An LLM – a large language model – is an AI model trained on enormous amounts of text to predict the most probable next word, which is why it can express itself fluently on almost anything.
The most important thing to understand is what the model has learned. It has not read the world – it has read text about the world. It does not know what is true; it knows what sentences about a subject tend to look like.
That is why a language model writes just as convincingly when it is wrong as when it is right. The confidence of the language says something about the training material – not about the facts. It is not a flaw in your prompt. It is how the technology works.
In practice
Ask a language model for a draft press release, and within seconds you get a text with the right structure, a credible tone and a quote that sounds like a chief executive. But dates, figures and quotes may be invented – phrased just as fluently as the rest. The draft is a gift; the fact-check is yours.
How to explain it to senior management
“A language model is trained to write convincingly – not to be right. That is why everything it writes has to pass a human who knows the facts.”
If you want to dig a little deeper, I have written a longer introduction to what artificial intelligence actually is.