Glossary · Term

self-critique

← all terms

Definition

Plain language

Asking the model to look over its own work and flag mistakes.

As stated in the literature

A refinement loop where a model reviews and revises its own output; empirically subtractive, raising precision while leaving omitted mass untouched.

Also called: self-critiquing

Why it matters: Self-review tends only to delete questionable material, so it cleans up mistakes the model made while leaving the items it never thought of still missing.

For example, after writing ten tests the model is asked to review them, and it removes two it now judges to be wrong about what the program should return.

Heard on the show

“So self-critique, cross-model review, adversarial revision — every subtractive loop acts as a directional filter.”
Episode 233 — Why a Model Can Grade an Answer But Not Write the Answer Key

Mentioned in 5 episodes

  1. 233
    Why a Model Can Grade an Answer But Not Write the Answer Key
  2. 229
    One Word Flips a Chatbot From Backbone to Yes-Man
  3. 048
    How a 30B Open Model Reached Olympiad Gold With the Right Recipe
  4. 030
    Why Your AI Agent Won't Stop Working — and Each Model Falls for a Different Trap
  5. 029
    Why Forty-Eight Percent on FrontierMath Isn't the Real Story in DeepMind's New Math Paper

Related terms