What is a Large Language Model?

The name, decoded

Large Language Model (LLM), three words, three clues.

• Large: it has billions of tiny knobs (called parameters) inside it.
• Language: it works with words and sentences.
• Model: it’s a math machine that predicts something.

What does it predict? The next word. Over and over and over again.

Really big autocomplete

You know how your phone suggests the next word when you type? An LLM is that idea, taken about a million steps further.

1. You type: “The capital of France is”
2. The LLM thinks: “Hmm, in everything I’ve read, what word usually follows that?”
3. It picks the most likely word: “Paris”.
4. Then it does it again, picking the next word, and the next, and the next.

It writes whole essays one word at a time.

Why does it feel so smart?

Because it has read so much. Books, websites, forums, code, conversations. To get good at guessing the next word, it had to learn:

• How sentences are built.
• What words usually go together.
• Who Cleopatra was. What gravity does. How a recipe is shaped.

Predicting words well requires a tiny bit of “world knowledge”, so the LLM picks up a lot of that for free.

But… it doesn’t understand

Not like you do. It has never seen a sunset or felt cold rain. It just has a giant blurry memory of how humans talk about sunsets and cold rain.

That’s important to remember. It’s a parrot with a library card. A very, very good parrot.