Investors began to corner Yao Shunyu

"VCs around me are inquiring about Yao Shunyu's startup plans."

This prodigy is already well-known in the venture capital circle: a legend in the Department of Physics of Tsinghua University, a PhD in Physics from Stanford University, a top AI researcher, and only 29 years old. In September 2025, Yao Shunyu left Anthropic in a high-profile manner, and later joined Google DeepMind as a Principal Senior Research Scientist.

"I won't stay at Google for long," Yao Shunyu said frankly in an interview earlier this year.

Recently, he joined an online domestic event to discuss the currently trending RSI (Recursive Self-Improvement). This has subtly made the outside world more curious about his possible startup moves. "Leading VCs have already reached out to Yao Shunyu," a person familiar with the matter revealed.

This is always the case: the more turbulent the era, the more attention prodigies and "Yao Shunyu-like talents" will receive. Investors firmly believe that this group of young AI talents will change the direction of the world, and securing investments in them means seizing the alpha of the AI era.

The Most Sought-After Talent for Chinese Tech Giants: The Past of a Physics Prodigy

The 29-year-old Yao Shunyu is a prodigy that all Chinese tech giants want to recruit.

Looking through his resume, you can see he has taken a typical path of a top academic achiever. Born in Ningxia in 1997, he moved to Shanghai with his parents when he was a teenager. With remarkable talent in physics, he was admitted to the Department of Physics of Tsinghua University and won the Special Prize Scholarship for Undergraduates of Tsinghua University.

What caught the public's attention was in 2018, Yao Shunyu published a high-level paper in Physical Review Letters, one of the top academic journals in the field of international physics, which first proposed the topological band theory for non-Hermitian systems globally.

The research result is so high in gold content that an associate professor from a 211 university admitted: Even the professors in our university cannot match the physics level Yao Shunyu had achieved during his undergraduate study.

After graduating with a bachelor's degree, he went to Stanford University to pursue his PhD, focusing on cutting-edge issues such as quantum many-body chaos and open quantum system dynamics, studying under renowned physicists including Douglas Stanford and Zhenbin Yang. After graduating with a doctorate, he went to UC Berkeley for postdoctoral research.

Unexpectedly, after nearly 10 years of theoretical physics research, Yao Shunyu switched to the AI industry.

In October 2024, when large language models were at the peak of their popularity, Yao Shunyu left his postdoctoral position at Berkeley and joined Anthropic's Claude team, responsible for the fundamental reinforcement learning theory behind the Claude 3.7 Sonnet framework and the Claude 4 series. Only one year later, this prodigy smoothly switched jobs again to join Google DeepMind as a Principal Senior Research Scientist.

As for the reason for leaving Anthropic, he explained that 40% came from "fundamental differences in values", and 60% came from internal information of Anthropic that he could not disclose.

So far, it is enough to explain why leading VCs are keeping a close eye on Yao Shunyu: Top academic background, cutting-edge practical AI experience, amazing insight and extremely high incubation value.

VCs Are Waiting for Yao Shunyu to Start a Business

At the frontiers of AI on both sides of the ocean, there is a Yao Shunyu standing out respectively.

One is Yao Shunyu who has already returned to China. Also born in 1997, Yao Shunyu got a score of over 700 in the national college entrance examination and was admitted to the Yao Class of Tsinghua University to study computer science, then went to Princeton University to pursue his PhD. After graduation, he joined OpenAI, led the development of agent products and participated in the Deep Research project. At the end of 2025, Yao Shunyu left OpenAI and returned to China to serve as the Chief AI Scientist of Tencent.

Now VCs are paying close attention to the movements of the other Yao Shunyu in Silicon Valley.

I still remember that in an interview in the first half of this year, Yao Shunyu mentioned: "I won't stay at Google for long, I will find something worthwhile to challenge myself."

But at least for now, Yao Shunyu is still working at Google — in early September, when Google officially launched the Gemini 3.8 model, Yao Shunyu, as a core member, posted on social media that this is just a small step for the model, but a huge leap for RSI (Recursive Self-Improvement).

RSI is undoubtedly one of the hottest topics in the technology circle recently. Simply put, it means AI builds stronger AI by itself, and the new AI continues to build even more powerful AI, forming a positive cycle to become increasingly capable.

Currently, RSI-related startups are booming.

The latest case is Chaoyan Intelligence, founded by Chen Yongchao, assistant professor of the School of Artificial Intelligence of Tsinghua University, which recently completed nearly 400 million yuan in angel round and angel+ round financing. Another example is Zhang Muhan, assistant professor of the Institute for Artificial Intelligence of Peking University, who is also reported to have launched his RSI startup. In August, former Google Chief Scientist Jeff Dean and other co-founders launched Discovery Loop, whose first step is to automate machine learning research.

For domestic large model companies, DeepSeek and Tsinghua University jointly published a paper, in which the part of "Build environments of Agents, by Agents, for Agents" also revealed some RSI-related practices. In mid-September, Zhipu AI publicly released China's first RSI recursive self-improvement engineering practice implemented in the production environment, which enables large models to act as intelligent engineers to independently optimize the underlying inference operation system.

A thought-provoking detail is that around the topic of RSI, Yao Shunyu, who had not appeared in public for a long time, recently joined the online event of domestic AGI House. "RSI is not equivalent to ordinary RL or model self-generated data training, but a systematic engineering that allows AI to participate in the design of training processes, optimization goals and system closed loops," Yao Shunyu said. The hardest part of RSI is not making AI stronger, but verifying that this improvement is real, sustainable, and not contaminated by score brushing or reward hacking.

This once again sparked public speculation about this prodigy's startup plans and business directions.

The Golden Age for Prodigies

Undoubtedly, China has become the world's largest source of top AI talents.

Recent data released by the Carnegie Endowment for International Peace shows that among the global top AI talents in 2025, 57% of the researchers got their bachelor's degrees from Chinese universities. If we extend the timeline further, this change is even more obvious — from 2019 to 2025, the proportion of top AI researchers from China rose from about 29% to about 57%.

Looking around, from Agent, embodied intelligence, to world model, AI for Science, and then to RSI, a group of "Yao Shunyu-like talents" have already started their own businesses.

A typical example is Yu Jiahui. This world-renowned top AI scientist was born in 1995. After graduating from the Special Class for Young Talents of the University of Science and Technology of China, he went to the University of Illinois for his PhD, focusing on computer vision, speech and multimodal large model research. He used to work at Google DeepMind, OpenAI and was a key member of Meta's Superintelligence Lab. In August this year, Yu Jiahui officially announced that he would start his independent entrepreneurial path.

Earlier in May, Tian Yuandong, a well-known Chinese AI scientist and former research director of Meta FAIR, posted on Twitter to officially announce the co-founding of a new company called Recursive, with a $650 million first-round financing and a post-investment valuation of $4.65 billion. The startup direction is also related to RSI — Tian Yuandong said that the new team is building an AI that can automatically discover knowledge and recursively improve itself, which will fundamentally change the way scientific and technological progress is made.

In addition, many post-85s AI leaders including Dai Jifeng, former head of Alibaba's Qwen team Lin Junyang, Huawei's prodigy Zhi Huijun, Hu Yuanming and Ji Yu have all started their own businesses. Without exception, they all have outstanding educational backgrounds, experience working at top tech companies and cutting-edge industry vision.

"This is the best era for prodigies."

For investors who always want to get ahead of others, keeping a close eye on those prodigies who may start businesses and not missing any of them means not missing this era.

This article is from the WeChat official account "Pedaily" (ID: pedaily2012), written by Feng Yuchen, authorized for release by 36Kr.