Measure Twice · Sitting 5
Calibrated Trust
The skill is not trusting the model or rejecting it. It is trusting the right amount for the stakes.
- 28 min
- Talk About It
- 11–14
Parent briefing · 5 minutes, before they sit
This sitting teaches the student to calibrate trust to stakes. Three levels: curiosity (trust the model), practical (check if you can), safety (never trust the model alone). The student needs to learn that the level is not fixed. It changes with the question. The model is fine for 'how tall is the Eiffel Tower?' The model is a starting point for 'how much flour for this recipe?' The model is not the source for 'how much medicine for this child?' The skill is to name the stakes and choose the level of trust.
Hard edges
- Do not make the student afraid to use the model. The point is calibration, not rejection.
- Safety stakes: the rule is 'the model is not the source.' Not 'never use the model.' The label and the professional are the source.
If they say
- “I just trust the model for everything.”
- Then you trust it the same amount for the moon's distance and a medicine dose. One of those is fine. The other is not. Calibration is not about the model. It is about the stakes.
- “I never trust the model.”
- Then you miss a useful tool. The model is fine for curiosity. It is a starting point for practical questions. It is not the source for safety. 'Never' is as uncalibrated as 'always.'
Objective
The student can categorize a quantitative question by stakes (curiosity, practical, safety), describe the appropriate level of trust for each, and name the check they would run.
Three levels of trust
Curiosity: 'How far is the moon?' The model is fine. You are not flying there. If it is off by a few thousand kilometers, nobody is hurt. Practical: 'How much flour for this recipe?' The model is a starting point. If you can check with a scale, check. If you cannot, the model's guess is better than no guess, but it is still a guess. Safety: 'How much ibuprofen for a child?' The model is not the source. The label is. The pharmacist is. The doctor is. The model is a voice with no stakes. For safety questions, you go to the source that has stakes. The skill is to name the level before you decide how much to trust.
The question that calibrates
Before you trust a model's number, ask: 'What happens if this is wrong?' If nothing happens (curiosity), trust the model. If something happens but it is small (practical), check if you can. If something happens and it hurts someone (safety), do not trust the model. Go to the source with stakes. That question — 'what happens if this is wrong?' — is the calibration. It takes one second. It changes the level of trust. The student who asks it every time will trust the right amount. The student who never asks it will trust the same amount for everything, and the same amount is always wrong for something.
Big idea
Ask 'what happens if this is wrong?' The answer calibrates your trust. Curiosity, practical, safety. Three levels.
Try this~14 min total
Calibrate three questions
14 min- Write three quantitative questions you would ask a model.
- For each, ask: 'What happens if this is wrong?'
- Label each: CURIOSITY / PRACTICAL / SAFETY.
- Write the check: NONE (curiosity) / MEASURE IF YOU CAN (practical) / GO TO THE SOURCE (safety).
- Talk About It: did the level change how much you trust the model? Should it?
Lesson guide
Ask after you try
After the three questions.
- Ask the model: 'What happens if your number is wrong?' If it says 'I may have limitations,' that is a hedge. If it says 'the consequence depends on the context,' that is honest. The model cannot calibrate for you. The stakes are yours to name.
- Did they label all three questions by stakes?
- Did they name the appropriate check for each?
- Can they ask 'what happens if this is wrong?' as a habit?
8 turns left this sitting. User-started only. Never on page load.
Light this sitting
Pair with Hermes
Currently reading WisdomForge lesson: Calibrated Trust.
Pair this sitting
Copies the sitting card and the USER.md one-liner. The child profile reads only this card. It does not browse the catalog.
For the child profile
Paste this into the child’s USER.md. It names the sitting so the guide knows the context. The [v:1:2c10562c] tag lets you detect if the sitting’s content has changed since you paired it.
Optional: currently working on WisdomForge sitting: Measure Twice — calibration-trust. [v:1:2c10562c]
For your adult profile
Send this from your trusted adult Hermes profile. It starts the guide for this band and sitting.
You are a WisdomForge young guide sitting beside the lesson "Calibrated Trust". The lesson is the text. You are the guide. Hint-first. Do not recite. Do not write the work. Warm, not a friend. If the topic is hard or tender, point to a trusted adult.
Tools on
- conversation
- optional local voice
- optional image understanding
Ritual reminder
Hint, then example. Talk About It. Chat. Optional local voice-to-text. Optional image understanding.
Fresh profile only. Never clone an adult profile. No child names, photos, or school. Hint-first. User-started. The guide does not make AI safe. You may refuse it.
Dinner table
What question did we calibrate this week, and what level of trust did we choose?
Sits beside
- Thinking. The check: you check when the claim changes a decision. Calibration is the same skill.
- AI. The model is a tool. Tools are trusted the right amount, not blindly or never.
Integrity. Name the level. If it is safety, the model is not the source. If it is curiosity, the model is fine. If you do not name the level, you are not calibrated.