Large language models are the technology behind many of the AI text tools that have become widely available. Understanding what they are — and what they are not — helps you use them more effectively and interpret their output more critically.
What Is a Large Language Model?
A large language model (LLM) is a type of AI system trained on very large amounts of text data to predict and generate language. The word "large" refers to the scale of both the training data and the model itself — these systems have billions of learned parameters.
LLMs learn statistical patterns in language: which words and phrases tend to follow other words and phrases, across an enormous range of topics and styles. They use these patterns to generate responses to prompts.
Tokens
LLMs do not process text word by word. They process tokens — chunks of text that may be a word, part of a word, or a punctuation mark. Tokenization is how the model breaks text into units it can process mathematically.
Prompts and Context Windows
A prompt is the input you provide to an LLM — a question, instruction, or piece of text you want the model to respond to or continue.
A context window is the amount of text the model can consider at one time. Text outside the context window is not available to the model when generating a response. This is why very long conversations or documents may cause the model to lose track of earlier content.
Why Fluent Output Can Still Be Inaccurate
LLMs are trained to produce text that is statistically plausible — text that sounds like something a person might write. They are not trained to verify facts. A model can produce a confident, well-structured, grammatically correct response that is factually wrong.
This is sometimes called hallucination — the model generates content that sounds plausible but is not accurate. It may fabricate names, dates, citations, statistics, or events.
An LLM Is Not a Search Engine
A search engine retrieves existing documents from an index. An LLM generates new text based on patterns in its training data. These are fundamentally different operations. An LLM does not search the web when you ask it a question — it generates a response based on what it learned during training, which has a knowledge cutoff date.
An LLM Is Not a Database
A database stores structured data and retrieves exact records. An LLM generates probabilistic text. It cannot reliably retrieve specific facts, figures, or records the way a database can.
Embeddings and Retrieval-Augmented Generation
Embeddings are numerical representations of text that capture meaning and relationships. They are used to find semantically similar content. Retrieval-augmented generation (RAG) is a technique that combines an LLM with a retrieval system — the model can access relevant documents before generating a response, which can improve accuracy on specific topics. These are more advanced concepts; see the glossary for details.
- LLMs generate text based on statistical patterns in training data
- Fluent, confident output can still be factually incorrect
- LLMs are not search engines and not databases
- Context windows limit how much text a model can consider at once
- Always verify important facts from LLM output independently