Unfaithful Explanations in Chain-of-Thought Prompting — AI Alignment Forum