Interactive guides

Deep learning, visually.

Guides that let you move the sliders, break the model and watch the math come alive. Start with a single neuron and work up to Stable Diffusion.

Work in progress. Opening soon.

Try it now · a neuron, live

Put every dot in its own color

This neuron votes on which side of a line each dot sits. Move the three sliders until all twelve dots match their background.

weight w₁ weight w₂ bias b

Dots on the right side

How it works

Drag a weight, shrink a step, add noise. The picture changes right away, so the idea sticks before the formula appears.

Each demo comes with short guided steps that say what to try and what to look for. Nothing to figure out alone.

Linear algebra and calculus built up with worked numbers, for people who know some math and learn best by seeing it.

What is coming

1
Neurons and backpropagation
A neuron is a weighted vote with a bend. See a network learn by rolling downhill.
2
Convolutional networks
Filters learned from data, and pretrained models worth reusing.
3
Detection and segmentation
Boxes, masks, and the numbers that tell you whether they are any good.
4
Attention and transformers
How a model looks things up, and how GPT writes one token at a time.
5
Vision transformers and speech
Images as patches, and sound as a picture a model can read.
6
Diffusion models
Generate images by learning to undo noise, all the way to Stable Diffusion.
7
Projects
Plan, measure and present a project of your own.