Preface

Hello. I'm glad you found this book.

I have been writing books and program code for more than twenty years. Within the last twelve months, AI has become good enough to do both at a fairly decent level. So my most important skills have already been more or less replaced. At first, that was a strange feeling. It took me quite a while to make my peace with it.

Today I see it differently. AI doesn't replace my skills, it complements them. I still plan, research, write, and program the way I did before, just much faster. Projects that used to be impossible for a single person can now reach a very good state in a fairly short time. Tasks that used to be beyond my abilities are now something I can take on. Of course, that brings new challenges. When we can do practically anything, focus and prioritization matter all the more. And keeping up with the speed of these AI systems is exhausting enough on its own.

The technology really became tangible for me on the evening I loaded a freely available AI onto my own computer. What sat on my hard drive was, at its core, one large file full of numbers. Next to it, a small program that reads those numbers and computes with them. That was all. Without an internet connection or any outside help, I could ask these “numbers” very personal questions, and the answers were surprisingly good and sensible. That naturally raised a whole series of questions about how this works. And out of those questions, in the end, came the book you now have in front of you.

But I also noticed how easy it is to misjudge what an AI can do. A convincing answer can be correct, incomplete, or freely invented. An AI can deliver good results on a hard task and fail at a simple detail shortly after.

That's why this book is about what happens beneath the surface. You'll learn how text is broken into tokens, how sequences of numbers represent meaning, how information is connected and stored, how you can compute with it, and how an AI is trained. After that, we look at how you can use AI sensibly, check its answers, and connect it to tools. Finally, we turn to data centers, privacy, freely available AI, work, education, safety, and where things are heading.

I wrote this book for everyone who feels the way I did. For those who want to dive deeper and really understand what is happening right now. What possibilities are opening up, and where the limits are. You don't need prior knowledge, programming experience, or a computer science degree. All you need to bring is a little time and curiosity.

Where a technical term is needed, I explain it first. The details go as far as you need them to understand. Formulas are not the focus. It's the connections that matter.

The focus is on large language models, or LLMs for short. They are the foundation of applications like ChatGPT, Claude, and Gemini. In everyday language, the broader term AI is often used for all of it. In this book, language model refers to the technical class of model. I speak of an AI system when model, data, tools, interface, and controls work together. We also look at other kinds of models, so this distinction stays clear.

The book has four parts.

Part I: Seeing the system. Four chapters. We start with a very concrete encounter, then go through the history of AI and the different kinds of models, and at the end comes the map. What the model is, what the application around it is, where the data, the tools, and the control sit. This part is the foundation for everything else.

Part II: Inside the language model. Eight chapters that follow the text through the machine. Tokens, embeddings, attention, training, reasoning, hallucinations, and the question of what actually happens inside the model. This is the technical core. Once you're through this part, you know what happens the moment you type a question and get an answer.

Part III: From model to reliable system. Six chapters where we move from theory to practice. Bringing in outside knowledge, phrasing tasks clearly, deciding when AI helps and when it hurts, measuring quality instead of guessing, agents with tools, and the path from a first attempt to an operation you can rely on.

Part IV: Infrastructure, power, and consequences. Four chapters that zoom out of the model. Chips and energy, data and open models, the consequences for work and education, and an outlook. These chapters stand more on their own. You can read them in whatever order interests you.

At the end, the conclusion brings everything back together. For your first pass, I still recommend the order of the book. I explain many terms in full only once and keep using them afterward.

Some names and numbers in the book capture the current state and important historical steps. They will change. The underlying mechanisms, and the questions that let you judge new developments, age more slowly. That is exactly where the focus lies.

This is not a classic textbook either. It is more a collection and structuring of my thoughts as I work through the subject and use AI to automate business processes in my own company and in others. The idea behind this book is that it might help you too, or that we learn together.

If you find an unclear passage or a mistake, you can report it at any time through the website at ai2718.com.

And now I wish you lots of fun reading,

Thomas