I’m a Research Scientist at Meta Superintelligence TBD Lab. I recently defended my PhD at CMU, advised by Maarten Sap, and received a Microsoft PhD Fellowship.
My goal is to build AI agents that help each of us realize our potential. To work well with people, agents need to (1) help us get things done (WebArena, TOM-SWE); (2) understand people and society (OdysSim, Social World Models, Sotopia); and (3) act safely and respect human values (HAICOSYSTEM, OpenAgentSafety).
News
| Oct 12, 2026 | Our team at Meta Superintelligence TBD Lab released Muse Wonder, a model family for personal and professional AI agents. |
| Sep 26, 2026 | OdysSim: Building Foundation Models for Human Behavior Simulation has been accepted to NeurIPS 2026! |
| Sep 2, 2026 | Our team at Meta Superintelligence TBD Lab released Muse Spark 1.3, with a focus on long-running agent workflows, user collaboration, and coding. |
| Aug 12, 2026 | I defended my PhD! [Slides] |
| Jul 8, 2026 | Excited that five papers were accepted to COLM 2026: Mind the Sim2Real Gap in User Simulation for Agentic Tasks, SOTOPIA-TOM, Social World Models, GoodPoint, and Training Proactive and Personalized LLM Agents. |
Connect
Feel free to contact me at xuhuiz@cs.cmu.edu