Commentary on AGI Safety from First Principles — LessWrong