Search papers, labs, and topics across Lattice.
The Hong Kong Polytechnic University
3
0
6
Collective Skill Tree Search transforms LLMs into versatile agents capable of mastering complex tasks through a structured skill tree that enhances their adaptability and performance.
Get RL-level multi-turn LLM performance with SFT-level efficiency by decoupling trajectory generation and optimization via importance weighting.
Global AI aggregators can actually *worsen* learning compared to specialized, local ones, especially when they update too quickly.
