Synthetic Restrictions — LessWrong