Moonshot AI's 17-year-old Kimi K3 co-author Chen Guangyu heads to Jeff Dean's Discovery Loop

# Moonshot AI's 17-year-old Kimi K3 co-author Chen Guangyu heads to Jeff Dean's Discovery Loop Source: [https://vectorwire.ai/article/moonshot-ai-s-17-year-old-kimi-k3-co-author-chen-guangyu-heads-to-jeff-dean-s-di-c5c8a8](https://vectorwire.ai/article/moonshot-ai-s-17-year-old-kimi-k3-co-author-chen-guangyu-heads-to-jeff-dean-s-di-c5c8a8) Chen Guangyu, the 17\-year\-old AI researcher who co\-authored a core architectural technique behind Moonshot AI's 2\.8\-trillion\-parameter[Kimi K3](https://vectorwire.ai/models/kimi-k3)model, has reportedly joined[Discovery Loop](https://vectorwire.ai/entity/discovery-loop), the U\.S\. startup founded by former[Google](https://vectorwire.ai/models/vendors/google)chief scientist Jeff Dean[1](https://vectorwire.ai/article/moonshot-ai-s-17-year-old-kimi-k3-co-author-chen-guangyu-heads-to-jeff-dean-s-di-c5c8a8#source-1),[2](https://vectorwire.ai/article/moonshot-ai-s-17-year-old-kimi-k3-co-author-chen-guangyu-heads-to-jeff-dean-s-di-c5c8a8#source-2),[3](https://vectorwire.ai/article/moonshot-ai-s-17-year-old-kimi-k3-co-author-chen-guangyu-heads-to-jeff-dean-s-di-c5c8a8#source-3)\. Chen announced his departure from Moonshot AI on October 15 via X, writing that the past year had been the best time of his life and thanking colleagues, particularly Du Yulun, who hired him at age 16\. He indicated he was heading to the Bay Area but did not name his next employer[4](https://vectorwire.ai/article/moonshot-ai-s-17-year-old-kimi-k3-co-author-chen-guangyu-heads-to-jeff-dean-s-di-c5c8a8#source-4)\. Multiple reports, including one citing the social media account Apper Grok, which tracks AI talent moves, identified Discovery Loop as his destination\. As of October 21, Chen had not publicly confirmed the move, and Discovery Loop declined to comment on individual personnel matters\. ## Chen's work at Moonshot AI Chen was a co\-first author, alongside Zhang Yu and Su Jianlin, of the paper Attention Residuals, a technique that became part of the core architecture of Kimi K3\. Kimi K3, released in July, is a model with 2\.8 trillion parameters\. Chen first came to industry attention after participating in an AI hackathon for middle school students hosted by Moonshot AI\. After his departure announcement, Sholto Douglas, head of scalable reinforcement learning at[Anthropic](https://vectorwire.ai/models/vendors/anthropic), publicly invited Chen to work together in San Francisco\. Chen declined, replying on X with a denial and later separately refuting rumors that he would join Anthropic\. ## Discovery Loop's ambitions Discovery Loop was founded in August by Jeff Dean, Sanjay Ghemawat, Quoc Le, and Oriol Vinyals, all former Google researchers\. The company pursues the automation of scientific and engineering research using AI\. Jeff Dean has described the work as similar to efforts to achieve recursive self\-improvement of AI\. Discovery Loop has not yet launched a product but is reportedly seeking funding at a target valuation of approximately $50 billion\. Moonshot AI, meanwhile, is reportedly preparing for a Hong Kong Stock Exchange listing, aiming to raise up to $5 billion\. ANALYSISThe move, if confirmed, would mark a notable instance of a China\-trained young researcher crossing to a U\.S\. frontier lab at a moment when both countries are competing intensely for AI talent\. Discovery Loop's recruitment of a teenager who contributed to a 2\.8\-trillion\-parameter model's architecture underscores the startup's appetite for researchers with hands\-on experience at scale, even before it has shipped a product\. The day’s vector, in your inbox\. **Vector Wire Daily**: one dispatch each publishing day\.