Intelligence Amplification and Friendly AI — LessWrong