Glossary · Term

world model

← all terms

Definition

Plain language

An AI's internal sense of how its environment works — what its actions do and how the world responds.

As stated in the literature

A model's learned representation of environment dynamics enabling prediction of next states; in ECHO, training a terminal agent to predict environment tokens is argued to instill a transferable world model, evidenced by cross-model prediction gains.

Also called: world models

Why it matters: An internal sense of how the world responds lets an AI plan ahead and adapt rather than blindly reacting step by step.

For example, an agent navigating a kitchen has one if it can predict that opening the fridge will reveal what's inside before it even acts.

Heard on the show

“And the deeper claim isn't about Hanoi at all — it's that a world model isn't a property a system has or doesn't have.”
Episode 237 — The Model Built a Perfect Map of the Puzzle, Then Lost It

Mentioned in 11 episodes

  1. 237
    The Model Built a Perfect Map of the Puzzle, Then Lost It
  2. 234
    Two Copies of Gemini Cooperated in a Game Where Betrayal Always Pays
  3. 213
    A Model Learned to Control a Robot by Watching Video It Never Acted On
  4. 183
    Why You Can't Fine-Tune Foresight Into an AI Agent
  5. 182
    How a Tiny Model Too Weak to Plan Cuts a Bigger Agent's Hallucinations by 80%
  6. 167
    How Teaching an AI to Predict, Not Act, Made It a Better Actor
  7. 163
    Why Training Only on Perfect Solutions Cripples a Model's Reasoning
  8. 115
    Teaching a Phone Agent to Reason Silently, And Keeping It Honest
  9. 113
    What If a Prompt Injection Never Left? Attacks That Wait in Agent Memory
  10. 084
    Terminal Agents Get Free Supervision From The Tokens We've Been Throwing Away
  11. 052
    An Old Reinforcement Learning Tradeoff Sneaks Back Into LLM Agents

Related concepts

Related terms