What is a Large Language Model?
The one core idea behind ChatGPT and friends — fluent pattern prediction — explained without math or hype.
It’s a pattern engine, not a database
The single most important thing to understand: a large language model (LLM) doesn’t search for answers. It predicts what comes next — token by token — based on patterns absorbed from a huge amount of text.
Think of an autocomplete that read every book, conversation, and codebase on the internet, then got extremely, extremely good at the guessing game.
A small vocabulary: the “words” it actually predicts are called tokens — chunks like
un,believ,able. Roughly three quarters of a word.
Why it’s so fluent (and sometimes wrong)
Because it’s optimizing for plausible, not true. A confident, smooth sentence is exactly what “the most likely continuation” produces — even when the facts are invented. That’s why AI can “hallucinate”: not because it’s lying, but because fluency and accuracy are different skills, and LLMs were trained to be great at the first one.
The three words you’ll hear most
| Word | Plain English |
|---|---|
| Token | The small chunk of text the model actually works with |
| Context window | The model’s short-term memory — what it can “see” at once |
| Inference | The act of running the model to produce an answer |
See it in action
When you type a prompt, the model turns it into tokens, then predicts forward:
Input: "The capital of France is"
Model: "...Paris."
There’s no magic in that small example: the model has seen the pattern “capital of France is …” millions of times, and “Paris” is what statistically follows. Scale that up millions of times and you get ChatGPT.
Where does the brain actually live?
You can run the same model two ways: in the cloud, where your prompt travels to a data center — or on your own machine, where a copy of the model lives in your RAM.
- ☁️ Cloud — the easiest start, the biggest models, pay-per-use, and your prompts leave the building.
- 🏠 Local — free to run, private by default, limited by your RAM and your patience.
Both paths are covered hands-on in the tutorials.