Glossary · Term

stochastic environment

← all terms

Definition

Plain language

A setting where, after you act, something random and outside your control also happens.

As stated in the literature

An environment with non-deterministic transitions where the next state isn't fully determined by the agent's action; the leftover value terms it introduces break the telescoping reward-recovery and motivate advantage-based signals.

Also called: stochastic, deterministic environment

Why it matters: Randomness in outcomes makes it hard to tell whether a result came from a good decision or luck, which is exactly why some training methods focus on advantage rather than raw reward.

For example, in a card game, after you choose to draw a card, which card you actually get is random and beyond your control.

Heard on the show

“My instinct is, forget the privileged extras, because a model is stochastic, and its own averaged weights aren't quite the same as its current weights.”
Episode 242 — Making a Vision Model Better by Showing It Blurry Images

Mentioned in 13 episodes

  1. 242
    Making a Vision Model Better by Showing It Blurry Images
  2. 199
    Finding a Model's Hidden Behaviors Without Knowing What You're Looking For
  3. 191
    How One Researcher Beat GPT-5.2 and Gemini 3 by Judging Their Answers, Not Improving Them
  4. 173
    The Free Step-Level Grader Hiding in Every RL Training Run
  5. 133
    How MiniMax Turned a Reward-Hacking Disaster Into Olympiad Gold
  6. 122
    When Your Coding Agent Lies About the Fix: Verifying the Plan Before the Model Runs
  7. 120
    How an AI Agent Rewrites Its Own Tools, Without an Answer Key
  8. 096
    How Treating an AI Agent's Execution Like Git Recovers a Coordination Penalty
  9. 095
    Seven Wins to Zero: How Organizing AI Agents Like a Lab Changes the Search
  10. 076
    Same Model, Organized Differently: How an Agent Architecture Beat Frontier Systems at Research Math
  11. 070
    When Models Know the Answer But Say the Wrong Thing Anyway
  12. 027
    When AI Agents Build the Serving Stack: A Bet on Bespoke Infrastructure
  13. 007
    Exploration Hacking: When Models Sabotage Their Own RL Training

Related terms