AI 101

Lesson 1 · about 6 min

What an LLM actually is

You have heard that you should “use AI.” Before you do, it helps to know what the thing actually is. Most of the confusion about these tools comes from imagining the wrong machine.

It predicts the next chunk of text

A large language model (LLM) is the engine inside tools like ChatGPT. Its one job is to predict the next chunk of text.

You give it some words. It asks itself: given this text so far, what chunk most likely comes next? It adds that chunk, then asks the same question again, and again, until the answer is done. A “chunk” is usually a short piece of a word, so the model builds a reply a fraction of a word at a time.

That is the whole trick. Everything the tool does — answering questions, writing an email, explaining a recipe — is this same act of continuing text.

The patterns come from a lot of human writing

The model learned to predict text by reading an enormous amount of it: books, articles, websites, conversations. Over and over it guessed the next chunk. Then it checked itself against the real text and adjusted.

What it kept from all that reading is patterns — a sense of which words and ideas tend to follow which. It did not keep the text itself. When people say “the model,” they mean this large bundle of compressed patterns.

It is not a database, and it is not thinking

Because it answers in full sentences, it is easy to picture a librarian looking up facts or a person reasoning through your problem. Neither is happening.

Keep this in mind and the mistakes stop being surprising.

Why it can still say something new

If it only keeps patterns, how does it write a sentence no one has written before? Because patterns combine. It can borrow the shape of a cover letter, the topic of your hobby, and the tone you asked for. Then it predicts a path through all three at once. The result is new text, built from old patterns.

You now have the core idea: an LLM continues text using patterns it learned from human writing. Next, you will see what that makes it good and bad at.

Knowledge check

  1. At its core, what is a large language model doing when it answers you?
  2. Where does a model's ability to write come from?
  3. A friend says: "The model just copies sentences it has already seen." What is the best response?