Reverse Regulatory Capture — LessWrong