I try to understand and harness general, robust, and emergent intelligence.

My research focuses on frontier deep learning and reinforcement learning architecture, as well as their application in LLM. Recently, I am working on multi-agent RL, meta RL (learning to learn), multi-step reasoning, and efficient training & inference (e.g., TTT). I am also interested in game theory and mechanism design.

Until recently, I researched stress-testing and red-teaming with Anthropic. I also contributed to various in-house projects and collaborated with ByteDance, Qwen, and Citadel Securities.

I love Go and Chess. I earned a 7th Dan and Candidate Master (CM). Some of my informal thoughts can be found here, which represent my thinking at some particular stage (though I am constantly learning, and my perspectives evolve).