Soares, Tallinn, and Yudkowsky discuss AGI cognition — LessWrong