Improving the modal UDT optimality result — LessWrong