Is AI a Student or an Actor

We already talked about the prompt. This is what you normally type in the chat box of an AI assistant. Now let's look at the answer which the AI assistant generates: the completion. We tend to think that the AI is like a student taking a test. You ask a question, it keeps quiet, thinks, checks its memory and then writes the answer. But that's not exactly what's happening. The AI is not a student. It's more like an improvisational actor. In improvisational acting, the number one rule is to say 'yes, and': you take the situation and you build on it.

You never stop the scene. And now a reality check. If you try a modern model like GPT-4 or later and ask something about an invented person like Elias P. Wiggle, it might look you in the eyes and say this person doesn't exist. Has the actor been replaced by a database? No. The actor has just learned a new role. It used to play the storyteller who makes things up, and now it plays the smart assistant who checks facts. When you ask about the fake person, the model scans its memory.

It doesn't find a match. So the 'yes, and' changes direction. Instead of saying 'yes, and he invented the toaster', it says 'yes, I hear your question, and I must correct you, because my internal pattern says that this is false.' It is still performing. It is still completing the pattern. But the pattern it follows now is: be helpful and truthful. The technology is improving really fast. Hallucinations are getting rarer because the models are better at playing the role of a skeptic. So the next time you see a perfect answer, remember it's not magic.

It's a world-class actor who has finally learned to say, 'I don't know.' But the core idea stays exactly the same. The model is still completing the pattern and finding the next most probable and most likely token or word. This was a note on AI for today. Thank you for watching and I'll see you in the next one. Bye.