Glossary · Term

Q-value

← all terms

Definition

Plain language

How well you expect things to turn out if you take one specific action right now and then play normally.

As stated in the literature

In RL, the expected return of taking a given action in a given state and following the policy thereafter; the advantage function equals the Q-value minus the state value.

Also called: Q-values

Why it matters: It lets a system compare specific actions by their expected long-term payoff, which is the foundation for choosing good moves and computing advantage.

For example, the Q-value of moving your chess knight to a certain square is how well you'd expect the game to go if you make that move and then keep playing as usual.

Heard on the show

“Eric, walk through the Q-value framing, because that's the move that makes this more than just "write a plan.”
Episode 183 — Why You Can't Fine-Tune Foresight Into an AI Agent

Mentioned in 2 episodes

  1. 183
    Why You Can't Fine-Tune Foresight Into an AI Agent
  2. 173
    The Free Step-Level Grader Hiding in Every RL Training Run

Related terms