China Association for Science and Technology Releases Two AI Large Model Achievements: “Lizhi” and “Caihua”

The “New Tiangong Kaiwu” scientific and technological achievements release conference, hosted by the China Association for Science and Technology, premiered its AI special session on September 30, releasing two independent innovation achievements from China’s artificial intelligence technology field, corresponding respectively to the “Lizhi” UniMoE series multimodal large model and the “Caihua · MatVerse” materials multimodal large model. Among them, the “Lizhi” system, developed by the team at Harbin Institute of Technology, Shenzhen, focuses on issues such as complex semantic understanding, natural language generation, and crossdomain adaptation, emphasizing a new AI paradigm native to language intelligence, and has already been implemented in multiple enterprises. The other achievement, “Caihua · MatVerse,” was jointly launched by the Suzhou National Laboratory, Shanghai Artificial Intelligence Laboratory, and Shanghai Jiao Tong University. It builds fullchain capabilities from data to application for materials research and has established an evaluation benchmark covering 19 categories of materials secondary disciplines. At the release conference, several experts introduced the two achievements. The report noted that “Lizhi” possesses complete independent intellectual property rights, while “Caihua · MatVerse” is the world’s first materials large model with tens of billions of parameters and the first native multimodal materials large model. The related achievements have also been included in the digital exhibit repository of the scientific and technological achievements exhibition.