Large Language Models Are Zero-Shot Text Classifiers