Yao Shunyu / 姚顺宇
Yao Shunyu is the AI researcher interviewed in 140. 对姚顺宇的4小时访谈：请允许我小疯一下！在Anthropic和Gemini训模型、技术预测、英雄主义已过去. The source distinguishes him from another AI researcher with the same Chinese name: this Yao came from theoretical physics, studied physics at Tsinghua, completed a Stanford high-energy theory PhD, briefly held a Berkeley postdoc, then joined Anthropic in 2024 and Google DeepMind in 2025 in that earlier episode’s timeline.
176: 姚顺宇，来到腾讯300天 adds the later Tencent chapter. The source says Yao wanted to return to China in 2025 and became the young leader Tencent backed to rebuild Tencent Hunyuan / 腾讯混元 after the DeepSeek shock. In that account, his value is not only technical reputation from OpenAI/frontier-model circles, but also his ability to reorganize pretraining, post-training, evaluation, and infrastructure teams under executive cover from Martin Lau / 刘炽平 and Lu Shan / 卢山.
Technical Position
Yao treats frontier model progress as a problem-definition and systems-engineering challenge. In the episode, he says coding succeeded early because feedback is clear and code data is strong, while the next frontier is ML Coding and Long-Horizon AI: models that can work through longer chains, run experiments, analyze failures, and manage limited context as if usage were much longer than training.
Operating View
The source presents Yao as anti-mythological about AI work. He argues that individual heroism has mostly passed in large language-model training and that the valuable researcher is reliable, detail-oriented, responsible for global system effects, and able to place local work inside a larger training and product pipeline.
Connections
- Anthropic — company where he worked on large-scale reinforcement learning and coding-related model work, according to the source.
- Google DeepMind and Gemini — current organization and model/product context in the episode.
- Claude Code — Anthropic product direction he later reads as important to Anthropic’s productization.
- Long-Horizon AI and ML Coding — main technical directions he names for his current work.
- Frontier Model Scaling, AI Coding Verification, and Context Engineering — technical themes he qualifies through training, feedback, and context-management arguments.
- AI Organization Design and Research Taste — organizational and research-method themes he develops through physics, Anthropic, and Google experience.
- Tencent Hunyuan / 腾讯混元, Tencent, Martin Lau / 刘炽平, Lu Shan / 卢山, and Federated AI Organization — later Tencent model-organization chapter added by episode 176.