What Is a Large Language Model (LLM)?

A large language model (LLM) is an AI system that generates text by predicting the next word. That’s it. Behind the impressive conversations, the creative writing, and the code generation, the core mechanism is surprisingly simple: given some text, the model guesses what word comes next.

How Next-Word Prediction Works

When you type a message to an LLM, the model doesn’t “understand” your question the way a person would. Instead, it looks at the words you’ve given it and calculates which word is most likely to follow — based on patterns it absorbed from enormous amounts of text during training.

You type:  "The capital of France is"
The model predicts: "Paris" (highest probability)
Then predicts the next word after "Paris": "."
Then the next, and the next, until the response is complete.

This happens one word at a time, incredibly fast. The entire response is built through a chain of individual predictions.

What Makes Them “Large”

The “large” in LLM refers to parameters — billions of adjustable values that the model tuned during training. Think of parameters as tiny dials. During training, each dial gets adjusted until the model becomes skilled at predicting what comes next. More parameters generally means the model can capture more nuanced patterns.

Modern LLMs have hundreds of billions of parameters and were trained on massive datasets — books, websites, code, and conversations spanning much of the internet.

Beyond Text

Early LLMs only worked with text, but modern models are multimodal — they can process images, audio, and even video alongside text. The same core idea applies: the model has learned patterns across different types of data.

What LLMs Don’t Do

LLMs don’t search the internet (unless given a tool to do so), don’t have feelings, and don’t “know” things the way humans do. They’re pattern-matching engines — extraordinarily powerful ones, but pattern matchers nonetheless.

Understanding this distinction matters. It explains both why LLMs are so useful and why they sometimes get things confidently wrong — something we’ll explore later in this course. Up next: how these models are actually built.