Large Language Model (LLM)
What it means
A large language model, or LLM, is a type of artificial intelligence model trained to recognise patterns in large amounts of text and use those patterns to process and generate language.
The training idea is surprisingly simple to describe. The model is shown enormous quantities of written material — books, articles, websites, transcripts — and learns to predict which words tend to follow which others. After enough exposure, it becomes very good at guessing the next likely word in a sentence, and the word after that, and so on. That ability to predict language is what lets an LLM read a prompt and produce a fluent reply.
An LLM does not store the text it was trained on like a filing cabinet. It learns statistical relationships — which words and concepts tend to appear together, how questions are usually answered, what a polite email sounds like, what a summary looks like. When you ask it to explain something, it draws on those patterns to produce text that reads like an explanation. Tools such as ChatGPT are built on top of an LLM, adding a conversation interface and other safeguards on top of the model itself.
It is tempting to imagine an LLM as a person who understands you. In practice, it is closer to a very sophisticated pattern-completer. Because it generates one likely word at a time, its writing can sound thoughtful, structured and confident — but that fluency does not necessarily mean the content is accurate. An LLM can produce a sentence that is grammatically perfect and factually wrong, because it is optimising for plausible language rather than for truth. This is the root of AI hallucination.
Context matters too. An LLM only considers what you give it (and what it can infer from training), so the same model can give a shallow answer to a vague prompt and a genuinely useful answer to a well-framed one. This is why the way you phrase a request — the prompt — has such an outsized effect on the result.
The relationship between an LLM and an AI assistant is worth keeping clear: the LLM is the engine; the assistant is the product wrapped around it, with a conversation view, memory of the current chat, safety filters and a friendlier interface. Knowing this helps you set realistic expectations. The model is remarkable with language; it is not a person, a database or an oracle, and it works best when you treat it as a capable language tool that still benefits from your judgement.
Why it matters
Understanding what an LLM actually does — generating likely language rather than retrieving verified facts — explains both its usefulness and its risks. It is why an LLM can write beautifully and still be wrong, and why fluent answers deserve the same scrutiny as a confident stranger's. Knowing this helps you use these tools well: lean on them for language tasks, give them enough context to be useful, and check anything that has to be right.
A practical example
Ask an LLM 'what's a good way to organise digital files?' and it will draw on countless articles it has seen on the topic to produce a sensible, well-structured answer. The response reads as though someone knowledgeable wrote it — and much of it may be genuinely useful. But if you then ask for a specific statistic ('how many people use cloud storage in the UK?'), it may produce a plausible-sounding number with no real source behind it. The first answer draws on general patterns; the second needs a fact the model has no reliable way to look up.
Related terms
This term is explored fully in ChatGPT Made Simple.