Short answer
A large language model is a neural network trained on large volumes of text to predict likely continuations, which lets it write, summarise, classify and translate without being programmed for each task. GPT, Claude and Gemini are examples. The model holds no live knowledge of a business unless that information is supplied to it.
Also called: LLM, Language model
Prediction is the whole mechanism, and understanding that explains most of the behaviour people find surprising. The model has no database to consult and no internal test for whether its output is true; it produces the continuation that fits the patterns in its training data. That is why it writes fluent prose about a topic it half-knows, why the same prompt can return two different answers, and why its knowledge stops at the date its training data ends.
For a business, the practical questions are narrower than the technology debate suggests. Which model, at what cost per request, with what response time, and how does it get access to your own information? The last one matters most: a model becomes useful for customer work only once it can read your prices, your stock or your policies, and that is done by supplying the content at request time rather than by retraining the model.
Common questions
Is ChatGPT the same thing as an LLM?
ChatGPT is an application built on top of a large language model, with a chat interface, memory features, search access and safety layers wrapped around it. The model is the engine; the product is the car. A business integrating AI into its own systems usually calls a model directly through an API rather than using the consumer chat product.
Where this comes up in our work
Related terms
Retrieval-Augmented Generation (RAG)
Retrieval-augmented generation is a technique that searches a private collection of documents for passages relevant to a question, then passes those passages to a language model as context for its answer.
Fine-Tuning
Fine-tuning is further training of an existing language model on a set of example inputs and outputs, so it adopts a specific style, format or classification behaviour by default.
Hallucination (AI)
A hallucination is output that is fluent, plausible and false: an invented citation, a policy the company never wrote, a confidently wrong figure.
Inference Cost
Inference cost is what a business pays each time a model produces an output, billed by most providers per token of input and output, where a token is roughly three-quarters of a word.
Reading definitions because you are scoping a project? Skip ahead and just ask.