LLM Confidence Calibration

A research direction examining how well the output probabilities of language models align with their actual accuracy rates. It investigates whether a model's expressed confidence genuinely reflects its knowledge state and is a core topic in AI reliability and interpretability research.