What is a language model?
A language model, often called an LLM for large language model, is an AI model trained on large amounts of text to understand and write language.
A language model predicts which text is likely to come next. It has learned this by reading huge amounts of text, and the result is answers that sound natural.
The model only knows what was in its training. Your own routines, contracts and decisions are not there. Ask about them and the model describes how things usually work, which can sound right and still be wrong for you.
That is why language models are increasingly used together with an organisation's own documents. The model phrases the answer, and the content comes from what you have written.
How it works
The model works with tokens, small pieces of text. It reads the question and whatever material it has been given, then works out the answer one word at a time based on probabilities. How much it can take into account at once is limited by its context window.
Why it matters
The language model decides how well answers are phrased and how well a vague question is understood. For an organisation, though, the material and where the model runs matter more. A model that answers from the right documents in the right place is worth more than a larger one that guesses.
What to keep in mind
Models differ in language, size and terms of use. Some can run in your own environment, fully offline. Others are only available as a service from a provider.
A larger model does not automatically give better answers for your work. Test with your own questions and documents before you decide.
What it means for you
Ask how an AI service gets access to your documents. That matters more than which model it is built on.
Common questions
How is a language model different from an AI assistant?
The language model is the engine that understands and writes text. An AI assistant is the tool built around it, with instructions, permissions and a connection to your documents.
Does the language model know what is in our documents?
No. It has learned from general text up to a certain point in time. To answer from your documents, it needs to read them when the question is asked, for example through RAG.
Where does the language model run?
That depends on the solution. It can run at an external service or in your own environment as local AI. Where it runs decides where your questions and documents are processed.
Related terms
Read more
All terms Updated