Another attempt to explain UDT — AI Alignment Forum