17-year-old Kimi researcher Chen Guangyu says he has left Moonshot AI

Chen Guangyu, a developer born after 2000 who has been described as a young prodigy, said on social media on Oct. 15 that he had recently left the Kimi team at Moonshot AI, according to a Southern Metropolis Daily report published on Oct. 16. "The past year working at Kimi has been the best time of my life," he wrote. "I am very grateful to have worked with such outstanding and supportive colleagues and to have learned from them."

Chen first drew public attention in March, when Moonshot AI's Kimi published a technical report titled Attention Residuals that redesigned residual connections, the core structure large models had left unchanged for a decade. Observers noticed that besides Su Jianlin, one of Kimi's key figures, the paper's first authors included Chen, then a 17-year-old high school student in Shenzhen, which put him on the radar of a wider technical community as a researcher working on core model architecture.

The paper drew attention overseas soon after publication, with Elon Musk commenting at the time that it was "impressive."

According to public reports, Chen was born in Shenzhen. In middle school he tried registering a Shopify store using a parent's identity, organized a youth open source community, built robots and at one point fantasized about becoming a professional skier.

He first connected with large models in February last year, when he presented an idea called ThirdArm at a creative hackathon for middle and high school students. It was there that he met judge Dong Kehan, who encouraged him to go deeper into more advanced model technology.

It is worth noting that the news first came from Chen himself on social media and was only later picked up by media, rather than being announced by the company, so his own account is currently the only basis available to outsiders.

The report does not say why he left or where he is going next. For a large model team in fierce competition, the departure of a young researcher credited on a core paper is a signal worth tracking, and once more information becomes clear it will be another data point for reading talent flows in China's large model industry.