What is an LLM, really?

Have you ever used autocomplete on your phone? When you type "Happy birth" it suggests "day", because that pair shows up so often in messages. A large language model (LLM) is a much bigger version of that idea. Your phone only peeks at the last few words. The model weighs everything typed so far and guesses the next small piece, called a token. A token is sometimes a whole word and sometimes only part of one. It learned those guesses from an enormous amount of writing: web pages, books, Wikipedia, forums, and more.

Here is the key thing most people never learn: the model itself is not looking your question up in a database, and it does not open Wikipedia while it writes. It adds one token, then another, until it hits a stop signal or a length limit. A product such as ChatGPT can look things up when search is turned on. That lookup is a separate tool. This playground uses the model alone.

That is why an answer can sound sure and still describe something that never happened. The model is not lying. It was trained to continue text the way people write, and people often sound sure. It has no built-in step that checks whether the sentence it just wrote is true.

This playground is for anyone who uses AI but has never seen the engine. You will move the pieces yourself, no coding, no math, no prior knowledge needed.

By the end, you will be able to explain what a token is, what temperature does, and most importantly: why a smooth, fluent answer is never automatically true.