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