Enhancing Vision-Language Model Reliability with Uncertainty-Guided Dropout Decoding | Read Paper on Bytez