micahcarroll comments on On Targeted Manipulation and Deception when Optimizing LLMs for User Feedback