Glossary · Term

policy gradient

← all terms

Definition

Plain language

A reinforcement-learning method that nudges a model's behavior toward actions that scored higher.

As stated in the literature

A family of RL algorithms that estimate gradients of expected return with respect to policy parameters by sampling trajectories.

Also called: policy-gradient

Why it matters: It's the foundation under PPO, GRPO, and nearly every modern RL post-training method for LLMs.

For example, after a successful rollout, the policy is nudged to make the actions it took slightly more likely next time.

Heard on the show

“So you can reward or penalize that action directly with standard policy-gradient reinforcement learning.”
Episode 114 — Agents That Rewrite Their Own Weights Instead of Just Taking Notes

Mentioned in 8 episodes

  1. 114
    Agents That Rewrite Their Own Weights Instead of Just Taking Notes
  2. 090
    How MiniMax-M2 Bets That Sparsity Plus Verifiable Rewards Can Match Frontier Agents
  3. 084
    Terminal Agents Get Free Supervision From The Tokens We've Been Throwing Away
  4. 079
    An Old Idea From Cognitive Psychology Reshapes How We Reward Reasoning Models
  5. 060
    When Splitting One Model Across Three Agents Doubles Its Accuracy
  6. 028
    Teaching a Model to Hire Copies of Itself: Recursive Agent Optimization
  7. 025
    The Missing Gradient Term That Predicts Sycophancy in RLHF
  8. 010
    When Reward Climbs But Reasoning Goes Generic: Diagnosing Template Collapse in Agentic RL

Related concepts

Related terms