An interactive course · 25 lessons · free
What does the model actually see?
Short, interactive lessons that slow a language model down until you can watch every step: your words becoming tokens, numbers and attention. No maths prerequisites. No hype.
01How it works
One sentence, five steps.
01 · Text
It starts as text.
You type a sentence. To you it’s meaning. To the model it’s a string of characters, and it can’t do maths on characters.
The cat sat on the
02Inside a lesson
Not a video. A machine you can poke.
Every lesson starts from something a chatbot really does, then opens up the part behind it, one short screen at a time.
- A
One idea per screen
A sentence and a picture, nothing else to read. Hard words come after you’ve seen the idea.
- B
Real numbers, honest pictures
Token numbers come from a real tokenizer. Any diagram we drew by hand says so on the screen.
- C
Try it to move on
Screens with a try-it stay locked until you’ve done it: place a word, guess the next one, find the clue.
03Who it’s for
Wherever you’re starting, start here.
Developers
You can read code, but every AI explainer is either a wall of linear algebra or a thread of hype.
See each mechanism on real text, like the actual token numbers, then try it yourself before you move on.
Curious beginners
Everyone talks about AI like you should already understand it. You don’t, and that’s fine.
Start from zero. Every lesson builds intuition before it introduces a single term.
Builders
Your app hallucinates, forgets context or retrieves the wrong chunk, and you don’t know why.
Understand context windows, retrieval and evaluation well enough to debug them.
04Curriculum
From zero to building with it.
25 lessons in three parts. Each one is self-contained, so start wherever your question is.
What is AI?
Intuition first. No maths, no jargon. Just what is actually happening.
How it works
The machinery: tokenizers, vectors, attention and the transformer itself.
Build with it
Ship real systems, and know why they fail before your users do.
In the works
Coming soon.
- —Evaluations at ScaleSoon
- —AI Safety & Red-teamingSoon
- —Production AI SystemsSoon
You can’t trust what you can’t see. So we made the model show its work.
05Pricing
Free.
$0 · 00 / 25 lessons included
- All 25 current lessons, every screen included
- No credit card, no trial clock, no upsell on existing lessons
- Your progress is saved to your account
06Questions
Asked, answered.
Something else? Write to us at hello@crin.ai. It’s a small team, and we read every email.
CRIN is a free, interactive course that shows how AI works. Each lesson is a few short screens with one idea each. It starts from something a chatbot really does, then lets you try the part behind it: cutting text into tokens, placing words on a map of meaning, or guessing the next word.
No. Part I assumes nothing. If you can follow a flowchart, you can follow CRIN. Parts II and III go deeper for developers who want the real mechanics.
A transformer splits text into tokens, turns each token into a vector (an embedding), then uses attention to score how much every token should influence every other. That repeats across many layers before the model predicts the next token. CRIN has a lesson for each stage, so you can understand the parts before the whole.
Tokens are the chunks a model splits text into, often whole words or pieces of words. Embeddings are the lists of numbers each token becomes; similar meanings end up with similar numbers. The tokenization lessons cut real text with a real tokenizer, and the embeddings lessons let you place words on a map of meaning.
Every lesson is short: 6 to 9 screens, about 4 to 5 minutes, one idea per screen. Some screens offer an optional “go deeper” for more detail. Your progress is saved to your account, so you can stop and pick up exactly where you left off.
Yes. Every current lesson is free, with no card and no trial. If a paid tier is ever added for new advanced content, everything available today stays free.
Yes. Lessons are built one screen at a time, so they work on a phone as well as a laptop.