Demonstrating Operator

I'm a research lead of Operator in OpenAI. And what is Operator? Operator is a research preview of an agent that uses browsers to, help user to do things.

So I have a two year old kid who likes pasta. So I, make linguini with clams, so I ask it to buy the groceries for it. So I will use the Instacart app.

Operator can actually basically use any website without. . .

And it is not particularly optimized for Instacart. But the reason why I’m using this app is that it provides the, But the reason why I’m using this app is that it provides the, detailed instruction of how this website can be best utilized, just like the tutorial that humans can benefit from. So I'll use Instacart tab and ask it to solve tasks could you find a recipe of linguine with clams, from Allrecipes website and add all the ingredients to the grocery cart, or Instacart.

I think I already have, some ingredients like butter vegetable oil, and black pepper. So you don't need to add them to the cart. So it says there I'll find recipe and then add everything in the, ingredients to the cart.

Okay, it says that, it’ll confirm the ingredients and store with me before adding them to the cart. So let's start by, finding the recipe. I am not doing anything, from now on.

Operator is just doing and I am just watching what it’s doing. What is interesting with Operator is that its, it is using browser that I built for the human, and it is using the, seeing the exactly same script that I am seeing right now and it is using the, seeing the exactly same script that I am seeing right now and using the keyboard typing and mouse clicking to control the browser. Just like a human would do.

This is different from other agents that uses, API or programing based interface, which programmers might understand, API or programing based interface, which programmers might understand, but no programmer users cannot understand it really well. So Operator, because it is using the, this natural human interface it’s very easy to follow by just looking at what it's doing in the screen. Can you follow its progress?

Yes! So one way to follow its progress is I can zoom in to see the screen better. And Operator is powered by the, text based, chain of thought reasonings.

And Operator is powered by the, text based, chain of thought reasonings. So whenever it is doing things, it, says, it makes plans of how things can be done. And this can be followed through this, list of the tasks.

And it says I found the recipe. And, which store would you prefer to use? So I ask, use Gus’s.

So often it asks clarifying question whenever it is needed, it has clarifying question whenever it is needed, is needed in the process of solving the task. There are cases the Operator has to, make sensitive actions like things like logging in or buying things. In this case, we built Operator to be safe in this situation.

So Operator is designed to ask, us to take control, So Operator is designed to ask, us to take control, to log in by ourselves or whenever it needed, checking it. It gives us the control so that I can double check or whether the list is correct and then checking it by myself. Amazing.

Thank you so much. I appreciate you showing it to us. Amazing.

Thank you so much. I appreciate you showing it to us. Thank you so much.