Large language models concept

What is an LLM?

Reference

What is an LLM? An LLM, or large language model, is a language model trained on a large collection of text to work with language. It estimates what token—a piece of text such as a word or part of one—might fit next in context, then can repeat that prediction to generate text. LLMs can be used in applications such as chatbots, but a fluent answer can still contain mistakes.

Reviewed 2026-10-09

What a large language model is

A large language model is a model that works with language by estimating how likely tokens or sequences of tokens are in context. A token is a unit of text used by the model; it might be a whole word, part of a word, or a single character. Large language models are trained on examples so they can make predictions about language and generate or process text.

“Large” refers to a model with many parameters, which are values used in its computations. There is no single size cutoff that settles every use of the term: Google’s glossary describes an LLM as having a very high number of parameters, and says that, more informally, the term can mean any Transformer-based language model.

How it learns and generates text

During training, a model makes predictions about text and adjusts its internal values based on how those predictions compare with the training objective. One common objective is next-word prediction: given text so far, predict what comes next. Some language-model training instead hides tokens and trains the model to predict the missing pieces.

When a model generates text, it uses the prompt and the text already produced as context for its next prediction. It adds a token and predicts again, continuing until it stops. This repeated process can produce sentences and longer passages, even though the model generates them piece by piece. The exact training process and the application around a model can include other steps.

Where LLMs fit

An LLM is one kind of language model. The term foundation model describes a broad, pre-trained model that can be adapted for different tasks. Some LLMs can serve as foundation models, but the terms describe different things: “language model” points to the kind of model, while “foundation model” points to its broad training and possible reuse.

A chatbot is an application that uses a language model to hold a conversation. The application gives people a way to interact with the model; the model produces the language. So “LLM” names the model, while “chatbot” names an application that may use one. Other applications can use language models for tasks such as generating, translating, or summarizing text.

Common confusions

Predicting likely language is not the same as checking an answer against a reliable source. A model can produce a prediction that contains mistakes; Google’s learning material calls this hallucination. A polished sentence is still a model output, so its fluency alone does not establish that its claims are correct.

Next-token prediction is a useful way to explain one common kind of text generation, but it is not a complete description of every language model. For example, some training asks a model to predict tokens deliberately hidden in text. And a chatbot may include an application around its language model, so the model’s predictions and the full application’s behavior are not identical.

Questions people ask

What does LLM stand for?

LLM stands for “large language model”: a language model with a very high number of parameters, or, more informally, a Transformer-based language model.

What does an LLM do when it writes a response?

It uses the prompt and earlier text as context to predict likely next tokens, adding them piece by piece.

Is an LLM the same thing as a chatbot?

No. An LLM is a model; a chatbot is an application that uses a language model for conversation.

Can an LLM give a wrong answer?

Yes. LLM predictions can contain mistakes, sometimes called hallucinations.

Learn more

Introduction to Large Language Models 45 min
This free module introduces language-model predictions, tokens, context, and how large language models are created.
Machine Learning Glossary
The glossary gives short definitions for large language models, foundation models, and chatbots.

What changed

Sources

Part of Large language models: From text to tokens and back How a model learns and continues patterns Where fluent output can fail