Inspiration
Over winter break, I had the chance to try ChatGPT. My first question was asking if it could explain Linked Lists at an intermediate level, a concept I got pretty well and could verify if it was wrong. I didn't expect much when I pressed enter, but then it started typing, coding and diagramming. It kept going and going, and the craziest thing was that it was accurate! That was when I knew Generative Chatbot AIs could be something special.
I told my brother, who's not in computer science, about my experience and we ended up having a 3 hour long conversation about how ChatGPT and machine learning work. The main thing that we came away with was the idea that the most important thing is remembering these AIs aren't some nebulous tools that can do incredible things but only if we can use them for their best purposes. You wouldn't hammer a nail with a screw right?
This gave me the idea for a simple program interface that would help users generate quick simple prompts to perform actions that generative Natural Language Processing algorithms like ChatGPT are uniquely good at.
What it does
If a user has a use case for services like ChatGPT, but doesn't quite know what to ask or how best to word their command, they can open the app and follow the menu options to make a prompt. This breaks the task of making a prompt up instead of having to imagine it all at once. It also prioritizes brief answers.
For example, if you're a lawyer trying to get a quick list of Pros and Cons of Living Constitutionalism, you could open the app, go to Research>Pros/Cons and enter your topic and the level of detail you need per point (perhaps you'd say, list the just the pros and cons and don't elaborate). From there, it will interpolate your answers into a simple and direct prompt you can copy and paste into any Generative Chat AI on the market right now.
Or another example, if you're an office worker toiling on a tough project and you wanted some motivation, you might ask it to act like a motivational speaker speaking on the benefits of a strong work ethic and lots of breaks. In which case you'd go to Create>Act, and input what it should act like and what it should talk about.
By generating prompts, not only does it make GAIpg make it easier for users to integrate it into their workflow and not use mental resources just to come up with an idea; It also teaches them how to make better prompts themselves by leading by example.
How we built it
I used Figma as I wanted to take a stab at UX design. The simplicity and approachability of this project are a must, as I want more people to get comfortable adding generative AIs to their workflows.
That's why I determined three primary functions of these NLPs: Researching, Analyzing, Creating. Then I color-coded the three different major functions of Generative Chat AIs.
Challenges we ran into
The main issue is that since the interface has a lot of branching paths, it would require many if-then statements to code. As a result, I started working on a command shell version of the generator in Java, but didn't have enough time to implement all of the cases. However, in Data Structure we're learning about the Tree data structure, which may make actually coding this project more efficient. As well, I think this is a project where the concept is one of the more important parts.
Accomplishments that we're proud of
I'm really proud of coming up with a simple yet powerful idea to make a new and emerging useful technology more accessible. I'm extremely passionate about science education, and I want the public to be more familiar with AI and be comfortable using it themselves.
What we learned
I learned how to use Figma for this project, which made it easy to get my graphic design side as well as make a partially functional prototype.
What's next for Generative AI Prompt Generator (GAIpg)
I'd like to finish my command shell version of it, then perhaps learn how to code on Android/iOS and make a mobile version. I'd also see if there's a way to make a macro that opens ChatGPT and pastes the prompt into the textbox automatically.
Built With
- figma
Log in or sign up for Devpost to join the conversation.