01 Lesson 1 of 15
What you are actually looking at
Some of what you see in a chatbot is the machine. Most of it is decoration. This lesson shows you which is which.
Something in there is thinking about my question. You type. There is a pause. Then words appear one by one, as if someone is writing them. The design invites you to picture someone there. The pause is real. The someone is a design choice.
The model produces text. Everything else was chosen by a design team, to make the chat feel natural.
See it in the lessonA chatbot has two parts. The first is a language model. The second is everything around it: the chat bubbles, the typing animation, the word “I”, the apology when you push back. The model produces text. Everything else was chosen by a design team, to make the chat feel natural.
That is fine. A friendly design is a sensible choice. But it means the chat window cannot tell you what is really going on. “I think” comes from a style guide, not a thought. “Sorry” does not mean anything was sorry. One part is real: the words do arrive in pieces, because that is how the text is made.
You cannot break it
Nothing you type breaks anything. There are no wrong questions and no limit on tries. Nobody reviews your clumsy attempts. If a chat goes badly, start a new one. That is a normal move, not a failure.
People who feel stuck with these tools are rarely short of brains or technical skill. What they lack is a picture of how the thing works, so they cannot tell when it will help and when it will mislead. This site builds that picture, one short lesson at a time. The whole course takes about forty minutes, and you can stop after any lesson with something useful.
The diagram below takes a chat window apart, one layer at a time. Peel the layers off and see what is left in the middle.
Read this diagram as text
From the outside in, the layers are:
Under all of them is the language model: text goes in, likely text comes out. Peel a layer to remove it, and the line under the diagram says what that layer was doing.
One analogy. It is closer to a tool than to a person. A camera does not know what a good photo is. You do, and your photos get better as you learn how the camera works.
Where the analogy breaks: a camera does not talk back in full sentences. That is exactly why this one is harder to see as a tool.
Open any chatbot you can use and type something with no goal at all. Ask it what it is. Ask it something silly. You are not trying to get anything done yet. You are just getting used to the empty box.