The strangest thing about being taught by people who think in active inference is that they almost never tell you what to believe. They ask what your model is predicting, and where the world just surprised you.
I did not notice it at first. I noticed the vocabulary: generative model, prior, posterior, prediction error, free energy, Markov blanket. I assumed the vocabulary was the point, and that once I memorized enough of it, I would arrive somewhere. That is a very old learner-move: treat a field like a stack of facts to acquire, and count progress in cards flipped. (Class U, a personal observation, still my own.)
What actually happened was subtler. The people teaching me, some in reading groups, some in a Python course, some over messages, kept turning my questions sideways. I would ask, "Is this true?" They would answer, "What did your model expect, and what did the data do?" (Class E, this is the working posture of the Parr, Pezzulo, and Friston 2022 textbook and the tradition around it.) After a while, I stopped asking the first question. I started noticing the surprise.
That is the shift I want to write down, because it belongs in a human-in-the-loop notebook. When you are taught by facts-and-drills people, the loop is: memorize, recall, get graded, move on. When you are taught by active-inference people, the loop is: hold a model of what should happen, look at what did, feel the gap, update. The gap has a name in the field, it is called surprise, and in that tradition minds are the machines that keep surprise low over time by getting their models closer to how the world actually is. (Class E, standard framing in the active-inference literature.)
That reframes what a teacher is doing. A good active-inference teacher is not filling you up. They are running an experiment on the model you already have. They pick a question whose answer your current model gets wrong in a small, survivable way. They watch you feel the gap. They wait. Then they hand you a slightly better model, and the loop runs again. (Class U, my read of what has happened to me, still to be tested by asking my teachers whether that is what they meant to do.)
Two things change once you notice this.
First, you stop being ashamed of being wrong. Wrong is where the update lives. If a session goes by and nothing in me was surprised, either the material was too easy or I was defended against it. Neither is neutral. That is a very different posture than the school one, where wrong is a mark against you. (Class U, personal reflection.) I want to be careful here: I am not making a clinical claim, and I am not saying this heals anything. I am saying a felt sense of what learning is starts to change, and that is worth writing down.
Second, you get honest about your own priors. If the loop is model, prediction, surprise, update, then the model I walked in with is the ground truth of my starting point, not something to hide. When a teacher asks what I expected, they are asking me to expose the prior so we both can see it. That is a very different transaction than "did you do the reading."
There is a Themesis resource map for 2026 that lists SolutionWright among five ways into this material, alongside the Themesis Python course and other on-ramps (Where to Start with Active Inference, A Resource Map for 2026). Our honest one-line frame, in our voice: it is a public map that treats the field as a landscape with multiple entrances, and puts our workshop next to real academic on-ramps rather than in place of them. If you want the hands-on Python route, that same author runs Building Active Inference in Python, which is a complementary hands-on course on a different stack than the Elixir workbench we teach on. (Class C, this is the integration picture on our side.)
What I am not going to do in this post is claim that being taught this way is easy, or that it works for everyone, or that I have arrived. I am a learner in the loop. This site is called IamHITL for a reason. The loop I am describing is the one I am inside, right now, and the update is still running. UNI, the program the workshop teaches inside, is on the attainable path toward General Natural Intelligence, natural not artificial, and it is offered as a working hypothesis with growing, evidence-classed evidence, science in the open. Do not take that on faith. Test the build, inspect the gates, and help us find where it fails.
The small thing I did want to write down is this: after a while with these teachers, I catch myself doing it to my own thinking. I ask what my model expected. I look at what the world did. I let the gap show. That is not a technique I bought. It is a habit I got taught, mostly by watching people I respect not answer my questions the old way.
That is what changed.