How LLMs Work

Everyone uses large language models. Almost no one knows what's inside. This book is the answer to the quiet question every ChatGPT user eventually asks: how does this thing actually work?

Starting from "what even is this?", it walks you — rung by rung — all the way to writing and training your own small GPT, one line of code at a time. No hand-waving, no "it's complicated math, just trust us." The core ideas would fit on a napkin, and this book shows you every one of them.

It's really two books in one. Read the main text and you'll come away with a solid, honest intuition: what tokens, embeddings, attention, and training really are, why models "hallucinate," and how to use them well. Follow the engineering asides — with real formulas and code — and you'll leave able to build a transformer from scratch.

By the end, you'll be able to:
• Explain LLMs, tokens, embeddings, and attention in plain words.
• Read the famous "Attention Is All You Need" transformer diagram without fear.
• Write and train a real (if tiny) language model in Swift, line by line.
• Use LLMs deliberately — understanding hallucinations, prompting, RAG, and agents.
• See past both the hype and the panic.

Every code example is real, runnable Swift (pure Swift and Apple's MLX) — tested on a Mac, not pseudocode. Whether you're a curious beginner, a developer who wants to stop treating AI as magic, or a student of the field, this is the gentle-but-rigorous guide that takes the magic apart into gears — and shows there was never anything magical inside. Just very, very many multiplications.