EN: Passt: An Introduction to AI
DE: Passt: KI-Grundlagen
A twelve-week introduction to machine learning for people who will have to live with these systems, argue with them, buy them, and explain them to somebody in a meeting who has not read anything.
Passt is German for "it fits". It is also what you say when something is good enough. A model that fits its data is doing its job, and a model that fits its data too well is lying to you, and the whole course is about telling those two apart.
Live at https://i-yam.github.io/passt/
index.html course homepage, bilingual
modules/ 42 self-contained mini-modules, one folder each
notebooks/ Colab tinker sessions
assets/ shared theme tokens and the i18n contract
Every module is a single self-contained HTML file. No build step, no bundler, no framework. Open one in a browser and it works. The only external dependency is the KaTeX CDN, and only on pages with formulas.
| # | Class | Modules |
|---|---|---|
| 1 | What Is Intelligence? What Is Knowledge? | can_machines_think spiders_and_legs is_ai quiz1 |
| 2 | Learning Has Many Faces | duction anscombe parametric code_comprehension |
| 3 | Linear Regression | hot_or_cold sgd linregtutorial |
| 4 | Local Models and Geometry | glocal neighbours bv dimcurse |
| 5 | Regularisation | regularisation ridge lasso + Colab |
| 6 | Classification Fundamentals | decisions shrooms bayes test6 |
| 7 | Decision Trees | akinator entropy titanic |
| 8 | Ensembles | ensambles boosting |
| 9 | Validation and Model Selection | testtrain folds simpson |
| 10 | Recommendations and Neural Networks | people_like_you boards_of_destiny game_of_life nets |
| 11 | LLMs and Generative AI | gaps apophenic prompts |
| 12 | AI in Practice | persona same_cv magenta_line |
| — | Exam | exam open |
Every page carries a language toggle in the top right. English renders first, German sits behind the toggle, and the button always shows the other language.
The German is not a translation of the English, it is written to be read. Target reader is B2 in either language.
GitHub Pages from the repository root. .nojekyll is present so directories
starting with an underscore are not swallowed. Nothing else is required.
Some modules correct claims that this course itself used to make. dimcurse
carries a note that an earlier version stated a probability that was wrong by a
factor of two. That is deliberate. A course built on falsifiability that never
catches itself is not making the argument it thinks it is. If you find more mistakes, open issues, we'll fix them and the "scars" will make the course even better.
Sections about people who died, were wrongly convicted or wrongly arrested carry
no jokes in either language. bayes on Sally Clark, magenta_line on Air France
447, and titanic and entropy on the sinking are written plain on purpose.
Content CC BY-SA 4.0. Code MIT. See LICENSE.