If you teach and you actually want the math under active inference, you need a math-prep track. Here is one that is honest about the work.
Who this note is for
Not every teacher needs the equations behind free-energy minimization. Many educators do fine with the conceptual layer: generative models as expectations, KL divergence as a mismatch score, gates that reward showing your work. If that is you, skip this post and read KL divergence for teachers who do not love math.
But some of you are the math-hungry ones. You want the notation. You want to know what a partition function is, why entropy shows up next to information, and how statistical mechanics connects to the Bayesian machinery Karl Friston and colleagues build on (Class E). If that is you, keep reading.
The recommendation
Themesis, run by Alianna J. Maren, offers a course titled T3, Top Ten Terms in Statistical Mechanics for AI. It is a paid, self-paced course on Thinkific.
Link: T3, Top Ten Terms in Statistical Mechanics for AI
Our honest one-line frame: we recommend T3 as prep for the UNI Workshop for learners who want the statistical mechanics vocabulary before they meet active inference in a classroom setting.
We are not repeating her syllabus here. Go read her course page in her own words. What we are doing is telling you where T3 fits in a teacher's preparation path (Class C, integration note).
What T3 covers, in plain summary
Ten terms from statistical mechanics that recur in the machine learning and active inference literature. Things like energy, entropy, free energy, partition function, and the ideas that link them. Foundational vocabulary, not applied AI recipes.
Why does that matter for a teacher who wants to teach UNI honestly? Because the phrase "free-energy minimization" is not a metaphor. It has a specific mathematical shape, and that shape borrows from physics. If you want to answer a curious student's question about why the word "energy" shows up in a theory about brains and learning, T3 gives you the vocabulary to answer without hand-waving (Class E, expert citation to the Themesis materials).
Honest about the effort
This is a math course. Not a math-appreciation course. You will see equations. You will need to sit with them.
If you have not touched calculus or probability in a decade, budget real time. Two evenings a week, minimum, for several weeks. Have paper and a pencil ready. Do not try to skim it on a phone between classes.
If you finish T3 and want more, the next honest step is not another course. It is opening the Parr, Pezzulo, and Friston (2022) active inference textbook and reading the early chapters slowly (Class E). T3 gives you the vocabulary. The textbook is where the vocabulary gets used.
Where this fits in the readiness path
For most educators, the sequence looks like this:
- Conceptual grounding (our blog cluster, no math required).
- If math-hungry, T3 for statistical mechanics vocabulary.
- Optional: chapters 1 to 3 of Parr, Pezzulo, and Friston (2022).
- UNI Workshop, where the concepts get applied to classroom gate design.
You do not need step 2 or step 3 to attend the workshop. You need them if you want to teach the deeper layer with your own confidence intact.
For a broader map of what Themesis offers and how it slots into a teacher's plan, see Themesis resource map, a teacher's note. For the wider reading list, see Recommended reading order for teachers.
What we are not claiming
We are not claiming T3 is a UNI course. It is not. It is a Themesis course about statistical mechanics vocabulary.
We are not claiming T3 will make you an active-inference practitioner. It will not. It will make you literate in a set of terms that active inference borrows.
We are not claiming any partnership or affiliation. This is a recommendation based on reading the course description and knowing the terrain (Class E). If you take it, you are Alianna Maren's customer, not ours.
Next steps
- Read Themesis resource map, a teacher's note for the full picture of Themesis materials.
- Read Recommended reading order for teachers to place T3 in your reading plan.
- If you want the conceptual layer without the equations, start with KL divergence for teachers who do not love math.
- Ready to apply this to classroom practice? See the UNI Workshop.