an open notebook on modern machine learning · entry 000
One idea at a time.
AI writing babies you or buries you. Brain Drip takes one idea at a time, and lets you watch it work.
this page is the demo ↓
lab 01 · tokenization
Watch this page read itself.
A drip is a lesson with its hands on the machine, and this page is the smallest drip on the site. Start where every model starts: it never sees words. It sees tokens.
brain drip takes one idea at a time
lab 02 · prediction
toy bigram model · trained on this page’s own words · runs in your browser
Now the page reads itself. Every word above this line, counted while you scrolled.
One question. On this page, after the word “a”, what usually comes next?
Tallies become weights. Weights become a bet.
It bet right. That is the entire trick, and it is learnable.
brain drip takes one idea at atime
the notebook
The shelves.
56 drips, one idea at a time. Then the rest of the notebook.
56 · interactive explainers
Drips
One idea, one small lab, ten minutes.
latest · Natural Language Autoencoders
open the shelf →
16 · read in order
Courses
Drips arranged into a path.
latest · LLM Concepts
open the shelf →
6 · essays & arguments
Notes
What building this stuff actually teaches.
latest · Four Agent Blueprints You Can Run in Under 80 Lines
open the shelf →
33 · working code
Blueprints
Guides that end with something running.
latest · Run an Open Model with Ollama
open the shelf →
all of it free.
no ads. no sign-in
to read. just paper.
Your turn at the pencil.
The same toy model, your temperature. Low is careful, high is chaos. It writes the last line of this page.
Start reading → 56 drips