Qinghua Liu

I'm a Research Scientist at Meta TBD Lab. I work on STEM RL and synthetic data.

Before that, I was a Research Scientist in (mid, post]-training at OpenAI, where I worked on the critical path of training Codex and ChatGPT, including GPT-5.5, GPT-5.6, and GPT-6 Astra. I also very indirectly contributed to specialized models such as GPT‑5.5‑Cyber and GPT‑Rosalind. Even earlier, I worked on GPT-5 and GPT-5.1 posttraining.

I received Ph.D. from Princeton, advised by Chi Jin. My PhD works focused on RL theory, including partially observable RL (POMDP, PSR, Survey), multi-agent RL (Stochastic Game), and RL with large state spaces (Function Approximation). During PhD, I've interned at DeepMind London with Csaba Szepesvári. After PhD, I spent a wonderful year as a postdoctoral researcher at Microsoft Research NYC.

I received a B.E. in Electrical Engineering and a B.S. in Mathematics from Tsinghua University.

Google Scholar, Twitter