About Me

I build large language model agents that improve themselves: agents that learn from their own experience and pick up new capabilities through exploration and discovery. My work combines reinforcement learning and post-training with inference-time scaling through search and planning. I am especially interested in how agents learn from feedback, adapt to unfamiliar tasks, and develop strategic reasoning and social skills, with applications to coding agents and multi-agent collaboration in open-ended environments.

Along the way this has touched reward modeling and credit assignment, adaptive curricula and online learning, and learning from synthetic data, self-play, and self-reflection.

I came to this from an unusual direction. I trained as a mathematician and economist at Reed, studied financial mathematics at Chicago, and worked on market and mechanism design before moving into reinforcement learning. That background still shapes how I think about agents: as systems whose incentives, algorithms, and learning dynamics have to be designed together.

I enjoy collaborations. If you have an interesting problem, reach out.

"Know what you know and know what you do not know. That is true wisdom."

In modern terms: know the known knowns, the known unknowns, and the unknown unknowns. This idea guides my research and reflects what I see as the central challenge in building intelligent systems: understanding and expanding the boundaries of one's own knowledge.

Outside of research, I enjoy playing and designing board games, reading science fiction, electronic music composition, grand strategy games, fencing, and squash. I find well-designed games to be elegant systems and rich sources of insight into planning, reasoning, and interaction.