UDT from an RL perspective — AI Alignment Forum