Unfaithful chain-of-thought as nudged reasoning — AI Alignment Forum