Venture capitalist Lee Kai-fu launches AI start-up to seize on ‘historical opportunity’ to build Chinese LLMs

Lee’s new company Lingyi Wanwu, headquartered in Beijing, plans to develop its own large language model from scratch, according to Sinovation Ventures
The former head of Google China said China can only gain a competitive edge in the field of AI by creating its own LLMs rather than relying on foreign technology

Artificial intelligence

Xinmei Shen

Published: 5:30pm, 4 Jul 2023

Why you can trust SCMP

Lee Kai-fu, a prominent venture capitalist and former president of Google China, has unveiled his new artificial intelligence (AI) start-up, which aims to build its own large language model (LLM) as Chinese firms race to catch up with their US peers, including ChatGPT maker OpenAI.

Lee’s new company Lingyi Wanwu, headquartered in Beijing, has assembled dozens of core members that include former executives from Big Tech firms such as Alibaba Group Holding, Baidu and Didi Chuxing, according to Sinovation Ventures, a venture capital firm run by Lee, in a WeChat article published on Monday.

Alibaba is the owner of the South China Morning Post.

02:19

Is AI better at maths than mathematicians?

Sinovation said Lingyi Wanwu plans to take “the most difficult path”: developing its own LLM from scratch.

China had at least 79 LLMs with more than one billion parameters as of May and the number is still increasing as investors and entrepreneurs rush to tap the Chinese market, which is likely to remain closed to ChatGPT and Google’s Bard.

The investment frenzy has already fanned debates on whether China has poured too much money into LLMs, which are expensive to develop partly because the process requires many advanced chips.

Lee appears unfazed by the scepticism, saying that LLMs presents a “historical opportunity” that China must not miss, according to Sinovation.

If the country keeps relying on open-source models built by foreign companies, the firm said, it will face risks such as high fees and restricted access. Only by developing its own LLMs can China gain a competitive edge in the field of AI, Lee was quoted as saying.

Lingyi Wanwu, conceived in March, was officially established on May 16, according to Chinese corporate database Qichacha. It is helmed by Ma Jie, former head of metaverse operations at Baidu, who holds a 99 per cent stake. The remaining 1 per cent is owned by Sinovation.

Sinovation said within three months of Lingyi Wanwu’s conception, the start-up has already begun internal testing of an LLM that has tens of billions of parameters.

In general, the more parameters an AI model has, the more sophisticated it is. GPT-3, launched by OpenAI in 2020, had 175 billion parameters, while the latest GPT-4 is thought to have many more parameters.

Lee’s new venture is one of the most high-profile generative AI endeavours launched by Chinese tech executives.

Meituan buys out 4-month-old AI start-up Light Year for US$281 million

Wang Huiwen, co-founder of on-demand food delivery giant Meituan, announced in February that his new AI venture Beijing Light Year Technology had received a US$50 million investment.

In June, Wang stepped down from his corporate roles at Meituan, citing “personal health reasons”. The company subsequently acquired Light Year in a deal that involved US$233.7 million in cash and 336.9 million yuan (US$47 million) in assumed liabilities.

Meanwhile Alibaba, Tencent Holdings and Baidu are all developing their own LLMs, with internet search giant Baidu claiming recently that the LLM used in its Ernie Bot had outperformed ChatGPT in key metrics.