What is large language model (LLM)?

A large language model (LLM) is a neural network trained on a vast amount of text to predict the next token, the next word or part of a word. That one skill, scaled up, lets it write replies, summarize conversations, translate, label a message by topic and follow instructions written in plain language. Every generative support chatbot has one underneath.

It matters because the model's training is where its general knowledge comes from, and training ends on a date. A model does not know your return policy, this week's price or an order that shipped yesterday. Support products supply those facts at reply time, through retrieval and tool calls, and tell the model to answer only from them. The model writes; the application decides what it may write from.

Where definitions blur: "LLM" names the model, not the product. A vendor that says "powered by GPT" is naming its supplier, and two products on the same model can behave very differently depending on their prompts, retrieval and guardrails. Model providers bill per token, so some chatbot vendors pass that cost through as credits while others sell flat plans, and the choice of model also sets reply speed. The model is necessary and not sufficient: a weak knowledge base on a strong model still gives weak answers.