Introduction
What you are actually talking to
ChatGPT, Claude, Gemini and Copilot are different names on the front of the same kind of machine. Before we take that machine apart, it is worth being precise about what it is and what the rest of this course will show you.
The one-sentence version
A large language model is a system that predicts the next fragment of text, and everything it appears to do is that one operation, repeated very fast.
That one-sentence version sounds like it must be an oversimplification, and it genuinely is not. When a model answers a question, translates a document, writes code or talks you through a decision, it is running the same loop: look at all the text so far, score every possible next fragment, pick one, add it to the end, and go again. The rest of this course is that loop, slowed down.
Five stops, and the whole thing runs again for every fragment. Each stop draws what actually happens at it.
Five stops, and the whole thing runs again for every fragment. Nothing here is a box with a stage name in it — each stop draws what happens at it.
Like your phone's autocomplete, taken absurdly far
Your phone suggests the next word from the last two or three. Now imagine it reads the entire conversation, has absorbed a significant fraction of everything ever written down, and re-runs that judgement for every single word it produces. The mechanism is not more sophisticated than autocomplete. The scale is what changes the result from a party trick into something you can work with.
When a model writes a five-paragraph answer, when does it decide how the answer will end?