Google Machine Learning Crash Course

Google rebuilt the Machine Learning Crash Course from scratch in late 2024, and the new version is a substantially different thing from the one people remember. It's now organised as self-contained modules with animated explanations, interactive visualisations and more than a hundred exercise questions, and it added material on large language models and AutoML that the original never had.

The course comes out of Google's own internal ML education work, which shows in the emphasis. There is far more attention to data quality, generalisation and fairness than a typical intro course bothers with, and rather less on the mathematics.

What it covers

• ML models — linear regression with loss and gradient descent, logistic regression, classification with confusion matrices and metrics
• Data — numerical data, categorical data and encoding, feature crosses, datasets, generalisation and overfitting
• Advanced models — neural networks, embeddings, an introduction to large language models
• Real-world ML — production systems, AutoML, and ML fairness

The certificate question

Google is clear that there's no formal certification for the Crash Course. What you get is a badge per module, earned by scoring at least 80% — four out of five — on that module's end-of-quiz. That's fine if you're learning for yourself and worth knowing if you were planning to put something on LinkedIn. The Google Cloud professional ML certifications are an entirely separate, paid track.

Who it's for

Engineers and analysts who need working intuition about how models behave and where they fail, without a semester of linear algebra. The fairness and production modules are the ones you won't easily find elsewhere free, and they're the ones that matter once a model touches real users. Skip it if you want to build models properly — the mathematical grounding is deliberately thin and you'll need Andrew Ng's material or a textbook. Skip it too if you can't write Python or read a NumPy array; the exercises assume both.

Our take

The best free conceptual introduction to ML there is, with the caveat that conceptual is exactly what it stays. The interactive visualisations genuinely help — watching a decision boundary move as you change a hyperparameter beats reading about it — but the exercises are quiz-shaped rather than project-shaped, and you finish able to discuss models rather than train one on your own data. Pair it with a real dataset and a notebook or it won't stick.