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

In three linesStudy of alignment between LLM uncertainty and human uncertainty through behavioral analysis and internal activation patterns. Authors measure calibration and alignment across multiple-choice and open-ended factual recall datasets, assessing impact of instruction fine-tuning.

Summary generated by Claude — human-verified