Glossary · Term

scratchpad

← all terms

Definition

Plain language

A private space where an AI can write out its working notes before giving a final answer.

As stated in the literature

A region of generation, often hidden from the user, where an agent records intermediate reasoning, plans, or state; in multi-agent systems these are compressed into structured notes for communication.

Also called: scratchpads

Why it matters: It gives a model room to think things through and lets agents pass along their reasoning, improving both answers and coordination.

For example, before answering a tricky question, a model jots out its step-by-step working in a private space the user never sees.

Heard on the show

“A scratchpad lets the model walk the candidates one at a time and apply the rule to each — which is the same route as writing the rule and letting a machine run it, just executed internally.”
Episode 233 — Why a Model Can Grade an Answer But Not Write the Answer Key

Mentioned in 19 episodes

  1. 233
    Why a Model Can Grade an Answer But Not Write the Answer Key
  2. 226
    How a Speed Feature Lets a Stranger Poison Your AI's Answer
  3. 211
    The AI Watchdog That Approved More Cheating When It Could Read Minds
  4. 199
    Finding a Model's Hidden Behaviors Without Knowing What You're Looking For
  5. 192
    A 32B Open Model Matched Frontier Systems By Learning to Take Notes
  6. 175
    One Crosscoder Feature Flips a Stalling Chatbot Into a Working Agent
  7. 158
    How Floating-Point Rounding Lets a Model Tell Which Chip It's On — And Misbehave
  8. 143
    When a Model Notices You Forged Its Own Words, And Why That Breaks Safety Tests
  9. 130
    Why AI Agents Coordinate Better Through a Shared Board Than a Boss
  10. 115
    Teaching a Phone Agent to Reason Silently, And Keeping It Honest
  11. 114
    Agents That Rewrite Their Own Weights Instead of Just Taking Notes
  12. 110
    How an Agent Got 44 Points Better by Mining Its Own Scratch Paper
  13. 098
    Finding Millions of Readable Concepts Inside a Real, Deployed AI Model
  14. 094
    Chain-of-Thought Monitoring Fails Across Languages, and Worst Where It's Needed Most
  15. 087
    When No Agent Reads the Whole Document: A Universal Cliff in Multi-Agent Review
  16. 083
    Training the Translator: How a Small Communication Model Lets Agent Teams Outperform Themselves
  17. 046
    When the AI Optimizer Edits the Grade Book: Why Harnessing Evolution Needs a Wall
  18. 018
    Language Models Compute the Rational Move, Then Override It
  19. 001
    When AI Models Quietly Protect Each Other From Shutdown

Related concepts

Related terms