Teen AI researcher leaves Kimi after a year, heads to Bay Area

Why it matters: China's AI labs are competing fiercely for top research talent, and stories of very young researchers landing big roles draw wide attention. Moonshot AI's Kimi models are among the country's best-known large language models, so its staffing moves are watched closely by the tech industry. The pull of Silicon Valley is also a recurring theme in discussions of China's AI talent pipeline.

Chen Guangyu, a 17-year-old high school student, said on social media that he has left Moonshot AI's Kimi team, calling the past year the best of his life and thanking colleagues he worked with and learned from.

Chen did not explain the departure or give next steps, but thanked the team for hiring him at 16 and said there are more things worth exploring. In a reply to a Kimi researcher who said the news was sad but understandable, Chen said he is going to the Bay Area and denied speculation that he would join Anthropic.

Chen first drew wide attention in March 2026 as a co-first author, with equal contribution, of the paper "Attention Residuals" alongside Kimi researchers Zhang Yu and Su Jianlin. The AttnRes method and its Block AttnRes engineering version later became one of the underlying technologies of Kimi K3.

Chen joined Kimi as a machine learning research intern in November 2025. His first key task was putting the attention-residual architecture into practice: the idea came from Su Jianlin, and Zhang Yu and Chen proposed Block AttnRes, which split the model's 93-layer network into eight blocks to keep memory and communication costs manageable, lifting training efficiency by about 20%.

Kimi K3 was released in July 2026 with 2.8 trillion total parameters, and AttnRes was one of its two underlying architecture updates. Around the same time as Chen's paper, Silicon Valley investor Deedy Das mentioned a 17-year-old holding five AI job offers each worth more than $1 million a year; observers guessed it was Chen, though the source does not confirm this.