"Solving" selfishness for UDT — AI Alignment Forum