Chinese scientists use homegrown LLM to solve world-class math problem - Global Times

Chinese scientists have used a homegrown large language model to solve a world-class math problem, according to the headline of a report in the Global Times. The article states that Chinese researchers cracked a world-class mathematical conjecture using a Chinese AI model, and that this was achieved for the first time. A Shanghai-based AI lab gave the account to the newspaper. According to that account, the work was carried out through human-machine collaboration on what the lab called a 'general-specialized fusion' technical path.

The article also states that a formal verification of the proof was completed by two research teams at Peking University. Those teams used their self-developed SHEAF framework to carry out the verification. The excerpt does not name the conjecture, the model, or the Shanghai-based lab. It does not describe the size of the model, how it was trained, or the role each side played in the collaboration.

The framing in the report centres on the nationality of the model and on the Chinese institutions involved in the work. The excerpt presents the result as the first time a Chinese model has achieved this, without comparing it to models from other countries. Beyond the Peking University verification and the 'general-specialized fusion' description, no further technical detail is given. The report does not state when the work was completed, how long it took, or where the proof will be published.