“Reinforcement Learning from Human Feedback (RLHF) makes AI models appear aligned by shaping their behavior, but it may not be truly instilling aligned underlying values.”