AI Makers
Gervigreindarsmiðjan
They build with AI — and know exactly how it works.
What they'll learn
- How language models really work Understand tokens, training and the real limits behind the AI everyone is talking about.
- Build an AI app with guardrails Ship a real AI-powered app with safety limits they design in deliberately.
- Train on Icelandic data Train a model on real Icelandic data and see why the data you choose matters.
- Judge AI like a scientist Evaluate AI output critically and on evidence — not as a fan of the hype.
A working AI application and a trained model — each with its honest accuracy card.
The 13-week journey
-
Sprint 1 — Under the hood
What's actually inside: tokens, training and prediction; the hallucination hunt; prompt patterns that work and why. Ends with a personal prompt playbook plus a documented hallucination report.
-
Sprint 2 — Build with AI
Python against a managed AI API: each student ships a real assistant — a study buddy, an Icelandic-practice partner, a club helper — with a written system prompt and tested guardrails.
-
Sprint 3 — Train your own
Notebooks, pandas-light, and a classifier trained on Icelandic open data — Veðurstofan weather and earthquake feeds are the house favorites. Bias and honesty labs throughout.
-
Week 13 — Showcase
App demos plus a gallery of accuracy cards — the school's public proof that AI education here means substance.
What we cover
Every topic, unit by unit — so you know exactly what your child builds and learns.
Under the hood
- What's actually inside an LLM: tokens, training and prediction
- The hallucination hunt — catching the model confidently wrong
- Prompt patterns that work, and why
- Building a personal prompt playbook + hallucination report
Build with AI
- Calling a managed AI API from Python
- Shipping a real assistant: study buddy, Icelandic-practice partner, club helper
- Writing a system prompt
- Testing guardrails and edge cases
Train your own
- Notebooks and pandas-light for real data
- Loading and charting Icelandic open data (Veðurstofan weather & earthquakes)
- Training and evaluating a classifier
- Bias and honesty labs — reporting limits truthfully
Showcase & accuracy cards
- Writing an honest accuracy card for a model
- Arguing both sides of an AI-ethics question with specifics
- App demos plus a gallery of accuracy cards
A working AI application plus a model trained on Icelandic open data, each with an honest accuracy card.
The flagship of the flagship band, and the first course where students legally hold the tools everyone argues about. Three moves in one term: understand how language models actually work, build a real AI application with them, then train their own model on Icelandic open data — and grade it honestly.
The hooks
Kid hook: “I built an AI study buddy. And I trained a model on real earthquake data.” Parent hook: “They learn what AI actually is — and the honest way to use it for school. I stopped worrying about ChatGPT and homework.”
Who it’s for
One prior course (or a placement chat) so basics don’t slow the build. Thrives: AI Trainers graduates, the philosophically curious, and every teen using AI daily with no idea how it works — which is most of them.
Outcomes — by the end, students can
Explain how LLMs work at a real level (tokens, training, prediction, hallucination); prompt with structure and verify outputs against sources; build an application on an AI API with a designed system prompt and guardrails; load, clean and chart a real dataset in a notebook; train and evaluate a classifier — and report its limits truthfully; argue an AI-ethics question with specifics.
Tools & compliance
Python notebooks (Jupyter/Colab-style on school accounts), school-managed AI API keys with caps, 13+ tool accounts with signed parental consent, Icelandic open datasets, camera-free by default.
Where this course fits
AI Engineer at 16–18 is the direct continuation; pairs with any Developers elective.
Parent questions
Is this teaching them to cheat with AI?
The opposite — it's where they learn the difference between AI as tutor and AI as ghostwriter, with rules they help write.
Do they get real AI accounts?
Yes, at 13+ with your signed consent, on school-managed and capped access.
What makes this different from watching AI videos online?
They build and train — by week 13 they have made things most adults only talk about.
The first lesson is a free trial.
Book a no-commitment trial — pay nothing if it's not a fit.
Book a free trial