At 16, most French teenagers are in the first year of the lycée. They are preparing for the French baccalauréat, revising their equations, wondering what they will do later.
At the same age, Chen Guangyu was joining the research team of one of China's most prominent AI start-ups. A year later, his name appears at the top of a paper that served as the basis for one of the largest open AI models in the world.
On 15 October, he announced that he was leaving the company.
A dizzying trajectory
Born in 2009, Chen Guangyu, who also goes by Nathan Chen, began taking an interest in machine learning around 2024. He did a short internship at a young Silicon Valley company, then joined the Kimi team at Moonshot AI, a Beijing start-up, as a researcher in November 2025. He was 16.
There he works on the very architecture of the models, that is, on the way they are built. He is one of the first three authors, with equal contributions, of a paper titled "Attention Residuals", one of the innovations at the heart of Kimi K3. Released in July, this 2.8 trillion parameter model is one of the largest open models ever published, and its performance has worried even Washington, where there is concern about the gap narrowing between American and Chinese AI.
An AI model is built like a building with many floors, where information rises from floor to floor. Since 2015, researchers have been adding "residual connections": on each floor, a copy of the information goes straight through a lift, so that it does not get lost along the way. The idea of Chen Guangyu's team is to make that lift intelligent: instead of automatically receiving what comes from the floor below, each floor can choose which previous floors it wants to listen to. A discreet change, which makes very large models more efficient.
A much-discussed departure
On the social network X, he described his year at Moonshot as the best of his life so far. He did not explain the reasons for his departure. According to several media outlets, he is reportedly moving to the San Francisco area. Asked by users about a possible move to Anthropic, he replied that it was not on the cards. His next step remains unknown.
His departure comes a few months before Moonshot's planned listing in Hong Kong, and in the middle of a talent war between American and Chinese labs. The biggest researchers are being fought over at gold-plated prices, whatever their age.
The symbol of an era
Chen Guangyu's story says several things about AI today.
The field is so young that you can shine very quickly. Modern AI, the AI of giant neural networks, is barely fifteen years old. There is no decades-long accumulation of experience, as in medicine or law, closing the door on newcomers.
You can learn alone, and fast. Online courses, open-access articles, and AI assistants themselves allow a motivated teenager to learn in a few years what once required a long course of study. We explain in detail at the end of the day how to learn AI by yourself.
But that is the exception, not the rule. At a time when labs are fighting over prodigies, ordinary companies are hiring fewer and fewer beginners: in France, permanent contracts for 21-29 year-olds have fallen by 28% in the occupations most exposed to AI, as the Banque de France and HEC study showed. The contrast is striking.
What we take away
At 17, Chen Guangyu has already left his mark on one of the most powerful models in the world. His journey is fascinating, and he deserves it.
It also raises a broader question, to which we devote our reflection this evening: do you have to be young to invent?