Asymptotic Risk Calibration for Selective Question Answering

Shufan Lin, Sijin Dong

Large language models (LLMs) may generate fluent but incorrect answers, making uncertainty quantification important for reliable question answering. However, heuristic uncertainty…