A reward is a scalar signal that tells a reinforcement learning agent how good its action was, used to update its policy toward higher-reward behavior. Rewards can be sparse (only at episode end) or dense (every step), and reward design critically shapes what the agent learns. In RLHF for language models, a reward model trained on human preferences provides the reward signal. Reward hacking — where agents exploit loopholes to maximize reward without achieving the intended goal — is a persistent challenge.