Table of Contents
ChatGPT is one of the most popular AI chatbots in the world, powered by OpenAI’s cutting-edge technology. But what if you could improve on it? What if you could customize it to suit your needs and preferences? That’s exactly what custom GPTs allow you to do.
Custom GPTs are a new feature that OpenAI launched last week, which gives users more control and flexibility over how ChatGPT behaves and what it can do. With these, you can modify the chatbot’s communication style, provide specific data sources, and even integrate it with other services and APIs. But is this feature a game changer for generative AI, as some AI enthusiasts claim? Or is it just a gimmick that won’t make much difference?
What are custom GPTs and how do they function?
GPT stands for “generative pre-trained transformer”, which is a type of AI model that can generate text based on a large corpus of data, often scraped from the internet. ChatGPT uses GPT-4, the latest and most advanced version of this model, which can also analyze images and work with other OpenAI tools, such as DALL-E 3, an image generator.
ChatGPT is designed to be a conversational AI that can chat with you about anything, from trivia to philosophy. However, until now, if you wanted ChatGPT to have a specific personality or knowledge base, you had to persuade it to adopt a role or upload your data. And if you wanted to share your customized version of ChatGPT with others, you had to tell them the exact inputs you used and hope they got the same results.
Custom GPTs simplify this process by letting you create and save your versions of ChatGPT, which you can then share with others or use for your purposes. You can create a custom GPT by clicking on the “Explore” option in the upper-left corner of ChatGPT, where you will find some official custom GPTs made by OpenAI, as well as a chat interface where you can build your own. You can either describe what you want your custom GPT to do in natural language, or enter more specific parameters manually.
What can you do with custom GPTs?
The possibilities of custom GPTs are limited only by your imagination and the data you provide. You can create these for various purposes, such as entertainment, education, productivity, or creativity. For example, some of the official custom GPTs made by OpenAI include:
- A coloring book maker, which generates coloring pages based on your requests.
- A tech support assistant, which answers your questions about common tech issues.
- A Genz 4 meme chatbot, which talks like a Gen Z internet user and gives you advice on how to be cool.
These are just some examples of what custom GPTs can do, but there are many more that users and developers have created and shared on platforms like X and Reddit. You can find it that can help you with personal finance, weather forecasts, music production, graphic design, web development, and more.
To help you discover new custom GPTs, some people have created their ways of curating and showcasing them. For instance, Rowan Cheung, the author of a popular AI newsletter called The Rundown AI, has created a custom GPT that acts as an app store for ChatGPT, where you can browse and try different custom GPTs.
Cheung says that custom GPTs are the first version of an app store for ChatGPT and that anyone can create their apps with no code and natural language. However, he also warns that custom GPTs are not perfect and have some limitations.
What are the challenges and limitations of custom GPTs?
Custom GPTs are a new and exciting feature, but they are not without flaws and challenges. Cheung compares ChatGPT to the first iPhone and to the launch of the App Store, which was also very limited at first.
One of the main challenges of custom GPTs is that they are still prone to errors and hallucinations, meaning that they can generate inaccurate or nonsensical text. This is because GPT models are not very good at verifying the facts or the logic of their outputs, and they rely on the data they are given, which may be incomplete or biased. Therefore, these should not be used for critical or sensitive tasks, such as legal or financial advice.
Another challenge of custom GPTs is that they are still slow and buggy, especially when they interact with other services and APIs. APIs are application programming interfaces, which are pieces of code that allow different programs to communicate and exchange data. Custom GPTs can use APIs to access and manipulate other services, such as Canva or Replit, which can enhance their functionality and usefulness.
However, using APIs also adds complexity and latency to custom GPTs, which can affect their performance and reliability. For example, I tried a custom GPT that used Canva to create an image for this article, and it took several minutes to generate a rather messy image of a robot. I also tried a custom GPT called DesignerGPT, which used Replit to create a website, and it also took a long time to produce a mediocre result. I asked it to make a retro website, but it did not seem to understand what that meant.
Pietro Schirano, the AI developer behind DesignerGPT, says that he was not impressed by OpenAI’s previous attempt to integrate ChatGPT with other services, which were called plugins. Plugins were launched last March, but they did not get much attention or traction, because they were not very user-friendly or monetizable. Schirano says that custom GPTs are a better way to connect ChatGPT with other services, but they still need improvement.
What is the future of custom GPTs?
Custom GPTs are a promising feature that opens up new possibilities for ChatGPT and generative AI. They give users more control and flexibility over how ChatGPT works and what it can do, and they enable developers to create and share their own AI apps with no code and natural language. It also has the potential to connect ChatGPT with other services and APIs, which can increase its functionality and usefulness.
However, custom GPTs are also a new and experimental feature, which means that they are not perfect and have some limitations. They can still generate errors and hallucinations, which can affect their accuracy and reliability. They can also be slow and buggy, especially when they work with other services and APIs, which can affect their performance and usability. They also depend on the quality and availability of the data they are given, which may be incomplete or biased.
Therefore, custom GPTs should be used with caution and awareness, and not for critical or sensitive tasks. They should also be evaluated and improved constantly, and not taken for granted. These are great features, but they are not a magic solution.
The future of custom GPTs may depend on how OpenAI and the ChatGPT community continue to innovate and collaborate on the platform, and how they address the challenges and limitations of the feature. Schirano says that he is cautiously optimistic about custom GPTs and that he hopes that OpenAI will not screw up. He also says that developers need to start building things that people want to use.
Custom GPTs are a new way to customize ChatGPT, and they have a lot of potential. But they also have a lot of room for improvement. Custom GPTs are not a game changer yet, but they could be one day.
