What is actually in the library: 80 courses, 640 lessons

Basis Lab writes a course for you on request, which is the product. The library is the other half: 80 courses already written, 8 lessons each, 640 lessons in total, arranged on ten shelves. It exists so that you can see the output before asking for anything, and so that there is a sensible answer when the thing you want to learn is a thing lots of people want to learn.

Here is what is on the shelves.

The mathematics underneath, 17 courses

The largest shelf, and the one we wrote first. Reading the notation, logarithms and scale, counting without listing, chance and expectation, vectors and distance, matrices as maps, change and accumulation, cross entropy, gradient descent, backpropagation in full, the Fourier transform, Markov chains, systems that evolve, keeping numbers in range, arguing that it always holds, answering with random samples, and why learning works at all.

Nearly every score in the model courses is the log of something, and nearly every claim in them is an expectation. This shelf is where those stop being decoration.

How models work, 16 courses

The simplest model that learns, why a filter reads a picture the way it does, the mathematics of attention, how a model picks its next word, how noise becomes a picture, learning from reward, models that carry a memory, many specialists in one model, trees and committees of them, training from scratch, adapting a model you did not train, where training data comes from, cutting text into pieces, how text becomes numbers, finding structure without labels, and when a model is sure and wrong.

Putting a model into service, 11 courses

Writing what the model reads, finding things by meaning, models that use tools, what a model refuses and why, reading a benchmark honestly, judging a system rather than a model, running an online experiment, watching a model after launch, serving a model to many people, making a model smaller, and what it costs to run one.

The systems around the model, 7 courses

Doing two things at once, keeping the answer nearby, work that waits in a queue, when one machine is not enough, training across many machines, why a query is slow, and running a service that stays up.

Where the data lives, 7 courses

Designing a table to live with, how an index actually works, what the planner decides, what a transaction promises, moving data without losing any, counting things as they arrive, and bytes, text and file formats.

How a program runs, 6 courses

Memory all the way down, what happens while you wait, what makes code fast, turning text into instructions, what two threads can break, and reading a profile honestly.

Making it fast enough, 5 courses

How long it takes, finding it instantly, keeping things in order, answers built from smaller answers, and graphs and finding a path.

The network and the web, 5 courses

From an address to a page, how a browser draws a page, designing an interface to call, how a message stays private, and proving who you are.

Keeping it safe, 4 courses

How a password is stored, keeping a secret in production, thinking like an attacker, and the mistakes behind most breaches.

The physics underneath, 2 courses

Motion and what is conserved, and waves, modes and frequency. The smallest shelf, and the one that will grow next.

What you can read for nothing

Forty-one of the eighty sit on one of five reading tracks, in an order, with a line on each position saying why it is there. The rest are reachable from the shelves.

If what you want is not here, that is the other half of the product: say the goal and your level, and the course is written. The first one is free from beginning to end.