A 17-year-old researcher whose work helped reshape part of a leading Chinese AI lab's model architecture has left the company, and the tech industry is still trying to work out exactly where he is headed next.
Chen Guangyu, known internationally as Nathan Chen, was born in Shenzhen in 2009 and began teaching himself artificial intelligence around age 14, using large language models to work through the material on his own. A technical blog he kept drew attention from researchers after he took part in an AI hackathon in 2025, which led to an internship at a Silicon Valley start-up. By November of that year, at just 16, he had been hired as a research intern at Moonshot AI, the Beijing lab behind the Kimi family of models.
At Moonshot, Chen became one of three co-first authors on a paper describing attention residuals, a technique that alters a part of large language model design, the residual connection, that had gone largely unchanged for roughly a decade. The method improved the cost-efficiency and memory use of Moonshot's models and drew public praise from prominent figures in the field, including OpenAI co-founder Andrej Karpathy and Tesla's Elon Musk, who called it impressive work from Kimi.