An event has occurred in the world of cryptography and network flow theory that could revolutionize our understanding of artificial intelligence's capabilities. Dmitry Rybin, a graduate of the Higher School of Economics (HSE) and a professional mathematician, has managed to disprove the Dinic — Garg — Goemans conjecture, which had remained unproven since 1999. And he did this not through classical mathematical methods, but through interaction with the GPT-5.6 Pro neural network.
The most remarkable aspect is that the model received no direct prompts or algorithmic guidance. Rybin simply gave it general commands like "make a breakthrough" or "disprove the general case." And GPT-5.6 Pro completed the task in four prompts and several hours of work. The specialist admitted that he perceives this chat as a "pure meme" — so unexpected and impressive was the result.
The Dinic — Garg — Goemans conjecture relates to fundamental problems in network flow theory, which underlies many cryptographic protocols and distributed ledger systems. Its refutation could have far-reaching consequences for data flow optimization algorithms, including those used in blockchain networks and DeFi protocols.
Pending Independent Verification
At present, the result has not undergone peer review or independent verification by the scientific community. However, the very fact that a neural network without specialized training was able to find a counterexample to a complex mathematical hypothesis indicates a qualitative leap in AI capabilities.
My expertise: If this result is confirmed, we will witness a moment when AI begins not just to process data, but to generate new mathematical truths. For the crypto industry, this means that in the coming years we may see automated proof of theorems underlying blockchain security and, possibly, the discovery of previously unknown vulnerabilities in consensus algorithms. GPT-5.6 Pro has shown that a "pure meme" can become a scientific sensation.