Public preview

AI-vs-AI Critique

Feed one model's output to another and request harsh critique. Disagreements expose weak points.

10 min Intermediate

Goal

Use two different models to audit each other, vs accepting one model's output.

Steps

1

Get output from model A (e.g., Claude). Copy it.

4 more steps locked

Expected Outcome

Immediate

You catch 2-4 errors you'd have published without double-check.

Evidence base

School: Self-Refine / AI Critique

Founders: Madaan et al. · Anthropic (2023)

Madaan et al. 2023 "Self-Refine" proves second-model review raises accuracy 20%+. Anthropic uses similar constitutional AI techniques.

Frequently asked questions

How long does "AI-vs-AI Critique" take?

This exercise takes about 10 minutes, practiced solo in a Creative format.

When will I notice the effect of "AI-vs-AI Critique"?

You catch 2-4 errors you'd have published without double-check. In the short term: You habituate not trusting single-model output, especially factual.

Is "AI-vs-AI Critique" evidence-based?

Yes — it draws on Self-Refine / AI Critique (Madaan et al., Anthropic (2023)). Madaan et al. 2023 "Self-Refine" proves second-model review raises accuracy 20%+. Anthropic uses similar constitutional AI techniques.

Is "AI-vs-AI Critique" 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.

Interactive practice locked · 4 more steps locked

7 complete practices (one per domain) are free forever, no sign-in.