Reinforcement learning with imperceptible rewards — AI Alignment Forum