You have used one — ChatGPT, Claude, Gemini, or the AI feature inside a product at work. This guide explains what a large language model actually is, without requiring any math or programming.
The core idea: predict the next word
An LLM is a program trained on a huge portion of the internet’s text with one objective: given some text, predict what comes next. Ask it to complete “The capital of France is” and it predicts “Paris” — not because it looked up a fact, but because that continuation is overwhelmingly likely in its training data.
The surprise of the last few years is that this simple objective, at enormous scale, produces abilities nobody explicitly programmed: translation, summarization, coding, reasoning through problems step by step.
Why they get things wrong
Because an LLM generates rather than retrieves, it can produce text that sounds authoritative and is simply false — a failure called hallucination. The model has no built-in “I don’t know” reflex; it always produces its best-guess continuation. This is why serious applications pair LLMs with search or documents (so answers are grounded in real sources) and why you should verify anything important.
Four terms that unlock everything
- Token — the chunk of text a model processes at once (roughly ¾ of a word). Pricing and limits are measured in tokens.
- Context window — how much text the model can consider at once: its working memory.
- Parameters — the internal dials set during training; “bigger model” usually means more parameters.
- Fine-tuning — additional training that specializes a general model for a task or style.
Where to go next
Try the same prompt in two different assistants and compare. Then read our profiles of tools like Ollama to run a small model on your own laptop — nothing demystifies LLMs faster.