LLMs: Baidu in US crosshairs

On January 12, 2024, the South China Morning Post (SCMP), Hong-Kong’s main English language newspaper, revealed that the Chinese army was leading a research project on AI. The latter aims to use large language models (LLMs) in strategic decision-making processes. A military laboratory is thought to have tested two LLMs: Spark, designed by voice recognition specialist iFlyTek, and Ernie, published by Baidu, China’s leading search engine.

As US generative AI models are banned, Ernie is considered the most efficient LLM available in China. With 100 million monthly users, it is moreover an undisputed industry leader in the country.

With the SCMP’s revelations, Baidu runs the risk of US sanctions, as the United States seeks to limit the use of AI by the Chinese army as much as possible. This is one of the motives behind the ban on exporting the most advanced Western chips to China. The harshest sanction would be adding Baidu to the Entity List: this would prohibit the Chinese company from entering any trade relations with companies using US tech.

Investors have taken this threat seriously: Baidu’s share price on the Hong-Kong stock exchange suffered an 11% drop on January 15, 2024. Aware of the risk, the Chinese company denied any ties to or partnership with the Chinese army laboratory. “If they used our large language model, it was the publicly-available online version,” argues Baidu.