Yao Shunyu Denies Founding RSI Company: Still a Google DeepMind Employee

Download the app

+1 Exclusive | Yao Shunyu denies starting a business: "I am still a DeepMind employee" The Paper reporter Yu Yan 2026-11-03 16:45 Source: The Paper ∙ Future 2% > Listen to full text Font size

"I have not founded a company, and I am still an employee of Google DeepMind." On the evening of November 2, regarding rumors that he was preparing an RSI (recursive self-improvement) company, Chinese AI scientist Yao Shunyu denied the news to The Paper Tech (www.thepaper.cn).

On November 2, reports said that market rumors claimed Yao Shunyu was advancing an AI company focused on recursive self-improvement (RSI), that the co-founding team had basically taken shape, and that one co-founder was a senior Google DeepMind employee. The team was also said to be attracting other research and engineering talent from leading AI labs such as Google DeepMind, Anthropic, and OpenAI, with the project valued at as much as $2 billion.

In response, Yao Shunyu said, "I have absolutely no idea where this fake news came from."

He was born in Ningxia in 1997 and later moved to Shanghai with his parents. He entered the Department of Physics at Tsinghua University for his undergraduate studies and won Tsinghua's top undergraduate scholarship. In 2018, Yao Shunyu collaborated with his advisor Wang Zhong to publish a high-level paper in Physical Review Letters, one of the top academic journals in international physics, giving the world's first theory on boundary states and topological invariants in non-Hermitian systems.

After graduating from Tsinghua in 2019, Yao Shunyu went to the Stanford Institute for Theoretical Physics to pursue a doctorate, and subsequently conducted postdoctoral research at the University of California, Berkeley.

Yao Shunyu once discussed on his personal blog why he gave up physics and chose AI. Source: Yao Shunyu's GitHub

He once took the initiative on his personal blog to discuss why he gave up physics and why he chose to join the field of artificial intelligence. "Mainly I wanted to find a direction that gives young people more opportunities." In his view, theoretical physics is very good for training people, tests people's intelligence and depth, and also requires methods from many fields; however, for many years, fewer and fewer experiments could be done in this field, and his remaining choices were mainly AI and quantum computing. "I believe quantum computing will be very important in the future, but in my view its current bottleneck is mainly the experimental platform. Therefore I chose AI."

In 2024, Yao Shunyu joined Anthropic and participated in Claude's tool calling and coding, as well as large-scale reinforcement learning and numerical stability work. In September 2025, Yao Shunyu joined Google DeepMind as a senior research scientist, mainly researching next-generation reinforcement learning algorithms.

Investors are paying close attention to RSI

RSI is the abbreviation of Recursive Self-Improvement. Take model training as an example: if an AI system finds a better training method, making the next-generation model better at designing experiments, and the new model then uses this ability to improve the next round of training, this constitutes a process of recursive self-improvement.

There have already been concrete attempts to let AI participate in its own R&D. In February this year, when OpenAI released the coding model GPT-5.3-Codex, it disclosed that the R&D team had used an early version of the model to debug the training process, analyze test and evaluation results, and manage deployment. In May, Google DeepMind announced that AlphaEvolve formed a loop of "generating algorithms - automatic evaluation - retaining winning solutions," with results entering data centers, chips, and AI training. In September this year, Yao Shunyu spoke publicly about RSI. At that time Google released Gemini 3.8 Flash, and as a core member he posted about it, saying: this release is one small step for the model, but a giant leap for RSI.

A financial advisor (FA) once told The Paper Tech that from July to September this year, investors' focus shifted to RSI and world models, and capital was more eager to find profiles resembling "young entrepreneurial prodigies" born after 2000. Yao Shunyu's resume fits very well.

A headhunter who has long followed embodied intelligence and the AI industry told The Paper Tech that since the second half of this year, investors and headhunters around them had been asking around, wanting to know whether Yao Shunyu would return to China to start a business. On October 12 this year, Yao Shunyu refuted rumors on a social platform, saying, "Currently I have no plan to raise funds in China. Some reports have left me confused as well. Please do not waste time on me. I am just an ordinary researcher."

Domestic capital has already begun betting on RSI. Public information shows that on September 16, Chaoyan Intelligence, founded by Chen Yongchao, an assistant professor at Tsinghua University's Institute for AI, announced the completion of angel and angel+ funding rounds totaling nearly 400 million yuan, for R&D of self-evolving foundation models. In late October, HyperVolve, founded by Chen Kai, former head of the large model center at Shanghai AI Laboratory, disclosed that it had completed an angel round of hundreds of millions of yuan led by BlueRun Ventures. The company lists recursive self-improvement as a key technical path for developing research-oriented foundation models.

In Silicon Valley, Tian Yuandong, former research director of Meta's Fundamental AI Research (FAIR) team, and others co-founded Recursive Superintelligence at the end of 2025. On October 21, the company released its first model, Flint, a 40-billion-parameter coding model that can run on a laptop. According to the company, its automated research system Eureka Machine participated in Flint's R&D, including optimizing training plans and operational efficiency. Responsible editor: Huan Yanhong Image editor: Chen Feiyan Proofreader: Yao Yiqi The Paper news tip line: 021-962866 The Paper, unauthorized reproduction prohibited +1 Favorite I want to report #ChineseAIScientist #YaoShunyu #DeepMind View more