Meet Your AI Friends: Understanding Large Language Models

Chapter 1: Meet Your AI Friends: Understanding Large Language Models

The Amazing World of AI Assistants

Imagine for a moment that you could talk to a computer that actually listens to you. Not just follows simple commands like "play my favorite song" or "set a timer for 10 minutes," but actually understands what you're saying and can have a real conversation with you. Imagine asking it to tell you a story about a dragon who loves pizza, and it creates an entirely new, never-before-heard story right there on the spot, customized just for you. Imagine asking it to help you with your math homework, explaining things in a way that makes sense to you specifically, checking your answers, and then explaining where you went wrong if you made a mistake, all with infinite patience and kindness. This isn't a scene from a science fiction movie anymore. This is the world we live in today, and it's a world that's changing faster than any of us could have imagined just a few years ago.

You've probably already met some of these digital friends without even realizing it. When you ask Siri on your iPhone where the nearest ice cream shop is, or when Alexa tells you what the weather will be like tomorrow, or when you use ChatGPT to help you brainstorm ideas for a school project—these are all examples of something called Artificial Intelligence, or AI for short. AI is becoming such a big part of our daily lives that it's important to understand what it really is, how it works, and most importantly, what it's NOT. Understanding these digital assistants will help you become better at talking to them, getting useful answers from them, and using them as powerful tools for learning and creativity.

But here's something really interesting that many adults don't even fully understand yet: the most common type of AI that regular people interact with today isn't actually as "intelligent" as it might seem. It doesn't think like you do. It doesn't understand concepts the way you understand them. It doesn't have feelings, opinions, or beliefs. And yet, it can do some pretty amazing things that seem almost magical. So what exactly is going on inside these AI systems? How do they process what we say and come up with responses? And why does it matter how we ask them questions? These are exactly the questions we'll explore together in this book. By the time you finish this chapter, you'll have a whole new way of thinking about the AI systems you interact with every day.

What Is AI Really?

Before we dive into the specifics of how these AI systems work, let's take a step back and think about what artificial intelligence actually means. The word "intelligence" is thrown around a lot when people talk about AI, but it's important to understand that when scientists and engineers talk about AI, they don't necessarily mean intelligence in the way we think of human intelligence. Human intelligence is incredibly complex. It involves consciousness, emotions, intuition, creativity, moral reasoning, and so much more. The AI systems we'll be talking about in this book don't have any of that—at least not yet, and probably not in the way you might imagine.

So what are these AI systems really? At their most basic level, they are extremely sophisticated pattern recognition machines. Think about how you learned to recognize a cat. When you were very young, you saw many different cats—big cats, small cats, orange cats, black cats, fluffy cats, sleek cats. Your brain automatically looked at all these different examples and started to notice patterns. All cats have whiskers. Most cats have pointy ears. Cats typically move in a certain way. Cats make meowing sounds. Over time, your brain got so good at recognizing these patterns that now you can look at a completely new cat you've never seen before and immediately know it's a cat, even if it's a weird-looking cat wearing a tiny hat (yes, that's a thing people do with their cats sometimes, for some reason).

Large Language Models, which are the specific type of AI we'll be focusing on in this book, work in a surprisingly similar way. These systems were trained by looking at billions and billions of examples of text—every book ever digitized, every article ever written, every website on the internet, every conversation ever recorded. That's an almost incomprehensible amount of text. Imagine if you tried to read all the text that was used to train one of these AI systems. You would need to read for tens of thousands of years without stopping! And that's just one AI system. There are many of them, each trained on different collections of text.

What these AI systems learned during all that reading wasn't meaning or understanding in the human sense. What they learned was statistical patterns in language. They learned which words tend to appear near other words. They learned the patterns of how sentences are structured. They learned what kinds of responses are typically given to certain questions. They learned the patterns of millions and millions of conversations. And then, when you ask one of these systems a question, it uses everything it learned about language patterns to predict what the most likely helpful response would be. This is a huge simplification, but it's actually quite close to what's really happening under the hood.

Understanding Tokens: The Building Blocks of AI Language

Now that we have a basic understanding of what AI is, let's talk about one of the most important concepts for understanding how these systems process text: tokens. Tokens are essentially the "atoms" that AI systems work with. Just like matter is made up of atoms, text is made up of tokens. But here's where it gets interesting: tokens aren't exactly the same as words.

In many cases, a token is a word. The word "cat" is usually one token. The word "dog" is usually one token. But sometimes a single word can be made up of multiple tokens. For example, the word "unbelievable" might be broken into several smaller pieces, depending on how the specific AI system was designed. This is because sometimes it helps the AI to break complex words into smaller, more fundamental parts. The prefix "un-" means "not," "believe" is its own concept, and "-able" means "can be done." By breaking words into these smaller pieces, the AI can sometimes understand the meaning of words it's never seen before by recognizing the familiar parts.

Numbers can also be interesting. Sometimes "2024" might be one token, and sometimes it might be several. It depends on how the specific AI model was designed. Punctuation marks are often tokens too. The period at the end of a sentence, the comma in a list, even spaces between words—these can all be treated as tokens in some systems.

This is where our pizza analogy comes in really handy! Imagine you have a pizza in front of you. If you wanted to share that pizza with your friends, you'd probably cut it into slices first, right? A pizza with eight slices is easier to share than one giant circular pizza. Each slice is a manageable piece that you can think about, distribute, and eat independently.

Tokens work the same way for AI systems. When you type something into an AI chatbot, the first thing that happens is that your text gets "cut into slices"—or more technically, it gets tokenized. The AI system breaks your text down into individual tokens that it can process. Each token becomes a small piece of data that the AI can analyze, compare with patterns it learned during training, and use to generate a response.

Here's a fun example to make this crystal clear. Let's say you type the sentence "I love eating pizza with my friends." How might an AI tokenize this sentence? It might look something like this:

Token 1: "I" Token 2: "love" Token 3: "eating" Token 4: "pizza" Token 5: "with" Token 6: "my" Token 7: "friends"

In this case, each word is one token. But now imagine you're a really hungry pizza lover and you type: "Yummmmmmmy pineapple pineapple p