From Physics to DeepMind: How a Tsinghua Star Is Shaping AI Research

Google DeepMind recently hired a new senior research scientist, Shunyu Yao, a Tsinghua University physics alumnus and former recipient of the Tsinghua Undergraduate Special Scholarship who published in Physical Review Letters during his undergraduate years.

Yao left Anthropic on September 19 and joined DeepMind ten days later, continuing his work in AI research. During his year at Anthropic he helped build the reinforcement‑learning foundation team, contributed to the Claude 3.7 Sonnet framework and the theoretical foundations behind the Claude 4 series.

So Ant, it was good with you, but it is better without you :)

He describes his departure from Anthropic as driven by two main reasons: a fundamental values clash (about 40 % of the decision) and undisclosed internal issues (the remaining 60 %). He also cited the intense workload at Anthropic, which left him no time to write about his transition from physics to AI.

Yao explains that theoretical physics offers excellent training but has faced a lack of experimental progress for years, prompting him to seek a field more favorable to young researchers. After considering quantum computing, he chose AI because it resembles physics in its intellectual challenge while offering rapid experimental feedback.

He likens the current chaotic state of large‑model research to 17th‑century thermodynamics, where fundamental principles were unclear yet empirical laws emerged quickly. As a physicist‑turned‑AI researcher, he enjoys “not understanding the principles but continuously finding patterns.”

At Anthropic, Yao saw his research immediately impact cutting‑edge models and enjoyed the fast‑paced evolution of AI capabilities, a contrast to the slower feedback loops in physics. However, he felt unable to stay due to the values mismatch and other internal factors.

Yao’s academic background includes a breakthrough in condensed‑matter physics published in Physical Review Letters in 2018, where he introduced topological band theory for non‑Hermitian systems and defined two new physical concepts. He earned his PhD at Stanford under Douglas Stanford and Zhenbin Yang, working on quantum many‑body chaos and open quantum system dynamics.

After a brief post‑doctoral stint at UC Berkeley, he joined Anthropic, which actively recruits physicists for their rapid learning ability. During his year there he transitioned from academic research to industry AI, but notes that core researchers at Anthropic no longer write papers.

Now at DeepMind, Yao will continue AI research, bringing his cross‑disciplinary experience to the next stage of his career.