Kimi Researcher Chen Guangyu announces departure from the Moonshot AI team
BIBIBI
AT A GLANCE
17-year-old AI researcher Chen Guangyu said he has left the Moonshot AI Kimi team, but did not disclose the reason or his next destination.
Article
Kimi Genius youth Chen Guangyu resigns: 16 joined at age, participated in K3 underlying R&D; Musk liked it
Beating AI news, 17-year-old AI researcher Chen Guangyu (Nathan Chen) announced that he recently left the Moonshot AI Kimi team. He joined in 2025 November at the age of only 16, and served for less than 1 years. Chen Guangyu said that the past 1 years were the best experience of his life so far, but did not disclose his reason for leaving or his next destination.
Chen Guangyu previously drew attention for Kimi's Attention Residuals(Attention Residuals) paper. This research, which received public praise from Musk, listed him, Zhang Yu, and Su Jianlin as co-first authors. Su Jianlin initially proposed replacing traditional residual connections with an attention mechanism and validated the preliminary results. Chen Guangyu and Zhang Yu subsequently joined the research and jointly proposed the more memory- and communication-efficient Block AttnRes, enabling this architecture to be used for large-scale model training; it was later adopted by Kimi K3.
In his resignation post, Chen Guangyu also reviewed other work. He optimized the Kimi Delta Attention of GPU computational kernels with Zhang Yu, researched model-training optimizers, and participated in producing showcase examples for the release of K3. He thanked the team for giving him an opportunity when he was only 16 years old, and said he wanted to continue exploring more directions.
Original link https://m.theblockbeats.info/flash/371927
Key points
01
Chen Guangyu said that he recently left the Moonshot AI Kimi team.
02
The article states that Chen Guangyu is now 17 years old and joined the team in 2025 November; he was 16 years old at the time and served for less than 1 years.
03
Chen Guangyu did not disclose his reason for leaving or his next destination.
04
The article states that Chen Guangyu, Zhang Yu, and Su Jianlin were co-first authors of the Attention Residuals paper.
05
The article states that Chen Guangyu and Zhang Yu jointly proposed Block AttnRes, which was later adopted by Kimi K3.
06
Chen Guangyu said that he optimized Kimi Delta Attention's GPU computational kernels with Zhang Yu, researched model-training optimizers, and participated in producing K3 showcase examples at its release.
AI-assisted interpretation
The following is analysis, separate from reported facts. Verify important claims independently.
This news brief primarily concerns Chen Guangyu's departure and the R&D experience he described. The article says that he joined the 16 team at age Kimi and participated in Attention Residuals、Kimi Delta Attention and K3 related work. No reason for his departure or subsequent destination has been provided.
Why it matters to readers
Chen Guangyu participated in the model architectures, computational optimization, and showcase examples listed in the article; therefore, his departure may draw outside attention to personnel changes within the relevant team. However, the available materials do not state that product delivery or R&D progress has been affected.
For beginners, this is a AI research personnel change story that also involves concepts such as model-training architectures, memory usage, communication overhead, and GPU computational optimization. At this stage, it is not advisable to infer solely from the departure news that Kimi products or technical roadmap will change.
Risks and unknowns
The materials do not provide a direct link to Chen Guangyu's resignation post, so its original wording cannot be verified on this basis.
The materials do not include official confirmation from Moonshot AI regarding the departure news.
The reason for departure and next destination are unknown.
The available materials do not state whether the departure affects Kimi K3 or other R&D work.
The specific original content of Musk's public praise for the related research is not shown in the materials.
Related Developments
Loading event timeline…
Related concepts
Attention Residuals
This term is not in the glossary yet. Browse related concepts in the glossary.