Large Language Models are Zero Shot Hypothesis Proposers