Back to the wiki

LLM (large language model)

The engine behind ChatGPT, Claude and Gemini.

The analogy

Imagine someone who has read millions of books, websites and conversations, and has become incredibly good at a single game: guessing the next word. So good that, by chaining one prediction after another, they can write essays, translate, or code. That's an LLM: autocomplete taken to the extreme.

In detail

An LLM is a neural network (usually a transformer) trained on vast amounts of text to predict the next token in a sequence. With enough data and parameters, abilities like reasoning, summarizing and translating emerge. It doesn't look things up in a database — it generates text by computing probabilities.

An example

You type “The sky is” and the model assigns probabilities: “blue” (very high), “cloudy” (high), “green” (very low). To generate a full answer, it repeats that calculation hundreds of times, word by word.

Related concepts