What Is an LLM (Large Language Model)?
An LLM (large language model) is a type of artificial intelligence trained on enormous amounts of text to understand and generate language. It works by predicting the next word in a sentence based on patterns it has learned. It is the technology behind chatbots and AI assistants, and the engine that modern AI agents are built on.
Table of contents
In short: the engine behind modern AI
When you write to a chatbot and get a coherent answer back, a language model is working in the background. The abbreviation LLM stands for large language model, in Danish "en stor sprogmodel". "Large" refers both to the amount of text the model is trained on and to the model's own size, measured in the parameters it uses to represent language.
For a business, it's worth understanding what a language model is, because it is the building block of almost everything that goes by the name "AI" today: customer service bots, writing assistants, search tools and the AI agents that can carry out tasks across systems. If you understand the engine, you also understand better what the technology can realistically do, and where its limits lie.
How does a language model work?
You don't need to be a data expert to understand the basic principle. It is about text, patterns and prediction.
Trained on large amounts of text
A language model is built by having it read gigantic amounts of text: books, articles, websites, documentation and much more. During this training the model doesn't memorise facts like a reference book. Instead it learns the statistical patterns in language: which words typically follow one another, how sentences are constructed, and how concepts relate to each other.
The result is a model with a very nuanced feel for how language looks and is used, across subjects, tones of voice and languages.
Predicting the next word
The core of how a language model works is surprisingly simple to describe: it predicts the next word. When you ask a question, the model calculates which word is most likely to come next, then the next, and the next, until a whole answer has been formed.
Because the model has seen so many examples of well-written text, these predictions become so good that the output appears coherent, relevant and often surprisingly accurate. It is important to remember, though, that the model is fundamentally calculating probabilities for language; it doesn't "understand" the world the way a human does.
What a language model can and can't do
A language model is a powerful tool, but it has clear strengths and equally clear limitations. It is essential to know both before basing business processes on the technology.
| Typically good at | Can't do (without help) |
|---|---|
| Writing, rewriting and summarising text | Guarantee that all facts are correct |
| Answering questions in natural language | Know about your internal data without access to it |
| Translating between languages | Act in your systems on its own |
| Classifying and structuring text | Know anything that came after its training |
| Extracting information from documents | Verify its own answers itself |
In short: a language model is excellent at working with language and the knowledge it has seen. But on its own it is passive: it answers when asked, and doesn't act by itself.
The difference between a language model and an AI agent
This is where many people mix up the terms, and the difference matters to any business considering AI.
The language model is the engine. The AI agent is what runs on the engine and actually gets the work done.
A language model can understand and formulate language, but it can't look something up in your calendar, send an email, update a CRM system or fetch data from a database on its own. An AI agent combines the language model with tools, access rights and the ability to take several steps in a row, so that it doesn't just answer but acts.
Put simply:
- Language model: "Here's a suggested reply to the customer."
- AI agent: Reads the customer's enquiry, looks up the relevant information, drafts the reply and sends it, or has it ready for approval.
If you want to understand the next layer (how an agent uses the language model to carry out real tasks), read on in our guide to what an AI agent is for a business.
At Agent360 we build AI agents on top of language models. The model itself is a ready-made component; the value arises when it is connected to your processes, data and tools and set to solve specific tasks.
Well-known language models
Several large language models are known by name today. Among the most widespread are:
- GPT, developed by OpenAI and known from ChatGPT, among others.
- Claude, developed by Anthropic.
- Gemini, developed by Google.
What they have in common is that they are general-purpose language models that can be used for a wide range of tasks. They differ in details such as style, strengths and price, but the basic principle (training on large amounts of text and prediction of language) is the same. For most businesses it matters more to choose the right use and setup than to focus single-mindedly on one particular model name.
Limitations you need to know about
A language model is a tool, not an oracle. Two limitations are particularly important to understand before you start using the technology in a professional setting.
Hallucination
A language model can produce answers that sound convincing but are factually wrong. This phenomenon is called hallucination. Because the model generates language from probabilities (and doesn't look things up in a verified source), it can come up with something that isn't correct, in the same confident tone as a right answer.
In practice this means you shouldn't use a plain language model as the only basis for important decisions without some form of check. A well-designed solution handles this by giving the model access to reliable sources and building checks into the process.
Outdated knowledge
A language model only knows what it saw during training. In other words, it has a "knowledge cutoff": events, prices or rules that have come about since the training are not something it knows by default.
This means a language model isn't up to date on recent developments by itself, and doesn't know your current internal data unless it is given access to it. That is precisely why language models are often connected in practice to your own systems and up-to-date sources, so that the answers are based on what actually applies at your company.
From technology to value
Understanding what a language model is is the first step. The next is to translate the technology into something that creates real value in a business. A language model on its own rarely solves a business task: it does so when it is placed in a context with the right data, access rights and controls around it.
That is precisely the link an AI agent provides: it uses the language model as its engine, but adds the ability to act across your tools and processes. The language model supplies the understanding of language; the agent gives it the ability to get the work done.
Want to understand what AI can do for your business?
At Agent360 we help Danish businesses move from "AI as a concept" to AI that solves concrete tasks. We build AI agents on top of language models and adapt them to your processes.
Book a no-obligation AI consultation with Agent360, and we'll talk about where the technology can realistically create value for you, and where it can't.
Contents
What takes up your time?
Write down the task that takes up the most time. We reply whether it is worth building, and we say so if it is not.
Write what takes up your timeRelated articles

MFA: multi-factor authentication for your business – how to get started
Learn how to implement MFA securely in your organisation. Guidance on methods, costs and best practices for protecting company data.

English-language invoices: requirements for invoices to foreign customers
Invoice in English to foreign customers legally. Danish regulations, EU and international VAT rules, zero-rating, reverse charge, and automation.

Drawing programme (CAD): 2D, 3D, licences and file formats
What is CAD? Learn about 2D and 3D design programmes, file formats, licensing models, and how to choose the right tool for technical drawings.