Prompt, context, token: the words of usage.

You'll encounter these as soon as you start using one seriously.

A prompt

Simply what you write to ask it for something. The English word stuck because there's no short translation; "request" or "instruction" say the same thing. When people talk about "prompt engineering," it's just the art of phrasing your request well.

The context

Everything it has "in view" when responding: your question, but also the previous messages in the conversation and any documents you've pasted. This is its working memory, and it has a limited size.

In a very long conversation, the beginning eventually falls out of context: it doesn't forget it out of distraction, it literally no longer sees it. Hence the advice to open a new conversation per topic.

A token

The unit it manipulates: a small piece of a word. "Hello" might be worth one token, "antidisestablishmentarianism" several. It doesn't read letters; it reads chunks.

This is a useful quirk to know about: it explains why it can make mistakes counting the letters in a word. It doesn't see them.

The temperature

A setting that decides whether it always picks the most likely word or takes risks. Low, it's predictable and repetitive. High, it's inventive and sometimes incoherent. On consumer-facing interfaces, you don't have access to it.