Could you be wrong: Debiasing LLMs using a metacognitive prompt for improving human decision making