What is a large language model (LLM)?

On some shopping sites, above the reviews, there's now a short paragraph that starts with something like "Customers say." Hundreds of reviews, boiled down to a few sentences that no person wrote. The software behind a paragraph like that is very likely a large language model (LLM), the same kind of engine that sits inside today's chatbots. You may be reading its work more often than you think.

The name is three plain words, taken one at a time.

• Large: its size, both in the amount of writing it learned from and in how many learned numbers it holds.
• Language: what it works with, which is text, the words people write.
• Model: the learned part, a large file of numbers that turns what goes in into what comes out.

Put together, that's a very big model, trained on a very large amount of writing.¹²

An LLM learns from a vast library of writing: books, articles, websites, computer code, and more. As of 2026, the biggest are trained on trillions of words. Nobody labels all of that by hand, because the writing already holds the answers: every passage shows what came next. As it works through all that text, the model picks up how language tends to go, from grammar and spelling to the facts people write about and the way an argument usually unfolds.¹²

All that training builds toward a single skill: predicting what text should come next. Give it the start of something, and it continues in the most likely way, based on everything it learned. Put plainly, a large language model is a model trained on huge amounts of text to predict what comes next, and that one skill is what lets it write, summarize, and answer.¹²

It sounds too simple to be useful, until you notice how many jobs can be framed as "what comes next." An answer is what comes next after a question. A summary is what comes next after a long text and a request to shorten it. A translation is what comes next after a sentence and a request for it in Spanish. A first draft is what comes next after a description of what you want.¹²

That skill explains what LLMs do well. They're fluent: the writing usually reads well, in almost any style you ask for. They're fast, turning out in seconds what might take you an hour. And they're good with words in general. They can rephrase, shorten, explain a hard paragraph in plainer language, or turn rough notes into a tidy email.¹

The same skill explains the limits. An LLM produces what's likely to come next, and likely can sound right whether or not it's true. So it can be fluent and wrong at once, with a made-up figure or quote that reads as smoothly as the rest. It also knows only what was in its training, plus whatever the app hands it, so recent events can be missing. Lean on it for the wording, and check the facts that matter yourself, the way you'd check anything a fast, fluent writer handed you.²