First experience creating a Custom GPT: "Garden Genius"

A couple of days ago I created a data journalist assistant using the new OpenAI Assistants API. But I've been excited to try the new Custom GPT functionality because I think it's going to be quite disruptive. After waiting for access and enduring some technical hiccups, I finally got a chance. I'll walk you through the process, which I just went through to create a new "Garden Genius" custom GPT that helps people do landscape and garden design.

Step 1: Click to create a new Custom GPT

In the top left corner of ChatGPT, you'll see this "Explore" button:

When you click that, you'll come to a page with this "Create a GPT" button:

Click that button and you'll start the process.

Step 2: Tell GPT Builder what to build

After clicking "Create a GPT," you come to a page where you can have a conversation with the GPT Builder:

Just tell it what you're trying to build. It will suggest a name, generate an icon, and iterate with you as needed.

Step 3: Preview, edit, and improve

For some people, that will literally be it. For simple use cases, GPT Builder will walk you through everything, and you can probably just publish the result.

But I think often you'll want to test the GPT and edit configuration settings to improve it. You can do this using the "Preview" conversation on the right and the fields on the "Configure" tab, as shown here:

I'm not going to go into detail on all the configuration options in this post, but I will say that there's sufficient functionality here to build pretty complex assistants.

Step 4: Share privately, publicly, or not at all

When you're done, you can save your GPT and make it available to yourself only, to people who have the link, or publicly. Enterprise users, as I understand it, can also choose to share their GPTs only with people in their company. It looks like this:

It also looks like OpenAI will provide some degree of information about usage, at least based on the chat icon with a zero beside it here:

And that's it!

It is incredibly simple. This entire process took less than 20 minutes, and it would have been even faster if I wasn't taking screenshots along the way.

Next up for me: Experimenting with features like uploading knowledge and adding API-driven actions.