Explanation of Paul's AI-Alignment agenda by Ajeya Cotra — LessWrong