Skip to content

Basics

What Is an LLM? Large Language Models Explained Simply

AI assistants are powered by large language models, known as LLMs. If you understand how they work in principle, you can use them better – and know when caution is called for. An explanation without formulas.

Lurus Editorial Team Published Reviewed 3 min read
Painted scene: an open notebook with sketches by a window overlooking a lake.

Key takeaways

  • An LLM is an AI model that has learned to continue language from large amounts of text.
  • It calculates probabilities for the next piece of text – it doesn’t look things up in a database.
  • Tokens and the context window determine how much text a model can process at once.
  • LLMs can produce convincingly worded but false statements.
Contents
  1. LLM: the definition
  2. How a language model works
  3. Tokens and the context window
  4. What LLMs do well
  5. Where the limits are
  6. One model or several?

LLM: the definition

LLM stands for large language model. It is an AI model trained on very large amounts of text that has learned to understand and generate language. LLMs are the foundation of today’s AI assistants, which write, summarize and translate text, answer questions or generate program code.

“Large” refers to two things: the amount of training data and the number of parameters – the internal settings the model adjusts during training.

How a language model works

Put simply, an LLM predicts which piece of text fits best next. It takes your input, calculates a probability for possible continuations, picks one, appends it and repeats – piece by piece – until the answer is complete.

That sounds simple, but by training on huge amounts of text, the model has captured patterns in language, facts, reasoning and style. That is why it can solve tasks that go far beyond completing sentences. Researchers have shown that large language models can often solve new tasks from just a few examples in the prompt, without being retrained.

The transformer architecture

Almost all of today’s LLMs are based on the transformer architecture, introduced in 2017 in the research paper “Attention Is All You Need”. Its core is the so-called attention mechanism: for each word, the model weighs which other words in the text matter for its meaning. That lets it capture relationships across long passages of text.

Tokens and the context window

LLMs don’t process text letter by letter, but in tokens. A token is a unit of text, often a word or part of a word. Long and compound words usually consist of several tokens.

The context window indicates how many tokens a model can consider at once – your input, attached documents, the conversation so far and the answer. Once the window is full, the model can no longer take earlier parts into account. For working with long documents, a large context window is therefore important. Tokens are also often the basis for billing; more on this in What does AI cost a business?.

What LLMs do well

The German Federal Office for Information Security (BSI) names these strengths of large language models, among others:

  • generating text, such as drafts of emails, reports or descriptions
  • editing text – rephrasing, shortening or changing the tone
  • summarizing and translating text
  • answering questions and structuring information
  • generating and explaining program code

Typical business applications are described in AI use cases for small businesses.

Where the limits are

  • No real factual knowledge: An LLM calculates plausible continuations. In doing so, it can produce convincing-sounding but false statements – so-called hallucinations.
  • Knowledge cutoff: The model only knows what was in its training data. It doesn’t know about recent events or your internal information on its own.
  • Arithmetic and logic: Language models make mistakes with exact calculations and long chains of logic when they don’t use tools.
  • Data protection: What you enter is processed by the provider. Pay attention to contracts and rules.

How to deal with these limits is shown in AI hallucinations. How an LLM works reliably with your own documents is explained in RAG: using AI with company knowledge.

One model or several?

There are many LLMs from different providers that differ in strengths, speed, context window and price. For simple tasks, a smaller, faster model is often enough. For complex analyses or long documents, a more powerful one pays off. Because the market is evolving quickly, it makes sense not to be tied to a single model. The models we offer are listed on our Models page.

Frequently asked questions

Is an LLM the same as AI?

No. Artificial intelligence is the umbrella term. LLMs are a particular type of AI model specialized in language. Other AI methods recognize images or predict measurements, for example.

Does an LLM know what is happening now?

Only to a limited extent. A model’s knowledge ends with its training data. It can only use current information if a tool provides it, for example through a web search or your own documents.

Does the model learn from my inputs?

That depends on the provider. With many business solutions, inputs are contractually excluded from training. Check this point in the contracts and settings before use.

Why does an LLM give different answers to the same question?

Because it uses probabilities when generating text and usually doesn’t always pick the most likely next piece. That makes answers more varied, but also less predictable.

Sources

  1. Attention Is All You Need (Vaswani et al., 2017) arXiv
  2. Language Models are Few-Shot Learners (Brown et al., 2020) arXiv
  3. Generative AI Models – Opportunities and Risks for Industry and Authorities German Federal Office for Information Security (BSI)

Written by the Lurus Editorial Team according to our source standard. Editorial policy

Planning to introduce AI?

We’ll discuss your tasks and the right starting point – free of charge.

Request a free consultation

Free initial consultation

Your AI start, planned together.

Show us your tasks. We’ll discuss data paths, rules and the right starting point for your team – no feature tour.

The initial consultation is free of charge. A phone number is not required.