(AsianFin)— Many large models are constructed using a "shell" and assembly method, and its computing power is also bottlenecked despite significant progress in artificial intelligence (AI), said Xue Lan, the Dean of Tsinghua University's School of Public Policy and Management.

Xue, who is also the Director of the International Governance Research Institute of Artificial Intelligence, made the comments at the China Development Forum.

Currently, many foreign models are open source, so it is possible to create a shell model based on the open source, and then assemble some such large models together to form a larger model, said Xue, adding that the originality behind these models created is “limited.” 

In addition, China’s computing power is restricted as Nvidia's GPUs A100 and H100 are banned from shipping to China, said Xue. Furthermore, he pointed out that China's data quality is relatively low, and yet to be industrialized, with relatively a small number of standardized data service providers.

With the advent of ChatGPT in November 2022, technological innovation in the AI field has accelerated, especially with leading US technology companies such as OpenAI in a tight race. This year, OpenAI launched a unique and cross-modal video generation tool, Sora, stunning the world and potentially causing significant disruption to many industries such as short videos and media entertainment. Meanwhile, different fields such as autonomous driving, advertising, education, healthcare, and security are also greatly impacted and may become potential markets for Sora, Xue remarked.

At the same time, the United States has also made many new breakthroughs in the core hardware of AI computing power, such as the Groq LPU, a new large model inference chip. Although its process is still 14nm, due to the TSA architecture, its specific processing capability is very fast, with inference speed 10 times faster than GPUs, but power consumption is only one-tenth of it, Xue added.

Xue admitted that with continuous breakthroughs in AI technology by American companies such as Groq, Nvidia, and OpenAI, the technological gap between China and the United States in large models is widening, posing a significant challenge to China's innovation system.

China's current innovation model still follows a relatively traditional linear model, starting from academic research, then engineering technology, and finally products. This model cannot form an ecosystem that is closely integrated between research, engineering, and the market, which is an urgent problem for China's innovation system to solve, Xue mentioned.