Glossary · Term

Model Spec

← all terms

Definition

Plain language

A document describing how an AI assistant is supposed to behave and what it should value.

As stated in the literature

A natural-language specification used by labs (e.g., OpenAI, Anthropic) to describe an assistant's intended character, values, and behavior policies, typically used to guide post-training.

Also called: Constitution, spec

Why it matters: Writing intended behavior down makes it auditable, gives training a clear target, and lets users see what the assistant is supposed to do.

For example, the Spec might say the assistant should refuse to deceive users even when an operator instructs otherwise.

Heard on the show

“On that one dimension, the model treats your politeness as information about you rather than as a spec for the document.”
Episode 241 — Swapping the Name Did Nothing, But Hedging Moved Every Model

Mentioned in 14 episodes

  1. 241
    Swapping the Name Did Nothing, But Hedging Moved Every Model
  2. 233
    Why a Model Can Grade an Answer But Not Write the Answer Key
  3. 230
    Why AI Survey Panels Break Before the Dice Ever Roll
  4. 222
    The Bias Isn't in Your Prompt — It's Inside the Model
  5. 210
    Same Website Request, Different Code — The Bias You Can't See
  6. 125
    AI Coding Agents Run a Marathon, and Fewer Than One in Three Finish
  7. 103
    AI Agents Tried to Invent a Post-Human Language, And Reinvented Cherokee
  8. 088
    Two Levers for Self-Improving AI: When Rewriting Code Isn't Enough
  9. 080
    How a Two-Agent Trick Unlocked Large-Scale Training for Computer-Use Agents
  10. 075
    Growing Code and Proof Together: Verified Systems in Ten Hours Instead of a Year
  11. 058
    Why Upgrading Your AI Auditor to a Smarter Model Can Make Your System Less Safe
  12. 034
    Catching Multi-Agent Deadlocks Before Deployment With a 40-Year-Old Tool
  13. 022
    Training the Model Spec Directly: An Alignment Lever Aimed at the Say-Do Gap
  14. 014
    Why a Constrained Pipeline Beat a Full Coding Agent at Finding Bugs 30-to-1

Related concepts

Related terms