An LLM is a model trained on very large amounts of text that predicts the most likely continuation of your input. ChatGPT, Claude and Gemini are applications built on top of such models.
The abbreviation stands for large language model. The core is less spectacular than the term suggests: the model continues text.
Everything that looks impressive afterwards — summarising, translating, programming, arguing — follows from that prediction having become remarkably good across enormous amounts of text.
What an LLM does well — and what it doesn't #
Strong at: rephrasing, summarising, structuring, translating, drafting, spotting patterns in text, writing code.
Weak at: exact individual facts from memory, events after its training cut-off, arithmetic without a tool, and judging whether it actually knows something. Hallucination follows directly from that last point.
For any task, ask: does it need language or knowledge? Language can largely be delegated. Knowledge you have to supply — through connected sources or tools.
Use in a company #
An LLM on its own is rarely the solution. It becomes useful in combination: with your documents (RAG), with tools and system access (AI agent), with clear instructions (prompt engineering).
And with one rule that often gets forgotten: where an LLM speaks directly with people, the EU AI Act has required since 2 August 2026 that those people be told they are talking to an AI.
Common questions #
Is ChatGPT an LLM?
ChatGPT is an application built on an LLM. The model is the engine; the application is the car around it — interface, memory, tools and safety rules.
What does "large" mean?
It refers to the volume of training data and the number of parameters, the internally learned settings. Bigger is not automatically better — many tasks are served well by smaller, faster, cheaper models.
Does an LLM know what happened today?
Not by itself. It knows its training cut-off. It only knows current events if the application gives it access to search or documents.
Does the model learn from my inputs?
Not automatically, and in business contexts it depends on the provider and plan. Check this explicitly before confidential content goes in.
Sources #
- Regulation (EU) 2024/1689 (AI Act) — Chapter V on general-purpose AI models; Art. 50(1) on disclosure
- Anthropic — Models overview
- OpenAI — Models documentation
Which model for which task?
Ask me — or ask Ada. She is my AI assistant, available around the clock, and will propose a slot with me if you need one. Free, no form.