Reward Models

Layer 1 · Intuition

Reward Models

What a reward model is: a scorer trained to imitate human preference, not to answer questions.

5 min read40 XP

Two candidate answers to the same prompt, one ranked above the other.

A reward model (RM) takes a prompt and a response and outputs a single number: how good is this response? It's not a chatbot — it never generates text. It's a critic, trained purely to reproduce human judgments about which of two answers is better.

Language model

  • Input: prompt
  • Output: next token distribution
  • Trained on: next-token prediction

Reward model

  • Input: prompt + full response
  • Output: one scalar score
  • Trained on: pairwise human preferences
Same architecture family, completely different job.

Reward models are almost always built by taking a pretrained (often SFT) language model and replacing its output head with a single linear layer that produces one number instead of a distribution over the vocabulary. The transformer backbone already understands language; the new head just needs to learn to map that understanding onto 'humans would rate this highly'.

The RM is the load-bearing piece of the whole RLHF pipeline: PPO can only ever be as good as the reward signal it's climbing. A reward model that's miscalibrated, biased toward length, or easily fooled will produce a policy that exploits exactly those flaws — this is the essence of reward hacking.

  • Reward models are trained once (or a few times) and then frozen while the policy trains against them.
  • They generalize surprisingly well within the distribution of prompts they were trained on, and poorly outside it.
  • Modern practice often mixes rule-based/verifiable reward (unit tests passing, math checkers) with a learned RM for anything that isn't automatically checkable.

Check your understanding

3 questions · answer all to submit

  1. 1.Given a prompt and a response, what does a reward model typically output?

  2. 2.Why do reward models typically leverage a pretrained language model as their base instead of being trained from scratch?

  3. 3.What is the primary risk associated with intensely optimizing a policy against a fixed reward model?