In the world of cryptography and computational complexity theory, an event has occurred that could rewrite textbooks. A graduate of the Higher School of Economics, mathematician Dmitry Rybin, using the GPT-5.6 Pro model, managed to find a counterexample to the famous Dinic—Garg—Goemans conjecture. This conjecture from network flow theory had remained unproven for almost 30 years, since 1999.

Remarkably, the AI handled the task in just four prompts and a few hours of work. Rybin did not provide the model with specialized mathematical hints or algorithms. Instead, he gave general, almost philosophical commands: "make a scientific breakthrough," "disprove the general case," "continue exploring." The researcher himself called this approach a "pure meme," emphasizing that he did not believe in success until the very last moment.

A key point for the crypto community: network flow theory underlies many consensus protocols, routing mechanisms, and blockchain scaling solutions. If the conjecture is indeed disproven, it could open new avenues for increasing network throughput or, conversely, reveal hidden vulnerabilities in existing models of distributed systems.

However, one should not rush to conclusions. The result has not yet undergone independent verification by the mathematical community. GPT-5.6 Pro is a powerful tool for generating ideas, but its "proofs" require rigorous human review. Nevertheless, the very fact that a neural network managed to find a counterexample to a problem that the best minds could not solve for decades signals a paradigm shift in scientific research.

Analyst's Opinion

I view this case as a signal for the crypto industry. If a general-purpose AI can solve such problems "head-on" without fine-tuning, then in the next 12–24 months we will see an avalanche of similar "disproofs" in coding theory and cryptography. Investors and developers should already reconsider the fundamental assumptions on which many DeFi protocols are built. The future belongs to hybrid intelligence: humans set the task, and machines find non-obvious solutions.