Paradigms of AI alignment: components and enablers — LessWrong