LLM (large language model)

LLM (large language model): the explanation

During training, the model adjusts many parameters to learn patterns in data. Common generative systems then produce an answer progressively by estimating which tokens fit the available context. A token can be a word, part of a word, or another unit. The response depends on your request, the instructions supplied, and the information included in context. This supports flexible language tasks, but it does not mean the model automatically checks a trusted source before producing each statement.

LLMs provide the language engine inside many assistants and text tools. You can ask one to rewrite an email, compare documents, or explain an unfamiliar concept. An application may also supply search results and external tools. The model is still only part of the overall product. Interfaces, access permissions, optional memory, and execution of actions are generally managed by surrounding software. The same underlying model can therefore behave differently in applications with different instructions, tools, and safeguards.

Large does not refer to a universal size threshold that makes model comparisons straightforward. Performance also depends on training data, training methods, adaptation, and the task being tested. A model can write fluently while making mistakes, lacking recent information, or misreading a document. Some language models also accept images or audio. You need to check those capabilities separately because the abbreviation alone does not describe every supported input, output, limitation, or integration available in a particular application.

A concrete example

You need to answer a long email about an upcoming work trip. You give the message to an assistant and request a summary of the agreed arrangements followed by a polite draft response. The LLM organizes the information and proposes suitable wording. You check names and travel times before sending it. Without a booking tool and the relevant authorization, the model does not itself reserve your train ticket or hotel room.

Why it matters

Understanding an LLM’s role helps you separate model performance from application quality. When choosing a tool, test your actual languages, documents, and constraints. Consider privacy, response times, and ways to verify answers. Strong writing ability does not guarantee dependable factual research, and a capable model does not automatically give an agent appropriate permissions. The surrounding application remains important for deciding what information it receives and what actions can follow.

Common misconceptions

• An LLM does not necessarily search the internet when answering: it may rely only on its learned parameters and supplied context unless the application actually provides access to search or another external source of information.
• A larger model is not automatically better for your needs: one adapted to your language, documents, or a narrower task may be more useful, respond faster, or fit more easily into the application you intend to use.

Related terms: language models

hallucination

A hallucination is an AI output that presents false, invented, or unsupported information as though it were reliable.

Transformer

A Transformer is a neural network architecture that processes elements, such as pieces of text, by calculating relationships between them through attention.

embedding

An embedding is a numerical representation of content, usually expressed as a list of numbers called a vector.

token

A token is a unit that a model uses to represent and process part of its input or output.

context window

A context window is the maximum amount of content a model can consider within a particular processing operation, usually measured in tokens.

open weights

Open weights are learned model values made available for others to download and use under the applicable terms.

open-source AI

Open-source AI refers to AI made available with freedoms to use, study, modify, and share it, together with the materials needed to exercise those freedoms.

quantization

Quantization is a technique that stores or processes some of a model's numbers with reduced precision.

Where this site covers it

The term appears on 664 pages of this site. A few of them:

• Getting started with an LLM API: keys, your first request, settings, costs, and best practices Guide
• Local LLM reference guide: software compared, models by machine size, performance, and uses Guide
• Understanding LLM API pricing: tokens, input and output, caching, batch processing, and budgeting Guide
• Quantization and GGUF: how they work, Q4 and Q8 levels, quality trade-offs, and choosing a file Guide
• Reduce AI agent costs: model selection, caching, context, routing, and budgets Guide
• Reasoning models: how they work, what they improve, and measured limits Guide
• Test-time adaptation: how it works and what results show Guide
• Choosing a language model: criteria, benchmarks, pricing, context, and licensing Guide