Most people who say they want to learn AI actually want
to understand it first. There is a difference. This
course starts with that distinction and builds from
there.
What the course actually covers
We begin with the mathematical scaffolding that makes
machine learning possible: probability distributions,
linear algebra basics, and how data gets represented
numerically. Week 3 introduces supervised learning
through 4 concrete problem types, each with a dataset
you work through yourself. By week 5, you are evaluating
models using precision, recall, and F1 scores rather
than just accuracy.
The second half shifts toward neural networks. Not the
abstract version, but the specific mechanics: how a
perceptron computes its output, what backpropagation
actually updates, and why learning rate matters more
than most beginners expect. You will train a small image
classifier using roughly 6,000 labeled examples and
watch it fail in instructive ways before it works.
Who tends to get the most from this
Students with some programming experience in Python and
basic algebra move through the material fastest. That
said, 3 of the 8 modules include optional review
sections for people who need to revisit foundational
math. The course is not designed to be easy. It is
designed to be honest about what AI actually
involves.
After completing this course, Frederike Oost wrote: I
finally stopped nodding along to AI articles and started
actually understanding what they were
describing.
Assignments and feedback
Each week includes 1 graded assignment and 2 optional
exercises. Instructors respond to submitted work within
48 hours on weekdays. There are no automated pass/fail
grades on the core assignments, only written
feedback.
