Chinese tech entrepreneur says large language models should be ‘kept in a cage’ during initial adoption

Government bodies and enterprises in China should take a cautious approach towards adopting large language models (LLMs), the underpinning technology behind ChatGPT-like services, according to Zhou Hongyi, founder and chairman of Chinese cybersecurity firm 360 Security Technology.

As LLMs are still “unreliable”, government institutions and enterprises should tread carefully when they start using the technology in their products, Zhou said at an industry conference in Beijing on Sunday.

They should “keep LLMs in a cage” by having them independent and separate from existing businesses, and treat the technology as a “co-pilot”, Zhou was quoted as saying by the China Fund, a local newspaper. He made the remarks in a speech at the Global Digital Economy Conference, co-organised by the Beijing municipal government, the Ministry of Industry and Information Technology, and the Cyberspace Administration of China.

“If LLMs are closely incorporated into a company’s operations right away and are used to produce complicated applications, from my experience it could be hard to restrain,” Zhou said.

The comments come as investment interest in generative AI remains high in China, where Big Tech and start-ups are racing to produce ChatGPT-like services. China had at least 79 LLMs at the end of May, a Chinese government official said, and the figure is growing. The Chinese Academy of Sciences, the country’s top academy, launched its open LLM in mid-June.

As tech giants including Alibaba Group Holding, the owner of the South China Morning Post, Tencent Holdings and Baidu make headway on their own LLMs, a debate over whether start-ups have a chance of benefiting from the technology is also heating up in China.