30 Apr 2023
Speed up and improve your workshops through the use of WorkshopGPT (#13)
Motivation
Everyone is talking about AI, ChatGPT, and prompts. Integrations and plug-ins are a little quieter, although work is certainly in progress in many areas. And sometimes it doesn't take much effort to combine great tools via a plugin.
*Note: Due to the rapid development, some flaws are no longer present. (31.12.2024)
Challenge
As a retrospective facilitator, it is essential for me to evaluate and observe a variety of perspectives and thoughts. To be able to process these ideas better, I reformulate them with the team into "How could we" (HMW). This leads to more effective problem-solving and the development of innovative, practical solutions that address real team needs.
This step of reformulation takes time and needs practice. Could AI make this hurdle a thing of the past or significantly improve it overall? Let’s give it a try.
Solution approach
Get ChatGPT integrated into FigJam via a plugin, so you can work with the content created, as the content written down on post-its, without leaving FigJam. This saves time and everything stays in one place.
How does it work?
After everyone presents their post-its, they are grouped and organized, with the plugin rephrasing the topics into "How might we" questions. The content is processed through a developed and refined prompt. Additionally, the "Rephrase" and "Abbreviate" buttons apply prompts to the generated HMWs, ensuring that everyone can create viable and well-structured questions. Ultimately, it’s important to review and validate the output. Think of the plugin as a collaborative companion designed to help you brainstorm and develop concepts without starting from scratch. While its primary purpose isn’t to solve your problems, it accelerates your progress and serves as your trusted “rubber duck” for creative thinking.
Current flaws
It takes a few seconds to create an HMW. However, the current APIs also do not support the familiar typewriter effect seen in ChatGPT, which would enhances interactions by making them feel more conversational and dynamic.
Prompt Misalignment: The 3.5 model doesn't consistently write the HMW in the language of the input and often defaults to the language of the prompt instead. The 4 model could handle it, but generating the HMW often took too long, sometimes exceeding 10 seconds.
Next possible steps
The plugin layout takes into account future functions that can be added. What features will follow remains to be seen. Due to the current shortcomings, I will wait for further improvements from the ChatGPT API.
*Update 31.12.2024
As mentioned at the beginning, a lot has happened and Figma has now also added AI features to FigJam. Nevertheless, it remains interesting to use your own plugin with your own prompts in FigJam and especially how to handle sensitive data. I will follow up once I find time again.