Universal agents and utility functions — LessWrong