Investigating Uncertainty Calibration of Aligned Language Models under the Multiple-Choice Setting