steva.studio

A book & a workshop for curious minds

Students don't learn to use AI. They build it.

Over eight sessions, a class of 9-to-13-year-olds builds a working language model from scratch — a crawler, a filter, a tokenizer, a word map, a predictor — and on the final day, they talk to a machine they assembled themselves.

Why this, why now

Passengers,
or builders?

The children in your classroom will grow up in an AI-saturated world. The question is whether they enter it as passengers — typing prompts into systems they treat as magic — or as people who understand, at a foundational level, what these machines actually are.

The AI Lab takes the second route. It strips the mystery out of language models by having students build one.

9–13
ages · IB PYP & MYP
8
sessions, 75 min each
1–5
students per lab

Two ways in

One idea, built over two complementary experiences.

The book explains how the machine works. The workshop lets a class build one. 

The Workshop

The AI Lab

Small teams gather text, choose what deserves to stay, watch language become numbers, train an actual model, teach it to behave, then test it and trace its answers.

  • Students make decisions and learn through inquiry
  • Not a competition — no leaderboard, no score, no "best" model
  • Built for the IB: a PYP unit of inquiry or an MYP unit, with its own key concepts and ATL skills
  • Runs entirely online — no software to download, secure access
The Book

Building the Machine

How large language models work. Explained curious minds

It began with one question from a ten-year-old — "But how does it know all this stuff?" — and a promise to answer it properly.

  • A real story, not a textbook
  • Every fact verified, every concept accurate
  • Written for kids, genuinely useful for parents
  • The perfect complementary reading after the workshop

The deck · 8 sessions

Every session is a card. Front asks the question. Back tells the truth.

Act I Feed the machine
Act II Turn language into math
Act III Shape & question it

The truth contract

Six promises we don't break.

01

Real, not a slideshow

Press a button and a genuine training job runs on a cloud GPU; the strange first sentences are the model's actual output.

02

Not a competition

No leaderboard, no score, no best model; different choices produce different models; the variety is the point.

03

No make-believe about the machine

Never "the model knew / wanted / woke up"; say what mechanically happened: the algorithm merged the most frequent pair; its settings changed when the prediction missed.

04

Where abilities come from

A model's behavior is shaped by the text it read, how big it is, how long it trained, and how people shaped it after.

05

The lab is the unit — not the child

No student accounts, no names on inputs, no live internet for students; a team joins with a one-time code, and the text and notes students create stay on the school's own devices.

06

An ever-present "Why?"

Every screen lets a kid ask why this step matters; formal terms wait behind a "Why?" layer, plain language first.

In your classroom

Designed around the IB — not mapped onto it afterward.

Built for PYP and MYP: students drive the inquiry, and every session ties to key concepts, ATL skills, and the learner profile.

  • Students drive inquiry through experimentation and failure — not lectures.
  • Every team chooses their own topic — a unique NanoChat each.
  • Transdisciplinary by nature: language, mathematics, science, and ethics.
  • Builds thinking, communication, and research ATL skills.

How it runs

8×75

Eight 75-minute sessions, ideally on consecutive school days.

1:3

One device per group of two or three students. Browser-based — no installs, no student accounts.

+1

A steva facilitator leads each session; the class teacher co-pilots — no technical prep.

ABC

Prerequisites for students: basic reading and curiosity. That's it.

What students take away

By the end, every child can explain — in their own words:

01where a model's words come from, and who chooses them
02how language becomes numbers a machine can work with
03why the same model can give different answers
04what "training" really changed — and what it didn't

The book

Building the Machine

How Large Language Models Work, Explained for Curious Minds.

Coming soon
buthowdoesitknowallthatstuff

Building the Machine


steva
1
Feeding the Machine
2
Speaking the Machine's Language
3
Building the Brain
4
Scaling to Reality
5
Teaching Manners & Measuring Minds
6
The Machine That Remembers

"But how does it know all this stuff? How does it understand what I'm asking?"

Coming soon

Be first to read it

Book a pilot

Bring the AI Lab to your class

Tell us a little about you — whether you're an IB coordinator or teacher planning a pilot, a school leader planning ahead, or a parent who wants the book.

We’ve suggested these from your approximate network location. Please correct them if needed.

We'll only use this to reply. No spam, no sharing.