Anfonwch hwn fel neges destun: Adaptive representations for reinforcement learning