How do large language models actually create new text?

Large language models are like super-smart text detectives who know how to write new stories by looking at old ones.

Imagine you have a big book full of different stories, fairy tales, jokes, and even your favorite bedtime story. A large language model is like someone who has read all those stories so many times that they can guess what comes next, just by looking at the beginning of a sentence or two.

They use patterns, just like you might notice that every time your friend starts telling a joke, it ends with a funny punchline.

Think about it like this: if you're writing a story and you want to say something new, you look at the words before it to figure out what makes sense. That’s exactly what large language models do, they check the words around a new word or sentence to decide which ones fit best.

They’re not magic, they're just really good at recognizing patterns in text, like how you know that after "Once upon a time," there's usually a story about a prince or princess.