Invalid Logic, Equivalent Gains: The Bizarreness of Reasoning in Language Model Prompting