Before the craft, the honest introduction. A chat box answered your question in six seconds — this chapter says what actually did the work: a model on a provider's computers, reading your text in tokens, holding one finite window of context, assembling every answer fresh with a deliberate touch of chance, and knowing nothing after its cutoff. Five facts, and half of the model's personality stops being mysterious.
Most people meet these tools the way Tessa does on the first page — a colleague's tip, a chat box, a surprisingly good answer — and then use them for months on guesswork and superstition. The guesswork is unnecessary. Behind every chat box sits the same small set of facts, and they are not hard; they have simply never been laid out in order for someone who does not code. That is this chapter's whole job.
Five topics, each one fact made solid. What the chat box actually is — a window, with the model elsewhere. Tokens — the chunks the model reads and the unit everything is measured in. The context window — the one working memory, and why long chats forget their own beginnings. Why the same question gets different answers — assembly with controlled chance, not lookup. And the three honest limits — the knowledge cutoff, no memory between chats, no internet by default — that mark exactly where confident answers deserve the least trust.
What you see, and what is there
What you see
A chat box, a blinking cursor, and a good draft six seconds after you press Enter.
What is there
A model on a provider's computers, reading your words as tokens inside one finite window, assembling every answer fresh and knowing nothing after its cutoff.