How much reliable is ChatGPT's prediction on Information Extraction under Input Perturbations?