Landmark 2 stops to get here · leads to 4

Reward

The single number an environment sends back after each action, telling a reinforcement learning agent how good that step was; the agent learns to maximize its long-run total.

Your route here

2 stops · basics first
  1. Machine Learning ✓ understood

    Building systems that learn patterns from data instead of following hand-written rules, getting better at a task as they see more examples.

  2. Reinforcement Learning ✓ understood

    Learning through interaction with an environment, receiving rewards or penalties to learn optimal behavior policies.

  3. Reward · you are here ✓ understood

Picture it

  1. 01 Agent acts Takes action a in state s
  2. 02 Scalar reward r One number: how good that step was
  3. 03 Return G Sum of future rewards, discounted by γ
  4. 04 Policy update Shift toward actions with higher return
Notice that the reward is just one number per step; the agent learns by maximizing the sum of rewards over time, not any single one.

A reward is the feedback signal in reinforcement learning: after every action, the environment hands the agent a single number. Positive means that step helped, negative means it hurt, and zero means nothing much happened. There are no labelled right answers, only this running score.

The idea behind it is known as the reward hypothesis: any goal can be expressed as maximizing the expected cumulative sum of one scalar signal. Winning a game, balancing a pole, keeping a user engaged: each gets boiled down to “more reward is better.”

From reward to return

The agent doesn’t try to grab the biggest reward right now. It tries to maximize the return, the sum of all the rewards still to come. That’s why the diagram adds them up before anything gets updated. A discount rate γ between 0 and 1 makes a reward k steps away worth γ^k of the same reward received now. With γ at 0 the agent is myopic and only cares about the next step; close to 1, it plans far ahead.

Because the payoff can arrive long after the move that earned it, the agent has to work out which earlier actions deserve the credit. A value function estimates the return expected from a state, and the policy shifts toward actions that lead to higher values.

Designing the signal

Choosing the reward is where the human’s intent enters the system. The standard advice is to reward the outcome you want, not the steps you think lead to it. Sutton and Barto’s example: reward a chess agent for winning, not for capturing pieces, or it may learn to capture pieces and lose.

Sometimes no one can write the reward down. Inverse reinforcement learning infers one from expert demonstrations. RLHF trains a reward model on human rankings of model outputs, then optimizes a language model against it; that’s how InstructGPT was tuned.

The catch

An agent optimizes the reward you wrote, not the one you meant. This is called reward hacking. A 2016 paper on AI safety gives the example of a cleaning robot rewarded for not seeing any mess. It might learn to disable its own vision rather than clean. A learned reward model can be gamed the same way, which is why InstructGPT’s training added a penalty for drifting too far from the model it started from.

Where it sits

Explore nearby

In the research

All papers →

5 papers that build on Reward .

Frontier · Jan 2026 · 114 citations GDPO: Group reward-Decoupled Normalization Policy Optimization for Multi-reward RL Optimization Multi-objective RL is the norm in post-training; this fixes a subtle flaw in the default algorithm. Frontier · Aug 2025 · 112 citations On the Generalization of SFT: A Reinforcement Learning Perspective with Reward Rectification A tiny, theory-backed change to the most common training step in the stack. Frontier · Aug 2025 · 90 citations Pref-GRPO: Pairwise Preference Reward-based GRPO for Stable Text-to-Image Reinforcement Learning More stable RL for image generators, plus a better yardstick for them. Frontier · Nov 2025 · 75 citations DR Tulu: Reinforcement Learning with Evolving Rubrics for Deep Research It shows how to use RL where there is no verifiable answer. Frontier · Nov 2025 · 60 citations DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning Self-verification is a path to reasoning where no answer key exists.

Sources

  1. Richard S. Sutton and Andrew G. Barto, Reinforcement Learning: An Introduction (2nd ed.) . MIT Press, 2018; section 3.2, Goals and Rewards; section 3.3, Returns and Episodes
  2. Amodei et al., "Concrete Problems in AI Safety" . 2016; section 4, Avoiding Reward Hacking
  3. Ouyang et al., "Training language models to follow instructions with human feedback" . 2022 (InstructGPT)