← All methodologies

Reinforcement Learning from AI Feedback

RLAIF
An AI helps judge the practice.

An AI evaluator supplies feedback. A common setup trains a reward model on its comparisons, then uses that model for reinforcement learning.

The idea

A teacher writes a grading rubric. An automated teaching assistant uses it to compare essays. More essays can be graded, but a mistake in the rubric or grader can affect many students.

The judge is a model following a rubric.

The task

Give useful feedback on a student's draft.

Answer A

Your example is clear. Add one sentence explaining how it supports your point.

Answer B

This is bad. Rewrite it.

Feedback

An AI judge uses the rule “be helpful and respectful” and prefers A.

AI preference labels → train a reward model

Its scores guide the assistant's RL practice.

One common RLAIF pipeline. Some variants use AI-generated scores directly instead of first training a separate reward model.

How RLAIF works

  1. Set the judging rules

    People specify the behavior they want.

  2. Get AI feedback

    An evaluator compares candidate answers using those rules.

  3. Turn feedback into rewards

    For example, train a model to predict the evaluator's preferences.

  4. Train and check

    Use RL to improve the assistant, then check its behavior independently.

Useful for

Collecting more preference feedback than people can feasibly label by hand.

The catch

An AI judge can be biased, fooled, or confidently wrong. AI feedback does not remove the need for human choices and evaluation.

A little more detail

Is this the same as Constitutional AI?

Not exactly. Constitutional AI is a broader recipe guided by written principles. Its original version includes both a critique-and-revision stage and an RLAIF stage.

RLAIF names the use of AI feedback; it is not one fixed set of principles.

Read the research

These guides simplify the methods. The papers describe the full training recipes. Examples on this page are illustrative, not experimental results.