Investigating Human-Aligned Large Language Model Uncertainty