Jargon Buster / Large Language Model (LLM)
Large Language Model (LLM).
AI & AutomationAn AI system trained on huge amounts of text that can write, summarise, answer questions and follow instructions.
What is a large language model?
A large language model (LLM) is an AI system trained on very large amounts of text to predict the next piece of text, which lets it write, summarise, translate, answer questions and follow instructions in everyday language.
LLMs are the engine behind tools such as ChatGPT, Claude, Gemini and Microsoft Copilot, and behind most of the AI features now appearing inside business software.
Why it matters
Almost every AI tool you buy or build runs on an LLM, so understanding the basics helps you judge claims, costs and risks. The model affects quality, speed and price. How you supply it with your own information affects accuracy far more. The plan you use decides what happens to your data. Getting those three choices right is the difference between AI that saves your team hours and AI that produces confident nonsense.
How it works
- Training on text The model learns patterns of language from vast quantities of text, then is further trained to follow instructions helpfully.
- Tokens as units Text is broken into small units called tokens, which is also how usage is measured and priced.
- Context window The model can only consider a limited amount of text at once: your instructions, documents and the conversation so far.
- Word-by-word prediction It generates its answer one token at a time, choosing likely continuations, which is why it can sound right and still be wrong.
- Tool use Many models can now search, run code or call other software, which extends what they can do.
The honest view
The choice of model matters less than most people think. The leading models leapfrog each other every few months, and for everyday business tasks several are more than good enough. The context you give the model matters much more: an average model with your price list beats a brilliant one guessing.
Two things are underrated. Smaller, cheaper models handle routine jobs perfectly well. And designing your set-up so you can swap models later protects you from price changes and vendor lock-in.
What to do about it
Check the data terms of whichever AI plan your team uses, and move to a business tier if inputs could otherwise be used for training. Choose two or three real tasks and test more than one model on your own examples rather than trusting public benchmarks. Keep prompts and documents outside any single tool where you can, so changing models later is a configuration change rather than a rebuild.
Related terms.
Token
The small unit of text an AI model reads and writes, used to measure its limits and price its use.
AI & AutomationSee more GGenerative AI
AI that creates new text, images, audio, video or code in response to plain-language instructions.
AI & AutomationSee more HHallucination
When an AI model states something false or invented with complete confidence.
AI & AutomationSee more RRetrieval-Augmented Generation (RAG)
A method where AI first looks up your trusted documents, then writes its answer from what it finds.
AI & AutomationSee moreGot a question about your own setup? Let's talk.
Your first call is with our founder, Rhodri Williams.