Large Language Model (LLM)
Last Updated: September 10, 2026 | By Mihail Sebastian | AI Dictionary
A neural network trained on vast text to predict the next token; at sufficient scale that single objective yields broad ability across language tasks.
What is a Large Language Model (LLM)?
A large language model (LLM) is a neural network trained on enormous volumes of text to predict the next token in a sequence. At sufficient scale, that single objective produces broad ability: answering questions, writing code, summarizing documents, translating languages.
LLMs are the engines behind generative AI products such as GPT (OpenAI), Claude (Anthropic), Gemini (Google), and Llama (Meta). The product a user talks to is an application; the LLM is the model underneath it.
How Large Language Models Work
Modern LLMs are built on the transformer architecture, whose attention mechanism lets the model weigh every token in its input against every other. Text is first split into tokens, subword units the model reads and writes; a token is roughly a short word or a piece of a longer one.
Training happens in two broad phases. In pretraining, the model reads internet-scale text and learns one thing: predict the next token. Grammar, facts, reasoning patterns, and style all emerge as means to that end. Afterwards, fine-tuning and human feedback shape the raw predictor into an assistant that follows instructions and refuses harmful requests.
The model generates output the same way it was trained: one token at a time, each drawn from a probability distribution over what comes next. This is why LLMs hallucinate: a fluent, plausible continuation is exactly what the model is built to produce, whether or not it is true. Everything the model considers while predicting must fit inside its context window.
Large Language Models vs Foundation Models
An LLM is one kind of foundation model: the foundation model is the broader class, and the LLM is what you get when the modality is language. Foundation models also include vision, audio, and multimodal systems adapted to many downstream tasks.
| Criterion | Large Language Model | Foundation Model |
|---|---|---|
| What it names | A large pretrained model for language | The broad class: any large pretrained model adapted to many tasks |
| Modality | Text, including code | Text, images, audio, video, or combinations |
| Relationship | An instance of the class | The class itself |
| Examples | GPT, Claude, Gemini, Llama | LLMs plus vision and multimodal models |
The distinction matters for governance. The EU AI Act regulates both under the heading of general-purpose AI models, with documentation and transparency duties for their providers.
Example of a Large Language Model
Ask an LLM “Explain our refund policy in two sentences” with the policy pasted in. The input is tokenized, and attention lets the model relate the word “refund” in your question to every clause of the policy that mentions one.
Then generation begins. The model predicts the most likely first token of a good answer, appends it to the input, and predicts the next, looping until the summary is complete.
Nothing was looked up and no rule was followed. The answer is a prediction of what a correct summary looks like, which is why it is usually right and occasionally, confidently, wrong.
Related AI terms: Foundation Model · GPT · Generative AI · Prompt Engineering · Hallucination
Did you like the Large Language Model (LLM) gist?
Learn about 250+ need-to-know artificial intelligence terms in the AI Dictionary.
