Kai-Fu Lee, chairman and executive director of Sinovation Ventures, spoke during the HICOOL Global Entrepreneur Summit on September 11, 2021 in Beijing, China.
China News Service | China News Service | Getty Images
BEIJING – China’s artificial intelligence model may take at least half a year longer than that developed in the US, but China’s AI applications will be faster, said Kai-Fu Lee, the former head of Google China.
They refer to large language models, which are trained on the amount of data they can process and generate text, images and videos.
Top Chinese companies’ LLMs are about six to nine months behind their U.S. counterparts, while less advanced Chinese models may be in the U.S. about 15 months, Lee said. He spoke at the AVCJ Private Equity Forum China on Wednesday.
Lee, author of “AI Superpowers: China, Silicon Valley, and the New World Order,” is a commentator on AI, and co-founder of startup 01.AI as well as venture capital firm Sinovation Ventures.
“Applications, I would predict, at the beginning of next year will develop in China faster than in the US,” said Lee, noting that the cost of training good AI models has dropped significantly.
“It is inevitable that China will (build) the best AI applications in the world,” he said. “But it’s not clear whether it will be built by a large company or a small company.”
Lee, whose initial focus is on search apps proper, says it will take five to eight years to take generative AI consumer apps to the next level — one “super app” that can perform multiple tasks.
The industry may need new devices and existing smartphones, he said, adding that “the right device should be always on, always listening.”
Major Chinese companies such as Alibaba and Tencent has released AI models and business products. These companies and investors have also backed several AI startups.
Beijing-based ShengShu Technology, backed by Alibaba-affiliated Ant Group, announced Wednesday that its Vidu text-to-video model has introduced a new feature to improve how key elements or characters in AI-generated clips can be portrayed consistently, without distortion. . . That can allow advertisers to create promotional videos for their products.
Vidu was released earlier this year and its basic tools are open to the public, with more advanced capabilities available via subscription. Co-founder and CEO Jiayu Tang told reporters Wednesday that several companies are interested in buying ShengShu’s services, and they are not just exploring technology.
