What are Custom GPTs?
You can tailor the existing ChatGPT to deliver the end result you need. In many ways, you can say that GPTs are the next step in plugin evolution.
I am excited to follow the development of OpenAI. Features are getting better, cheaper and more available. Almost seems to good to be true.
We are at a point where the 20$/month subscription is a no-brainer. And people are catching up to this as well - just recently OpenAI had to stop new ChatGPT Plus subscriptions due to overload of demand.
This last onslaught of hype has been brought by Custom GPTs. You can read the official announcement here:
OpenAI Introducing Custom GPTs
Quoting OpenAI themselves:
You can now create custom versions of ChatGPT that combine instructions, extra knowledge, and any combination of skills.
Or in simpler terms:
You can tailor the existing ChatGPT to deliver the end result you need.
In general, this is not entirely new. ChatGPT Plugins have existed for a while now, and in many ways are essentially built for the same purpose. GPTs are the next step in plugin evolution.
The revolutionary key of custom GPTs is that now it is available to the regular ChatGPT Plus users. You no longer need to be an expert at coding (although it helps) to build your own custom GPT (plugin). You can just feed in the information to the GPT builder and it will already start working better for you.
If you are still unsure what I am talking about, you can take a look at some of the examples we have gathered in our lists:
One thing you may notice in the list, these are all public. You can share your custom GPTs with others. There are 2 options to share - make them available with links only (in case you want to share with friends only), or make them public. Public means they will be indexed by Google and also eventually end up in OpenAI GPT marketplace (once it is lauched).
The marketpace is supposed to come out in November 2023. We are yet to see how it will look but it will be a huge step for creativity in GPTs. OpenAI is also indicating that verified creators in the community will be able to earn money based on how many people would use their GPTs.
Of course, you can also make the GPTs private - just for your own use. This will be useful for companies and individuals that have specific use-cases. For example, one idea that makes me excited is building a custom GPT to improve employee onboarding. Instead of having to read or watch the usual piles of documents, a custom GPT could make it simple to create interactive onboarding courses, knowledge base and similar tools to jump start the employee experience in the company. I am sure custom GPTs for business will bring real productivity increase.
Another exciting feature of GPTs is that you can connect your GPTs to the real world. This can be done by using GPT Actions, which is essentially the same as plugins. You can connect other services via APIs and access databases and email features. This is really the revolutionary part, because by accessing external features you make your GPT the only platform you will ever need. We are not quite there yet, but the future is bright.
At the moment it still requires quite a bit of coding experience to create a custom action. However I expect the process will become easier as new versions are released. We have an article on creating GPT Actions in case you will need to add an external service. You can see the GPT configuration form in the below image:
In addition to Actions, you can also upload files that will serve as Knowledge base for your GPT. Using the previous example of Employee Onboarding system, this is where you might upload some internal knowledge base data. Internal rules, office etiquette or coding standards used in the company.
Next we have Capabilities. Capabilities allow you to define which features of ChatGPT Plus should be available to your GPT. At the moment this includes:
- Web Browsing - which allows you GPT to search using Bing. In case you need recent data, or real time data from the web, you got it.
- DALL-E Image Generator - this allows you GPT to generate images using OpenAIs DALL-E. It is perfect for GPTs that need visual output.
- Code Interpreter - This is a fairly new feature released only a few weeks back. For now, code interpreter can write and execute Python code in a sandboxed execution environment. You can write code, do math, etc... This is a huge step for OpenAI, as it has been notoriously known for providing code that does not work. It is still in early ages, and only for python, but being able to execute code before returning is / and will be ground breaking.
Next you can also define the Conversation Starters. Those are the questions, that appear next to message box. Those give you the opportunity to showcase how to use your GPT to the first time user. Not amazing, but a nice to have thing.
Sorry, let me back up a bit. I started with the features that got me going. But there are also the basic, common sense configuration fields - name, description and instructions. Name and description is what users will see when they find your GPT, make sure that those are descriptive and inviting.
Instructions, provide the Instructions for the GPT. Here you have to explain the essential purpose and functionality of the GPT. What it is, what it does, how is it supposed to answer the questions. What information can it reveal and how it should use the capabilities and actions.
You might have noticed the Plus mark on top of the configuration. This allows you to upload a feature image for your GPT. You can also use DALL-E to generate the image based on your description.
That pretty much sums it up for the configuration options. There is a lot more to dive in for each of the advanced features, so stay tuned for further articles.
One last thing I want to mention is that most of the configuration can be done by simply talking to the GPT Builder.
GPT Builder is pretty good at asking questions and gathering information for general purpose for the GPT. The only thing I would (definitively) suggest doing manually is configuring Actions, since GPT Builder has trouble with understanding the schema and purpose of the action. It is also a good idea to manually review the Instructions, to ensure that GPT has understood it's purpose.
