Tag
This paper proposes TRUST, a method that integrates uncertainty quantification into reinforcement learning reward design to improve LLM agents' tool-calling decisions, enhancing decision quality and maintaining reliable uncertainty estimates.