How do you create a GPT?

We’re all familiar with ChatGPT, but now you can create a GPT – a custom version of the all-encompassing chatbot that performs specific tasks or behaves precisely like you want it to. Like the sound of that? Then let’s explore how to build your own GPT. Best of all, it doesn’t need a single line of code.

Even better news? A ChatGPT “app store” is coming soon, so you can sell your creations.

How to create a GPT

To make your own GPTs, you will first need a ChatGPT Plus account. You can sign up for one from the ChatGPT website, and it currently costs $20 per month.

Once you have access to the premium account, hit the Explore button in the top-left corner of the ChatGPT interface. This gives you the option to create a GPT or play with examples created by OpenAI.

Example GPTs include The Negotiator, a bot that promises to help improve your negotiation skills; Laundry Buddy, which gives you tips on how to get specific stains out of washing; and Coloring Book Hero, a bot that takes your ideas and turns them into colouring-book pages.

That last example is a key reminder that ChatGPT doesn’t only output words. It can also spit out images, code, spreadsheets and other forms of data.

Training your own GPT

Deciding on the type of GPT you want to create is the hard part. Training the bot to do exactly what you want is relatively simple, largely because the creation process is chatbot-based, with ChatGPT asking you precisely what you want it to do and not do.

You’ll be asked a series of questions, although in my experience, this process can be a bit fraught. You’ve not finished answering one of the questions about how you want the bot to behave, before ChatGPT is firing another at you. Don’t get flustered. Keep answering all the questions one by one and you’ll eventually get the desired result.

In my tests, I created Tech Simplifier – a bot designed to explain complicated tech jargon in Plain English. If you’re a ChatGPT Plus subscriber, you can click this Tech Simplifier link and see how it works.

ChatGPT will suggest a name for your new GPT and help you design the logo, but you can make suggestions if you don’t like what it comes up with.

The key part is to test your new GPT’s abilities in the right-hand ‘playground’ pane as you go along, so you can see how it’s behaving. To test my Tech Simplifier GPT, for example, I asked it to explain what FTP (file transfer protocol) was. It came back with an explanation that was still far too complicated and stuffed with jargon. Jargon such as client-server protocols.

When I asked the GPT to explain what client-server meant in more friendly language, it came back with a much clearer explanation, comparing a server to a library and the client as someone looking to borrow a book. I told ChatGPT in the left-hand configuration pane that this is the kind of answer I wanted to see much more of from this GPT, and it adapted accordingly, making subsequent answers far easier to understand. Providing your own example answers can also help point ChatGPT in the right direction.

Sharing your GPT

Once you’ve got the GPT behaving how you want it to, you have the option to keep it to yourself, make it usable by any ChatGPT subscriber with the link, or make it available to the public. OpenAI will soon open a store where you might even be able to make money from your well-trained bot.

You can keep refining the GPT’s skills with the Edit option, so even if you’ve shared it with the world, it isn’t too late to adjust its behaviour.

Avatar photo
Barry Collins

Barry has 20 years of experience working on national newspapers, websites and magazines. He was editor of PC Pro and is co-editor and co-owner of BigTechQuestion.com. He has published a number of articles on TechFinitive covering data, innovation and cybersecurity.

NEXT UP

Fully Homomorphic Encryption (FHE) EXPLAINED

Fully Homomorphic Encryption (FHE) explained

From Caesar’s cypher to Fully Homomorphic Encryption (FHE) – Jeremy Bradley, COO, Zama, explains, in this sponsored article, exactly what FHE is, how it has evolved, what it is now capable of and how far off truly universalised FHE is.