Beyond Overconfidence: Foundation Models Redefine Calibration in Deep Neural Networks