OpenAI has just released it’s latest updates, and while I’m still not convinced about the future of chatbots in education, I can see some immediate commercial appeal and use cases for information retrieval. This platform is going to make OpenAI a lot of money, and hopefully they’ll use it to do something more interesting with this incredibly powerful technology.
Here’s my quick experiment setting up a GPT based on my book, Practical Reading Strategies. As you’ll see, it’s ludicrously straightforward. Prepare yourselves for GPT-everything in a “there’s an app for that” fashion: as long as you have the cash to burn on a subscription to ChatGPT Plus…
Creating a GPT in GPT Builder
The first step to create a new GPT is to go to the Explore tab on the left hand side and then click Create a GPT. Note that at the time of writing, this feature is still rolling out, and it’s only for Plus subscribers.
Next, you can begin “programming” the GPT by giving natural language instructions in the left window, and testing it in the right. The default first few prompts ask you to describe your bot, then it suggests a name (which you can change) and generates a profile picture using DALL-E 3.
Once you’ve provided some basic instructions, it will continue prompting for things like the tone of voice, style, and how it should respond. Eventually, you’ll end up with a GPT that you can test out and continue to refine with further instructions and prompts.
In the configure tab on the right hand side, you can adjust parameters further, including changing the default starter prompts (the four in the right hand side above), and adding contextual documents.
In this case, I’ve added a PDF version of my book and the instructions to only refer to the book when giving answers, and to reject all other questions. In the configure window, you can see how that has been interpreted and turned into instructions:
Once you’re finished, you can make the GPT public and share the link:
And that is it… As I said before, ludicrously simple.
It’s easy to see why this latest effort from OpenAI has a few tech startups on the ropes. These GPTs can use internet access, DALL-E 3 image generation, and advanced data analysis to read and write code, all while being tethered to specific documents.
There is also an option to use actions which function in a similar fashion to the earlier ChatGPT plugins, connecting the GPT to things like Canva, weather apps, or other integrations.
Does it work?
Here’s a very quick video demonstrating how it performs using my book as the basis. I know this book pretty well, and it’s responses are accurate and reasonably helpful. Because I’m a shameless self-promoter, I’ve also added the instruction to place a link to my website and a suggestion to buy the book at the end of every response.
This is part chatbot, part helpdesk, and part marketing platform…
If you want to chat about GenAI, get in touch using the form below: