Some Thoughts on AI Alignment: Using AI to Control AI — LessWrong