# What Is Large Language Model (LLM)? A Plain-English Definition

> A large language model (LLM) is a model trained on huge amounts of text that predicts the most likely continuation of whatever you give it. It can draft, summarise and answer questions, but it does not act on its own. Acting requires an AI agent with tools.

Canonical URL: https://www.ruebstudio.de/en/glossary/large-language-model-llm  
Author: Dr.-Ing. Marcus Rüb, CEO  
Published: 2026-10-21

A large language model (LLM) is a language model with a very large number of parameters, trained on vast amounts of text to predict what comes next. That single skill is what lets it draft emails, summarise reports, translate and answer questions. In a business setting, the LLM is the component that assistants, chatbots and [AI agents](https://www.ruebstudio.de/en/ai-agents) are built on.

## How does a large language model work?

An LLM splits text into small units called tokens and generates its answer one token at a time, each time picking a likely next token. It learns this during pretraining on large text collections, and is then shaped into a useful assistant through fine-tuning and human feedback.

In practice there are three ingredients:

- **Pretraining:** [Anthropic's glossary](https://platform.claude.com/docs/en/about-claude/glossary) describes models being trained to predict the next word given the preceding text. A model at this stage is not yet good at following instructions.
- **Fine-tuning and RLHF:** people rank example answers, and reinforcement learning nudges the model towards the higher-ranked ones. This is what turns a text predictor into something that behaves like an assistant.
- **Architecture:** most current LLMs are based on the Transformer, introduced in 2017 in [Attention Is All You Need](https://arxiv.org/abs/1706.03762). Scale matters too: GPT-3, described in 2020, had 175 billion parameters ([Brown et al.](https://arxiv.org/abs/2005.14165)).

What a model can take into account in one request is limited by its context window, a kind of working memory. Anything outside its training data and outside that window, such as your current price list, it simply does not know.

## LLM vs chatbot vs AI agent vs search engine

Think of the LLM as an engine rather than a vehicle. A chatbot is a conversational interface around an LLM. An AI agent gives the model tools and lets it complete multi-step tasks. A search engine retrieves existing documents; an LLM writes new text.

- **LLM:** turns text into text. On its own it cannot open a ticket system or update a CRM record.
- **Chatbot:** a chat front end, often limited to a defined set of topics.
- **[AI agent](https://www.ruebstudio.de/en/glossary/ai-agent):** model plus tools, instructions and memory. It might read a support ticket, look up the manual and save a draft reply.
- **Search engine:** returns sources you can check, but adds nothing of its own.

If you are choosing a specific model for your organisation, start with a product view such as [Claude for business](https://www.ruebstudio.de/en/guides/ai-tools-and-platforms/claude-for-business).

## What should businesses watch out for?

An LLM sounds confident even when it is wrong. These errors, usually called hallucinations, can be reduced but not eliminated. For business processes, that means grounding answers in your own documents, allowing the model to admit uncertainty and keeping a person in the loop for decisions that matter.

[Anthropic's guidance](https://platform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/reduce-hallucinations) suggests explicitly permitting the model to say it does not know, asking it to base answers on direct quotes from supplied documents, and requiring a citation for every claim. In our own operations, giving the model only the documents a task needs works better than relying on its general knowledge.

An LLM is the wrong tool when a task must run exactly the same way every time, such as posting an invoice or calculating a price. Conventional software does that more reliably; at most, the model should call that software as a tool.

## Frequently asked questions

### Are ChatGPT and Claude LLMs?

They are products built on large language models. The model is the engine; the product adds the interface, conversation history, file handling and often tools such as web search.

### Does an LLM know our internal documents?

Not by default. Company knowledge has to be supplied with each request, typically through a retrieval step that passes the relevant passages into the model's context.

### Can we run an LLM on our own servers?

Yes. Open-weight models can run on your own GPU servers, which keeps data in-house. Smaller models are cheaper to operate but usually weaker on complex tasks.

### Why does the same question get different answers?

Generation samples from several likely continuations, and the temperature setting controls how much that varies. Anthropic notes that even at temperature zero, outputs are not fully deterministic.

## Sources

- [Anthropic: Glossary (LLM, pretraining, RLHF, temperature, context window)](https://platform.claude.com/docs/en/about-claude/glossary)
- [Vaswani et al.: Attention Is All You Need (12 June 2017)](https://arxiv.org/abs/1706.03762)
- [Brown et al.: Language Models are Few-Shot Learners (28 May 2020)](https://arxiv.org/abs/2005.14165)
- [Anthropic: Reduce hallucinations](https://platform.claude.com/docs/en/test-and-evaluate/strengthen-guardrails/reduce-hallucinations)
