: "make a breakthrough," "disprove the general case," "continue exploring." The mathematician himself called this process a "pure meme"—so unexpected and unconventional was the approach. The model independently constructed a logical chain that led to a formal counterexample.
Technical details and verification status
At present, the result has not undergone independent expert review, and this is a key point. In the scientific community, it is accepted that any counterexample to a long-standing conjecture requires thorough verification. However, the very fact that AI was able not just to generate a plausible argument, but to propose a specific refutation for a complex mathematical structure, indicates a qualitative leap in the capabilities of large language models.
Analyst's perspective
As a professional analyst of crypto and technology markets, I see in this event not just a scientific curiosity. It is a signal: if GPT-5.6 is capable of solving problems that have stumped mathematicians for decades, then its potential in analyzing complex network structures—from blockchain protocols to decentralized financial flows—could be enormous. We stand on the threshold of an era where AI becomes not a tool, but an independent researcher capable of making discoveries. If independent verification confirms the result, it will change the rules of the game not only in mathematics, but also in related high-tech disciplines.