Capability-Aligned Hierarchical Learning for Tool-Augmented LLMs
Researchers have introduced Capability-Aligned Hierarchical Learning (CAHL), a novel method for improving how large language models (LLMs) use external tools. CAHL addresses the common issue of misalignment between a high-level planning policy and a low-level tool-executing policy by jointly optimizing both. Experiments on various tool-use benchmarks, including API-Bank, BFCL, and Bamboogle, have shown CAHL's effectiveness in enhancing LLM performance. AI
IMPACT Improves LLM capabilities in complex, multi-step tasks requiring external tools.