Build something first

You cannot hear a weak answer from a learner until you have given one yourself.

Every session in this curriculum asks learners one question more than any other: how would you know it came back wrong?

You can ask that from a script. What you cannot do from a script is hear the answer. "I'll look at it and see if it's good" sounds like an answer. It passes if you have never had to give one, and it fails on contact if you have.

That is the whole reason this half exists.

What this is not

It is not a request that you learn to code. Conflating those two is what keeps most educators out of this entirely.

You do not need to read what the machine writes. You need to have decided what a thing was for, watched it come back wrong, and worked out what to change about your own request. That is a different skill and you already have most of it.

Two builds, in order

Build One is calibration. Something small and personal, about a subject where you are the expert and the model is not. Nothing rides on it. You are finding out what it feels like to know enough to judge what comes back.

Build Two is a real thing for your actual practice, made for somebody else to use. It is harder in a way that has almost nothing to do with the building.

The order matters more than it looks. Build One is the only time your knowledge column starts full, because the subject is yours. That is the sensation you are trying to bank. Start at Build Two and you have nothing to compare against when the model hands you something plausible about a subject you half know.

What happens if you skip this

You run a good session. Learners build things. Somebody says they would check their work by looking at it, and you say great, and the session moves on.

From the back of the room that looks like it went well.

How long

Build One is an afternoon. A project that takes a month mostly teaches you about the month.

Build Two takes longer, because most of it is not building. It is the Anchor card, and most people find that harder than the build.

Start here

Build One takes an afternoon. It exists to show you what a wrong answer feels like while you are still the person best placed to spot one.

  1. Build One

    Make one small thing about a subject you know and a model does not, then hand it to somebody and say nothing.

    You leave with

    • One thing that runs, with a link you can send
    • One sentence about what you had to change in your own request
  2. Build Two

    Make something for your actual practice, and answer whether it should exist before you make it.

    You leave with

    • A filled Anchor card for one real project
    • A STEER check run against the thing before anyone else uses it
    • A decision about whether to build it, which may be no