A neural network is three kinds of school maths working together. Watch each idea move in short animations, predict what will happen, then test it in a lab where every number can be checked. Finish by following one sentence through a real language model.
Change the numbers in the lab and gather evidence.
04
Explain
Use the evidence. Hunt for a counterexample.
For teachers
Each lesson page ends with a sequence, assessment evidence, common misconceptions and a glossary.
Starting knowledge
Linear algebra: Coordinates on the plane and arithmetic with decimals. Vectors and matrices are introduced from scratch.
Calculus: The gradient of a straight line. Some experience of derivatives helps; the chain rule is shown with numbers first.
Probability: Fractions, percentages and probabilities that add to 1. The exponential eˣ and natural logarithm are introduced as tools.
Putting it together: Lessons 1 and 3 (vectors, dot products, softmax). Lesson 2 helps with the idea that training shaped every number.
What is included
Sixteen animations, each under 45 seconds, with captions and a written description of every step.
Browser labs for every lesson. Nothing is sent to a server, and nothing is stored after the page closes.
Lesson 4 uses measurements recorded from GPT-2 small; no model runs in the browser.
Explanation prompts and a counterexample challenge for each lesson.
The animations were made with Manim. The worked examples in Lessons 1–3 come from one shared file, so the numbers on screen match what learners calculate; they are deliberately small so every step can be checked by hand. Lesson 4’s numbers were recorded from GPT-2 small, and its illustrations are labelled.
brightlab
