Understanding and Mitigating Miscalibration in Prompt Tuning for Vision-Language Models | Read Paper on Bytez