Fixing The Good Regulator Theorem — LessWrong