GPT-4 can exhibit a propensity to favor certain options in multiple choice answers over others (regardless of the option content), i.e., the model can show position bias