Improving the modal UDT optimality result — AI Alignment Forum