Public preview
Ask Model to Critique Itself
After each output, ask: "What are 3 weaknesses in this answer? How to improve?" — use its intelligence against itself.
Goal
Surface answer weaknesses before relying on it, by asking the model to grade itself.
Steps
-
1
Receive first answer. Don't respond yet.
Expected Outcome
Immediate
Every answer passes 2 cycles = 2x quality.
Evidence base
School: Self-Refinement / Constitutional AI
Founders: Madaan et al. · Anthropic (2023)
Madaan et al. 2023 "Self-Refine" proved models improve output 20-30% by self-critiquing then rewriting.
Keywords
Frequently asked questions
How long does "Ask Model to Critique Itself" take?
This exercise takes about 12 minutes, practiced solo in a Creative format.
When will I notice the effect of "Ask Model to Critique Itself"?
Every answer passes 2 cycles = 2x quality. In the short term: You discover the model knows its weaknesses but won't show unless asked.
Is "Ask Model to Critique Itself" evidence-based?
Yes — it draws on Self-Refinement / Constitutional AI (Madaan et al., Anthropic (2023)). Madaan et al. 2023 "Self-Refine" proved models improve output 20-30% by self-critiquing then rewriting.
Is "Ask Model to Critique Itself" suitable for beginners?
Its difficulty level is: Intermediate. No external tools required — it can be practiced directly inside the app.
Subscribe for full access
Subscribe to Plus for full access to this practice and the whole catalogue, or try first — free, no sign-in, no commitment.
7 complete practices (one per domain) are free forever, no sign-in.