Tag
This paper introduces methods to calibrate confidence in agentic systems using internal representations, demonstrating improved performance over baselines in multi-turn benchmarks.