An event has occurred in the world of cryptography and theoretical computer science that could fundamentally change the approach to solving complex mathematical problems. A graduate of the Higher School of Economics, mathematician Dmitry Rybin, using GPT-5.6 Pro, managed to find a counterexample to the Dinits — Garg — Goemans conjecture, formulated back in 1999 and remaining unproven for almost three decades. This conjecture from network flow theory was of fundamental importance for optimizing algorithms, including those used in blockchain networks and decentralized protocols.

How AI Made a Breakthrough

The most remarkable thing about this case is the complete absence of human guidance. Rybin stated that he did not give the model any technical hints or clues. Instead, he used general, almost absurd prompts: "make a breakthrough," "refute the general case," and "keep exploring." GPT-5.6 Pro completed the task in just a few hours, requiring only four iterations. The researcher himself called this chat a "pure meme," emphasizing the unexpectedness and, possibly, randomness of the success.

Implications for the Crypto Industry

Although the result has not yet undergone independent peer review, its potential consequences are enormous. Network flow theory underlies many mechanisms used in cryptocurrencies — from transaction routing in the Lightning Network to optimizing smart contract operations. If the conjecture is indeed disproven, it could lead to a revision of consensus algorithms and an increase in the efficiency of decentralized systems.

Expert Opinion

From my professional perspective, this case demonstrates that AI is already capable not just of analyzing data, but also of making intellectual discoveries that were previously considered the prerogative of humans. For the crypto community, this is a signal: in the coming years, we may see algorithms independently optimizing blockchain protocols, finding vulnerabilities, and even generating new mathematical proofs. However, full trust in such results is still a long way off — independent verification is mandatory, otherwise we risk implementing erroneous conclusions into critically important systems.