Human-Alignment, Calibration, and Activation Patterns in Large Language Model Uncertainty | ArxivCSExplorer