Large Language Model

A large language model (LLM) is an AI system trained on billions of pages of text, covering everything from Wikipedia articles and scientific papers to books, forums, and code repositories. By processing this enormous volume of material, the model learns the patterns that give language its meaning: how words relate to one another, how arguments are structured, and how tone shifts between a formal report and a casual message.

LLMs generate text by predicting the most likely next token, one small piece at a time, using a transformer architecture that weighs the context of everything written so far. They do this so fluently that the output often reads as though a person wrote it, which is why LLMs now sit at the heart of most AI assistants, from customer support bots to writing tools.

How it works#

During training, an LLM is shown vast amounts of text with portions deliberately hidden. It must predict the missing words, and its internal settings, called parameters, are adjusted billions of times until the predictions become reliably good. The more parameters a model has, the more nuance it can capture, which is why "large" is part of the name.

At inference time, when you ask the model a question, it reads your prompt and generates a response token by token. Each new token is chosen based on the probability of it making sense given everything before it. This process happens very quickly, which is why responses appear within seconds.

LLMs do not look things up in a database of stored answers. Everything they produce is generated fresh, drawing on the statistical patterns absorbed during training.

Examples#

Writing assistant. When you ask an AI chatbot to "write a polite email declining a meeting invitation", the LLM does not retrieve a template. It draws on patterns from thousands of similar emails it processed during training, then generates new text that fits your specific request. Change one word in the prompt and the output changes too.

Code helper. A developer types a comment describing what a function should do. The LLM reads the comment and generates working code in the appropriate programming language. It applies the same pattern-matching approach it uses for prose, because code is, in many ways, just another kind of structured text.

Why it matters for business#

LLMs are changing how organisations handle writing-heavy work. Tasks that once took a human an hour, such as drafting a first version of a report, summarising a long document, or translating content into another language, can now take seconds with AI assistance. Teams that learn to give clear, well-structured instructions get far better results, which is why prompt engineering has become a valued skill.

Understanding what an LLM actually is, a pattern-based text generator rather than an all-knowing oracle, also helps people use it more safely. Knowing that the model can generate confident-sounding errors, called hallucinations, encourages the habit of checking important outputs before acting on them.

To see how LLMs work in practice and start using them effectively, explore the AI Fundamentals path or read the tutorial How LLMs Work.