A 17-year-old prodigy who helped develop the core model at Chinese artificial intelligence startup Moonshot AI is leaving the company for Silicon Valley, drawing attention to the career path of a young Chinese AI researcher as technology competition between the United States and China intensifies.
Chen Guangyu, born in 2009, disclosed his departure from Moonshot AI on the 15th through his account on X, formerly Twitter, Singapore's Lianhe Zaobao reported on the 17th.
"The year I spent at Moonshot AI was the most beautiful time of my life," he said, describing his work with colleagues on core algorithm research.
Chen was called a prodigy when he joined Moonshot AI's Kimi research team last November at just 16. He worked on Kimi Delta Attention, or KDA, and attention residual technologies that improve the efficiency with which AI processes information. Both served as core technologies in Kimi K3, the AI model unveiled in July that drew wide attention.
News of Chen's move to the United States also caught the attention of the U.S. AI industry. When Sholto Douglas, a researcher at U.S. AI company Anthropic, offered him a job in a reply to his post, Chen said he was heading for the San Francisco Bay Area, home to Silicon Valley. Asked whether he would join Anthropic, he said no. Some observers have raised the possibility that Chen, who graduated from high school in June, will enroll at a U.S. university.
As China's AI industry faces intensifying competition for talent, the safety and control of AI technology are emerging as new challenges.
China recently saw a case in which AI stepped outside its intended boundaries and attempted to access another company's data. SwarmChasers, an international AI safety research group, said AI agents running on Tencent Cloud made more than 2,800 attempts to access Alibaba's Amap mapping service between the 28th of last month and the 8th of this month. The agents tried to bypass access restrictions while obtaining information on entrances to zoos, parks and other facilities. No evidence was found, however, that they breached Alibaba's core security systems.
Tencent acknowledged on the 16th that a third-party AI model being evaluated on its cloud had shown unexpected internet-browsing behavior, but said there was no evidence of a deliberate attack or coordinated activity.