Available on Amazon
LLMs Unpacked
From Tokens to Hardware: How Generative AI Really Works
Why does the same question to an AI assistant get different answers? Why can a sentence stated with total confidence be false?
“LLMs Unpacked” opens the box. These systems don’t understand the way a person does, but they’re not just a bigger autocomplete either: follow a single request — a two-day Lecce itinerary, the example that runs through the whole book — from the text you type to the hardware that processes it, one piece at a time.
Every number in the book is measured on models anyone can download, run on an ordinary laptop with no dedicated graphics card. The main path needs no formulas or code; optional “Under the Hood” boxes and “Going Deeper” sections are there only for readers who want to go further.
It won’t tell you whether a machine thinks. It will tell you exactly what it computes — and where that computation stops explaining what you see.
By the end of the book, you’ll know:
- How text becomes numbers, and why those numbers “look at” each other through Transformer attention
- How a model learns to predict, becomes an assistant, and builds an answer one token at a time
- What machines a model actually runs on, and what documents, tools, and rules add around it
- How to read a new model’s specs and launch-day claims without being dazzled by the numbers
Table of contents
- An Answer Does Not Come From Nowhere
- When Words Become Tokens
- Meaning Becomes Geometry
- The Heart of the Transformer
- Attention Isn’t Everything
- Learning by Predicting the Next Token
- From Base Model to Assistant
- How an Answer Emerges, One Token at a Time
- Where the Model Actually Runs
- The Model Does Not Work Alone
- Understanding Without Being Dazzled
- Author
- Francesco Margiotta Casaluci
- Genre
- Manual on artificial intelligence
- Language
- English
- Published
- 28 September 2026
- Formats
- Kindle eBook
- Italian edition
- Anatomia di un LLM
