Enviar por SMS: Adaptive representations for reinforcement learning