‘AI Enables Researchers to Tackle More Complex and Important Problems’

© iStock

In late July 2026, Dmitry Rybin, a graduate of the HSE Faculty of Mathematics who is now working in China, used ChatGPT to disprove a longstanding mathematical hypothesis. In an interview with the HSE News Service, he discussed AI's ability to make discoveries in mathematics, reflected on his time at HSE University, and spoke about his doctoral research at the Chinese University of Hong Kong.

— Dmitry, according to media reports, you asked an AI system to ‘make a scientific breakthrough’ by finding a counterexample to the Dinitz–Garg–Goemans conjecture. At first, the model refused, but after several prompts it agreed to take on the problem and soon produced a solution. Could you explain what this conjecture is and what makes the solution so remarkable?

— There is a branch of mathematics known as graph flows, whose results can be applied to optimising data traffic over internet cables or managing transport networks. Imagine you need to deliver goods to multiple destinations, with each shipment requiring its own route while avoiding traffic congestion. If you are allowed to split each shipment into smaller parts along the way, the problem is relatively easy for a computer to solve. In real life, however, cargo cannot simply be divided.

Dmitry Rybin

In 1999, mathematicians Dinitz, Garg, and Goemans proved a fundamental result: if shipments are transported intact rather than split into parts, the additional load on the transport network is only slightly greater. They also proposed a conjecture, now known as the Dinitz–Garg–Goemans conjecture, suggesting that under these conditions the overall transportation cost—calculated as the fee for using each road—would not increase. I first came across this conjecture in 2020, during my fourth year at the HSE Faculty of Mathematics, and I immediately wanted either to prove it or disprove it.

From time to time, I worked on the problem myself and with the help of OpenAI's o1 and o3 models. Then GPT-5.6 Pro, after reasoning for five hours, found a counterexample. It constructed a road network and carefully selected numerical values showing that, in this case, transportation costs necessarily increase.

— Is this counterexample really that significant for mathematics?

— Yes—not only for mathematics but also for practical applications such as logistics and internet traffic management. It will undoubtedly spark renewed interest in the problem and lead to further advances building on the work of Dinitz, Garg, and Goemans.

— Could no human researcher have arrived at this result independently?

— This represents genuine progress on a problem that had remained unsolved for 27 years. Once news of the result became public, I received emails from many colleagues working on graph flows, including researchers at MIT and other leading universities.

But the most interesting aspect is something else. What spread across social media was not only the news that the conjecture had been disproved, but also my rather absurd conversation with GPT-5.6 Pro. I kept insisting that it should ‘make a breakthrough’—and in the end, it did. My role was to choose the problem, remain persistent, and verify the solution.

— What conclusions can be drawn about the future use of AI in scientific research?

— The main conclusion is that researchers should become more ambitious and use AI to tackle far more complex and important problems. I have been observing AI's ability to make discoveries in mathematics for the past five years, and the field is now undergoing a profound transformation.

— Could you tell us a little about yourself? How did you end up at HSE University, and how did you eventually move to China?

— I grew up in Nizhny Tagil. During my final years at school, I studied at the lyceum (Specialised Educational and Scientific Centre) of the Ural Federal University. I scored the maximum 300 points across three Unified State Examinations (USE). I chose the HSE Faculty of Mathematics on the recommendation of friends.

Shenzhen

© Photo courtesy of Dmitry Rybin

After graduating from HSE University, I wanted to enrol in a PhD programme in AI in a major metropolitan city. In 2021, I moved to Shenzhen to study at the Chinese University of Hong Kong (CUHK). The city is located in southern China, on the border with Hong Kong.

I have now lived in China for five years. I completed my doctoral studies there and was awarded my PhD just a month ago. My dissertation, which lies at the intersection of mathematics and artificial intelligence, is on Machine Learning-Based Search Methods for Combinatorial Optimisation.

— What was the most challenging aspect of working on your dissertation?

— Adapting to the fact that research is a very long-term endeavour. My greatest strength is the ability to focus intensely on a specific problem over a short period of time. That works well in academic competitions and the Unified State Examinations, but research is a marathon rather than a sprint.

— While you were a student, at a meeting between top-performing USE graduates and the President of the Russian Academy of Sciences, you asked whether talented students could be paired with mentors from the Academy. Why was mentorship so important to you?

— Because a good mentor can accelerate a student's progress hundreds of times over. In that respect, the HSE Faculty of Mathematics is a truly unique place, where you can spend entire days listening to world-renowned researchers.

I remember arriving at the faculty at 11 am, when lectures began, and staying until midnight, when Prof. Dmitry Kaledin's final seminars ended. I also look back with great gratitude on Prof. Valery Gritsenko, Prof. Boris Feigin, and many of my other mentors.

During my PhD, my supervisor was the Canadian-Chinese researcher Tom Luo. The most important part of doctoral study is choosing the right research problem, and in that respect he is one of the world's leading experts.

Graduation ceremony

© Photo courtesy of Dmitry Rybin

— Is the way mathematics is taught in China very different from the approach traditionally used in Russia?

— In Chinese schools, children rarely ask teachers questions, and they tend to retain that habit after entering university. Professors try to change this by encouraging students to ask more questions and engage in debate, but they are not always successful. For mathematics—and indeed any scientific discipline—that is certainly a disadvantage. In Russia, the approach is more creative and individualised. A good example is the personal task system developed by mathematician Nikolay Konstantinov, in which each student receives an individual set of problems and discusses their solutions directly with the teacher.

As for the strengths of Chinese education, they lie in its highly systematic engineering approach. The university in Shenzhen is an extraordinary place, full of robots, drones, and artificial intelligence. Visitors from HSE University are often amazed by the scale of Chinese construction, the combination of skyscrapers and parks, the remarkably quiet roads thanks to electric vehicles, and, of course, the inexpensive bubble tea.

— Do visitors from HSE University come here often?

— I was the first to make the move, but in recent years students from the Faculty of Computer Science have started coming here on exchange programmes, along with research groups from both the Faculty of Mathematics and the Faculty of Computer Science. I help them find their feet in China.

— What are your plans for the next few years?

— Together with my Chinese friends, I am building a $100 million start-up focused on discovering new algorithms using AI. I plan to continue developing it here in Shenzhen. Before long, this city will be more important than New York, with Fields Medal and Nobel Prize winners relocating here from around the world.