China’s technology companies were taken by surprise by advancements in generative artificial intelligence, with Beijing’s regulations and a slowing economy adding to the challenges.
In November, a year following the launch of ChatGPT, a relatively obscure Chinese startup quickly rose to prominence on a leaderboard assessing the capabilities of open-source AI systems.
The startup, 01.AI, despite being just eight months old, boasted substantial financial support, a $1 billion valuation, and was established by a prominent investor and technologist, Kai-Fu Lee. In discussions, Mr. Lee positioned his AI system as a viable alternative to offerings like Meta’s generative AI model, known as LLaMA.
However, there was a significant revelation: parts of the technology in 01.AI’s system were derived from LLaMA. Leveraging Meta’s technology as a foundation, Mr. Lee’s startup enhanced its system by training it with new data to enhance its capabilities.
This scenario exemplifies a stark reality acknowledged by many in China. Despite the nation’s rapid progress in generative AI development, Chinese enterprises heavily rely on foundational systems originating from the United States. As per insights from numerous tech industry insiders and top engineers, China currently trails the U.S. in generative AI by at least a year, with the gap potentially widening. This dynamic sets the stage for an intensified technological rivalry reminiscent of a modern-day cold war between the two nations.
Chris Nicholson, an investor at Page One Ventures specializing in AI technologies, emphasized that Chinese firms face significant pressure to match U.S. innovations. The debut of ChatGPT marked a pivotal moment akin to the “Sputnik moment” that spurred China into action.
Jenny Xiao, a partner at Leonis Capital focusing on AI-powered ventures, highlighted that the AI models developed from scratch by Chinese companies often fall short in quality, leading them to adopt refined versions of Western models. She estimated that China trails the United States by two to three years in generative AI advancements.
