You do not need to master active inference in a week. You need to hold it long enough to recognize it when a student surprises you. This is a five-day journal for that.
Day 1 (Monday): read one page slowly
Open Parr, Pezzulo, and Friston, "Active Inference" (2022), and read just the preface plus the opening section of Chapter 1 (Class E). Do not push past it. If a symbol confuses you, circle it and keep going. The claim to hold in mind is simple: a living system acts to keep its predictions close to what actually happens. That is it for today.
If you want a companion for the vocabulary, Themesis offers a short course called T3, Top Ten Terms in Statistical Mechanics for AI. Our honest one-line summary: it is a compact vocabulary primer we recommend for math-hungry learners before the UNI Workshop. It is not a substitute for the book, and we do not speak for the author.
Day 2 (Tuesday): watch students, not screens
Pick one class period. Watch for the moment a student's face changes because what happened was not what they expected. That surprise is what the theory calls prediction error. Note three of them. Do not intervene. You are collecting Class E observations against a live classroom, which is what the reading will start to feel true against.
Confusion is normal on Day 2. Most teachers report that the vocabulary feels like it is describing something they already do, but with heavier words. That reaction is a good sign, not a bad one.
Day 3 (Wednesday): touch the workbench
Log in to the workbench (Class C, configuration checked in your onboarding email). Do not try to run anything ambitious. Open one saved generative model and read the fields. A generative model is a compact story about what a learner is likely to do next, given what they have done. That is the whole idea for today.
If a field name is opaque, hover it. Every field on the workbench has a plain-language tooltip written for teachers, not researchers. If a tooltip is missing, log it: that is a documentation gap we want to close. See The Teachers Workbench Tour for a guided pass through the panels.
Day 4 (Thursday): try one gate
A gate is a small, honest place in a lesson where a learner has to show they understood, not just that they were present. Design one. It should take you fifteen minutes to write and a student under three minutes to attempt. Do not automate it yet. Run it by hand with one small group and watch what the workbench notices when you log the outcomes.
You are not grading with this. You are checking whether your read of the class matches what the model predicted. When they disagree, the model is usually the one that has more to learn. That asymmetry is the point.
Day 5 (Friday): write down what confused you
Take twenty minutes. Write, in your own words, the three things that still do not sit right. Do not fix them yet. This list is the most valuable artifact of the week. Bring it to the next cohort call.
The honest posture we ask you to hold, and that we hold ourselves: UNI is a working hypothesis on an attainable path toward General Natural Intelligence, a natural, active-inference approach whose evidence is growing, evidence-classed, and tested in the open. Do not take the claim on faith. Test the build, inspect the gates, and help us find where it fails.
What to expect in Week Two
The vocabulary settles. The workbench stops feeling like a cockpit and starts feeling like a notebook. You will begin to notice that you are already doing a version of this work, and that the theory is giving you a shared language for it, not replacing your judgment.
Where to go next
- Educator Readiness, the Non-Programmer Path, the sequenced route for teachers who do not code.
- Recommended Reading Order for Teachers, a short, honest reading list with why each piece is on it.
- The Teachers Workbench Tour, a guided walkthrough of the panels you touched on Day 3.
- /workshop if you want the full guided cohort with weekly office hours.