If you are a SaaS business owner or a developer, you might want to think twice before investing in a ChatGPT plugin or custom GPT with actions. Writing this based on our experience with building Sivi AI Designer plugin on ChatGPT — the Multilingual AI Designer for your brand and content.
Our Experience building the Sivi AI Designer Plugin on ChatGPT
When we developed Sivi’s design generation API, we decided to test it out by creating the Sivi AI Designer plugin for ChatGPT. While the custom GPT with actions served its purpose, it has its own limitations and challenges. Here’s what we learned about creating a custom GPT with actions, what to do and what not to do when building your own custom GPTs with actions.
How the AI Designer plugin on ChatGPT works
The Sivi AI Designer plugin we built for ChatGPT allows users to generate designs via Sivi API for design and displays them directly within the ChatGPT interface. Here's how the AI Designer plugin works:
User Interaction: Users can initiate the design request by typing the prompt into the ChatGPT interface.
Design Generation: The plugin processes the user prompt and forms the design brief. Once the brief is ready, it calls the Sivi API to generate the designs.
Generated Designs Listing: Once the design generation is complete, the designs are shown to the user within ChatGPT. Users can download the designs directly or click on any design to edit it further using the Sivi Design Editor.
Limitations of custom GPTs and actions on ChatGPT
While custom GPTs with actions offer possibilities, they come with their own limitations.
Instructions limitations
Clarity required: Large Language Models (LLMs) naturally take the least resistant path when processing instructions. So, every step in the instructions needs to be crystal clear to avoid misinterpretation.
Character limit: Custom GPT instructions have a 8000 characters limit, making it challenging to include detailed instructions. The instructions need to be concise.
Knowledge files limitations
Single file limitation: Uploading more than one knowledge files leads to error though it doesn’t stop the user from using the GPT.
File formats: Text files work better compared to JSONs or other formats.
Synchronized update: While updating instructions, sometimes it is necessary that we update the knowledge file as well to ensure that the details in the knowledge files are used.
Capabilities and image handling issues
Image generation and handling: Unlike template design tools, Sivi can generate designs for user-provided content and images. So, we initially enabled all the custom GPT capabilities including, DALL-E Image Generation. But, passing the generated image via actions is not possible. So, we did a work around. We instructed our users to right-click on the generated image and copy-paste the image URL. However, images from the URL are available for download only for a limited time. So, we downloaded and added a mapping to our URL. Not just the images generated by Dall-E, images uploaded by the users cannot be passed as well. We also created a prompt engineering guide to assist users with this process.
GPT updates: Then came a GPT update where they stopped allowing custom GPTs to download the generated or uploaded images. So, we disabled DALL-E Image Generation capability and asked user to share the image URLs for their products, services, or other design elements.
Technical limitations
Unreliable Performance: ChatGPT often experiences downtime or unresponsiveness, which can disrupt user experience.
Exploration Errors: The "Explore GPTs" feature frequently throws errors, making it difficult for users to discover new GPTs.
Discoverability challenges
Review System Flaws: ChatGPT Plus users can review custom GPTs with actions without actually using them, leading to potentially skewed ratings.
Lack of Distinction: There’s no clear differentiation between personal GPTs and GPTs with actions.
Discoverability Challenges: Custom GPTs with actions are hidden in the ocean of GPTs, making it difficult for users to find them. Users need to search to discover or use direct link.
No Listing Page: There’s no dedicated listing page for custom GPTs. Only top GPTs are easily discoverable. A limited number of six GPTs get listed on featured GPTs. But, the probability of getting there is quite low.
We either need to attract ChatGPT users or bring our users to ChatGPT.
The Canva GPT case study
Consider Canva GPT as a benchmark. Canva has around 200M users, ChatGPT also has around 200M users, and Sam Altman showcased Canva GPT in the Opening Keynote of OpenAI DevDay. Still Canva GPT has only 3M conversations and a 3 star rating from 50K users. If Canva struggles to make a significant impact, it’s worth thinking through the challenges an upcoming SaaS business might face.
By the way, here’s the comparison between Canva GPT and Sivi plugin. While Canva tries to pick templates based on your prompt, Sivi generates designs from scratch in 72+ languages.
What could ChatGPT do to help SaaS businesses and developers building custom GPT with actions
To better support developers building custom GPTs with actions, ChatGPT could do the following:
Improve Reliability: Reduce downtime and errors to ensure a smoother experience.
Enhance Discoverability: Introduce a listing page for custom GPTs to make them easier to find. And provide a filter for custom GPTs with actions.
Refine the Review System: Ensure that only users who have actively used the plugin can leave reviews.
Refine the Review System: For GPTs that enable Dall-E Image Generation, provide a way to use it via API.
How we overcame the limitations
We ended up building our own in-app chat to overcome all these limitations. Yes, you heard it right.
This allowed us to provide a seamless design experience directly within Sivi, bypassing the issues we encountered with the ChatGPT plugin. More about it on another blog. Meanwhile, you can experience the Sivi in-app chat here and share your thoughts.
Here’s how it works:
Conclusion
Though building the ChatGPT plugin helped us improve our API and build our in-app chat, we faced multiple hiccups along the way. For SaaS business owners, it might be worth exploring other integration options that offer more control, reliability, and flexibility. By understanding the limitations and potential drawbacks, you can make an informed decision about whether building a custom GPT is the right path for your project.
Email us if you’re interested in integrating the Sivi API for design.
——
Author: Sona
Sona is the Co-Founder and Chief Design Officer of Sivi, a generative AI that instantly turns text into business visuals. With a distinguished academic background, Sona started her career as an engineer before transitioning to a designer. Over the last 14+ years, Sona has worked with multiple startups and MNCs in diverse sectors and designed over 35 products. She has evangelized design thinking and formed successful design teams from scratch. Sona is a gold medalist, TiE Women Bangalore Entrepreneurs finalist, MeitY Women Entrepreneurs Award finalist, and an author of a popular design portal.
LinkedIn | Twitter | Instagram
