On April 3rd, according to media reports, at the recent China Digital Economy Development and Governance Academic Annual Conference (2024), Professor Xue Lan, Dean of Tsinghua University's Schwarzman College and Director of the International Institute for AI Governance, stated in his speech that many large models in China are "shells".
Xue Lan stated that the current development trend of AI in China is basically stable, with more than 130 large models domestically. Although there has been significant progress in terms of quantity alone, in reality, there are still many problems with large models in China.
He believes that many large models are built using a method of "shelling" and assembly, and computational power is also being constrained.
Xue Lan said: "Currently, many foreign models are open source. By shelling on the basis of open source, a shell large model can be formed. Then, by assembling some of these large models together, it becomes even larger. The originality behind large models made in this way is limited."
"Furthermore, our computational power is also constrained. For example, Nvidia's GPU A100 and H100 are banned from sale in the United States, which has affected many companies."
At the same time, he also believes that China's data quality is relatively low, there is not yet true industrialization, there are relatively few data service providers that are standardized, and China's capital is short-sighted, etc., which are all issues that need to be considered and addressed for the next step in the development of the large model industry in China.
