Yao Shunyu walks in wearing a pair of amber-tinted glasses that shift color with the light.
There are two Yao Shunyus in Silicon Valley's AI circles—both graduated from Tsinghua in the same class, making Yao Shunyu(姚顺雨) and Yao Shunyu (姚顺宇) frequent subjects of conversation.
The first, Yao Shunyu, jumped from OpenAI to Tencent in 2025 and joined our show last year (Episode 115). The second, Yao Shunyu, also switched jobs that same year, moving from Anthropic to Google DeepMind.
Today's guest, Yao Shunyu(姚顺宇), graduated from Tsinghua and Stanford. His former research field was theoretical physics—non-Hermitian systems, quantum physics, and high-energy physics. His life philosophy? "Always wanting to challenge things I'm not very good at."
His biggest leap came after finishing his PhD, when he decisively left physics after luke combs norman ok nine years and entered the brand-new AI industry. Over the past two years, he has served as a research scientist at both Anthropic and Google DeepMind, contributing to the development of key models including Claude 3.7, 4.5, and Gemini 3.
Yao Shunyu is anything but a nerd. Sometimes, he'll catch you off guard with a sudden "little crazy" remark.
Our program was recorded in March 2026. Since then, the world has seen hydro many unexpected developments: Meta's acquisition of Manus was revoked, Cursor may be acquired by SpaceX, xAI will cease independent operations and merge into SpaceX under the new name SpaceXAI, and more. If our conversation feels somewhat dated, jacob bethell we ask for your understanding—the world of AI simply changes too fast and too unpredictably.
We should also note that some technical details involve corporate confidentiality, and certain topics are inconvenient for our guest to discuss. We appreciate your patience on this front.
Throughout the interview, we have done our utmost to learn about AI alongside you. You will gain insights into Yao Shunyu's technical exploration journey and thinking at Anthropic and Google Gemini.
As for those rebellious, offbeat moments—please, allow us to go a little crazy.
And now, here is my interview with Shunyu.
DISCLAIMER: This content is not intended as investment advice.
---
中文标题：DeepMind姚顺宇：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
姚顺宇戴了一副茶色眼镜走过来，这副眼镜会随光线变化而改变颜色。
硅谷AI业界有两位Yao Shunyu，他们曾是清华同一届毕业生，这让姚顺雨与姚顺宇时常成为话题人物。
前一位姚顺雨2025年从OpenAI跳槽到腾讯，他去年来过我们节目（第115集）；后一位姚顺宇也于同年跳槽，从Anthropic来到Google DeepMind。
今天的嘉宾姚顺宇，毕业于清华和斯坦福大学，曾经的研究方向是理论物理——非厄米系统、量子物理与高能物理。他的人生奋斗姿态是——“总想挑战一些自己不太会的事”。
他人生最大的一次跨步是博士毕业，毅然决然离开深造9年的物理，来到崭新的AI行业。过去两年，他先后在Anthropic和Google DeepMind出任研究科学家，参与了Claude 3.7、4.5、Gemini 3等关键模型的开发过程。
姚顺宇一点也不nerd。有时，他会令你猝不及防，突然发表一点“小疯”言论。
我们的节目录制于2026年3月，距离我们这次节目录制完，世界又发生了许多意想不到的变化：Meta对Manus的收购被撤销、Cursor可能被SpaceX收购、xAI将终止独立运营并入SpaceX，并更名为SpaceXAI等等。如果我们的谈话内容有一定滞后性，请大家多包涵——AI的世界实在变化太快、太出乎意料了。
可能还要特别说明的一点是，技术细节会涉及企业机密，有一些嘉宾是不方便分享的，也请大家能够包容。
在访谈中，我们尽最大可能和大家一起学习AI。你会收获姚顺宇在Anthroic和Google Gemini的技术探索历程与思考洞察。
而对于那些离经叛道的微小片刻，还请允许我们小疯一下。
接下来，就是我对顺宇的访谈。
DISCLAIMER: 本内容不作为投资建议。
---
OUTLINE:
00:01:26 Two Shunyu Yaos——两个Shunyu Yao
00:07:15 Competition and Escape——竞争与逃逸
00:25:22 "Pre-training Is Not Over"——“Pre-train没有到头”
00:35:08 The Coding Explosion——Coding的爆发
00:50:10 Seedance——Seedance
00:54:30 "Hard Distillation" and "Soft Distillation"——“硬蒸”和“软蒸”
01:04:07 Robotics——机器人
01:08:45 Betting on the Underdog Territory——在Underdog之地赌一把
01:19:44 Non-Hermitian Systems and Quantum Physics——非厄米系统与量子物理
01:36:27 High-Energy Physics——高能物理
01:43:09 Physics and AI——物理与AI
01:52:32 Training Claude 3.7 and 4.5 at Anthropic——在Anthropic训练Claude 3.7和4.5
02:35:03 "AI Is Fundamentally Simple"——“AI本质是简单的”
02:41:10 Training Gemini 3 at Google DeepMind——在Google DeepMind训练Gemini 3
03:01:28 Tech Predictions and Organization Building——技术预测和组织搭建
03:23:33 The Triumph of Collectivism——集体主义胜利
