(Yicai) Oct. 16 -- A research team led by Xu Disheng, a professor at Great Bay University, has used a scientific multimodal artificial intelligence model developed by Shanghai AI Laboratory to prove the three-dimensional Ghys conjecture, marking the first time a domestic AI model has solved a world-class mathematical problem.

The proof of the Ghys conjecture, which has puzzled the mathematical community for 30 years, was completed using Intern-S2, Shanghai AI Lab announced yesterday. A team led by Dong Bin, a professor at Peking University, and Liu Ruochuan, an academician of the Chinese Academy of Sciences, has verified the findings, it added.

The paper on the proof has been published on the preprint platform arXiv, while the formal code was made available to the public on the code-sharing platform GitHub.

The Ghys conjecture, proposed by French mathematician Etienne Ghys in 1995, asks whether the behavior of chaotic dynamical systems can determine their geometric classification. It has long been regarded as a key open question in dynamical systems and a benchmark for testing mathematical reasoning capabilities.

A student of Fields Medal winner Artur Avila, Xu has long committed to proving the Ghys conjecture. After noticing the potential of AI models in math research last year, he visited the Shanghai AI Lab and formed a team that began collaborative research on model training at the start of this year.

Intern-S2, the result of the team upgrading the previous-generation Intern-S1, began exploring the Ghys conjecture from multiple angles in August, producing thousands of pages of derivations, based on which Xu and his partners continuously provided feedback to the model, corrected reverse exploration‌, and successfully found the proof method through multiple rounds of human-computer interaction.

The significance of this study lies in verifying a new scientific research paradigm, which is different from traditional math research that relies on long exploration by individual geniuses, Shanghai AI Lab pointed out. The proof systematically integrates expert intuition into the computer solving loop and creates a closed loop of "system exploration -- expert judgment -- achievement accumulation -- feedback correction," it added.

This paradigm can be replicated and used in more basic research fields with "huge search space but scarce experts" to achieve more original innovation breakthroughs from 0 to 1, the lab stressed.

AI's role in math research has attracted widespread attention. OpenAI said it spent 88 hours using 10,000 AI agents to solve the Navier-Stokes Millennium Prize Problem last month, while releasing 722 mathematical manuscripts covering multiple fields, including number theory, geometry, and algebra, earlier this month.

However, OpenAI's radical attempts have sparked controversy within the industry. After the US firm announced a proof of the Navier-Stokes Millennium Prize Problem, 25 Fields Medal winners, including Terence Chi-Shen Tao and Deng Yu, said that AI companies treat solving math problems as benchmark tests, which harms the mathematical community.

Editor: Martin Kadiev