Naveo

STEP 11 / 24

A7 A/B

MCQ · NO COST

You asked the model for a plan to reduce oxygen consumption during an emergency. It returned a 4-step plan that sounds good. Before executing, you want to stress-test it. Which follow-up brings you real pushback?

Why?. optional

Look for: closed contract, explicit fallback, scaffold at the end.

GUEST MODE

You're viewing this lesson as a guest. To save your progress, earn XP, and keep your streak, sign in when you're ready to check.

The model is never "sure" of anything useful

If you ask the model "are you sure?" it'll answer one of two things: "yes" or "let me check". Both are noise. The model doesn't have an internal certainty thermostat. It's predicting the next token.

What does work: ask it to adopt an adversarial stance and critique your plan from there.

The devil's advocate pattern

code

"Act as <specific skeptical role>. List 3 concrete reasons this
could fail, and for each one tell me what data you'd need to see
to rule the risk out."

Three elements:

Role (not generic). "Skeptical senior engineer", "security auditor", "customer who has to sign the contract".
Fixed number ("3"). Without a number, it brings 0 or 15.
Associated test ("what data to verify"). Turns each criticism into an action.

Why raise the rigor before executing

There are decisions where the cost of being wrong is high: a migration, a reply to an important customer, a production change. In those cases, an extra devil's advocate turn costs little and prevents a lot.

Variant with humans: the same question works in a meeting. "Before we decide, someone play devil's advocate and tell us 3 reasons this could go wrong." People good at this raise the quality of any group decision.

When NOT to use it

For throwaway or low-risk outputs (an internal email, a quick summary), devil's advocate is overkill. You pay the cost on important decisions, not on every turn.

On the right, two ways to stress-test a model's plan. Which one brings actionable critique?