Large language model (LLM)
- AI basics
- Standard term
A large language model, or LLM, is the AI inside many chat apps. Training on vast amounts of text teaches it to predict tokens, small pieces of text, supporting writing, answering questions, and other language tasks.
The model learns patterns in language and relationships among ideas. It builds an answer one small piece at a time, using the request and the text already written to choose the next piece. Further training can help it follow instructions and produce the kinds of answers people prefer.
Where you’ll meet it
The AI app your company provides uses the same model as the assistant you use at home. Yet the work app can't search the web or open your files, even when you ask the same way.
What to do
Check which actions, such as web search and file reading, the work app allows before switching models. If file access is missing, the person managing the app may need to enable a connection or provide an upload option. The model's name alone doesn't tell you what the app can reach.
Easy to mix up
| Large language model (LLM) | AI assistant |
|---|---|
| The trained model that reads what it receives and writes a response | An app combining the model with instructions, actions such as searches, and other features |
Check yourself
You stop an LLM halfway through an email and change its last sentence. You send that version back and ask it to finish. Which draft is the starting point for the new continuation?
Go deeperThe Transformer paper introduced a design used by many language models
Vaswani and colleagues introduced the Transformer in their 2017 paper, “Attention Is All You Need.” Its attention mechanism helps the model relate one part of the text to another as it processes the input. Many large language models use this design.
Google's Machine Learning Glossary describes “large” in terms of a model having many parameters: numbers adjusted during training that control the model's calculations. It also uses “foundation model” for a broader category trained on large, varied data, which can extend beyond language.
Sources
- LLMs: What's a large language model?, Google Machine Learning Crash Course
- Attention Is All You Need, Vaswani et al., NeurIPS 2017
- How Claude Code works, Anthropic Claude Code docs