A single, continuous book told in plain English. It takes the five technical courses below — on large language models, encoder models, encoder–decoder models, mixture-of-experts models, and diffusion models — and retells their ideas as one story, with the mathematics removed and the intuition left standing.
It reaches for an analogy wherever the technical courses reach for an equation. Nothing here will let you build one of these systems, but by the end you will understand what they are, why they are shaped the way they are, and how the famous names (GPT, BERT, T5, Mixtral, Stable Diffusion) relate to one another.
Audience: curious and capable, no ML background needed. Prerequisites: none — no calculus, linear algebra, or probability assumed.
Read it front to back, like a book. Each chapter assumes the ones before it, and every page links to the next.
Part I — Foundations
| # | Chapter |
|---|---|
| 1 | What These Machines Actually Are |
| 2 | From Words to Numbers |
| 3 | Attention: The Idea That Changed Everything |
Part II — The Family of Language Models
| # | Chapter |
|---|---|
| 4 | The Storytellers: Decoder Models (GPT, Llama, Claude) |
| 5 | The Readers: Encoder Models (BERT and family) |
| 6 | The Translators: Encoder–Decoder Models (T5, BART) |
| 7 | The Committee: Mixture-of-Experts (Switch, Mixtral, DeepSeek) |
Part III — Beyond Language
| # | Chapter |
|---|---|
| 8 | Sculpting Images from Noise: Diffusion Models (Stable Diffusion, DALL·E) |
Part IV — The Lifecycle
| # | Chapter |
|---|---|
| 9 | How a Model Learns |
| 10 | Raising a Helpful Assistant |
| 11 | How a Model Thinks in Real Time |
Closing
| # | Chapter |
|---|---|
| 12 | The Whole Landscape |
Appendices
| A | A Field Guide to Real Models |
| B | Cheat Sheets: How the Architectures Compare |
| C | A Field Guide to Attention Variants |
| D | Telling a Model Where Words Sit |
| E | Compute-Optimal Scaling: The Chinchilla Story |
Reference
| Glossary |