[103rd TrustML Young Scientist Seminar] Talk by Prof. Huaxiu Yao (University of North Carolina at Chapel Hill) "Building AI Agents That Act, Learn, and Evolve"

The 103rd TrustML Young Scientist Seminar features Prof. Huaxiu Yao (University of North Carolina at Chapel Hill) presenting "Building AI Agents That Act, Learn, and Evolve." The talk introduces a full-stack approach to agentic LLMs that overcomes the static-after-deployment limitation of current models, covering Agent World Model (auto-synthesized executable training environments), SkillRL (extracting reusable skills from successes and failures), MetaClaw (learning from natural interactions and idle-time fine-tuning), SimpleMem (lifelong memory via semantic compression across text/image/audio/video), and ClawArena (benchmark for evolving information environments). Held hybrid; open-space attendance restricted to AIP researchers, online attendance is open.