Save Big on Coursera Plus. 7,000+ courses at $160 off. Limited Time Only! Grab it Watch a research seminar presented by UCLA PhD student Da Yin exploring innovative approaches to training adaptable Large Language Model (LLM) agents. Learn about the Agent Lumos framework that enables better cross-domain generalization through unified training on converted reasoning trajectories. Discover how Trial and Error and Q* Agent implementations promote self-exploration and trajectory collection for preference optimization. Explore future directions in agent critique and world models aimed at enhancing LLM adaptability with minimal resources. The talk covers groundbreaking work in weakly supervised learning methods that address key challenges in developing LLM agents capable of handling unseen tasks and environments. Gain insights from an emerging researcher who has earned recognition including an Amazon PhD Fellowship and Best Paper Award at EMNLP Pan-DL workshop, while contributing to the field through workshop organization and academic service roles.