What is a large language model (LLM)?
By Yassine Seghrouchni, founderUpdated
In short
A large language model (LLM) is an AI model trained on large amounts of text that predicts which words logically follow its input. It is the technology behind ChatGPT, Claude, Gemini and Copilot. An LLM doesn’t know facts: it generates text that is likely, given what you put in.
Also known as: LLM, language model
What it does and doesn’t do
The model receives text and calculates, word by word, what is likely to follow. Nothing is looked up and nothing is checked, unless the model is explicitly allowed to use a search function or database. That determines what an LLM is good and bad at.
Good at: wording, rewriting, summarising, translating, structuring, adjusting the tone and spotting patterns in text you provide.
Weak at: exact calculation, reproducing facts from memory, knowing what happened recently, and admitting that it doesn’t know something.
Why it matters in practice
The key insight: a language model isn’t a knowledge base but a text generator. People who understand that start checking factual statements and asking for sources as a matter of course.
People who don’t copy a made-up legal article or a quote that doesn’t exist. Exactly the kind of mistake that does damage you can’t undo. See also AI hallucination.
Give it context
For everyday use, this means: give the model the information instead of hoping it knows it. Paste in the contract, the product list, the previous quote. An LLM working with your documents is far more reliable than one drawing on its training.
That’s why tools that work in your own environment, with your email, your documents and your CRM, are often more useful than an empty chat window. How to set that up responsibly is covered in ChatGPT for business.
Frequently asked questions
Why does an LLM sometimes give a wrong answer that sounds convincing?
Because the model aims for likely text, not for truth. To the model, a fluent, confident wrong answer is just as likely as a right one. How convincing it sounds tells you nothing about whether it’s correct. That’s why you always check factual statements.
Does an LLM learn from what I type into it?
That depends on the version and the settings. With business versions from the major providers, training on your input is usually switched off under their terms. Free and consumer versions often have different defaults. Check it per tool, and don’t let staff use personal accounts for work.
