A Heuristic Proof of Practical Aligned Superintelligence — LessWrong