The Banhammer

Inspiration

Ctrl-C here at HackED 2024, University of Alberta came up with this idea because we thought that moderators on social media platforms such as Discord should use AI to more accurately flag users instead having to constantly pay attention to their servers' chat feeds. To keep our use of AI fair we still give the moderators the final say on punishing offenders. This is especially important for moderating misinformation as the APIs we are using, OpenAI moderation and GPT-3.5, are not as up to date to current events as we need them to be. Since they can be used to write misinformation themselves, they also often fail when scanning for fake news made by other Large Language Models (Llamas). But still, we knew that OpenAI would make the best option for this project as their moderation endpoint is completely free!

What it does

A Discord bot monitors the server for all messages through Python communicating with Discord API. Then, two checks are performed, calling OpenAI moderation and GPT-3.5 for hate/violence and misinformation respectively. All relevant data is sent to a Firebase database which is then used within the website to display the context for the flag action.

How we built it

We imagined this project from front to back and then built it from back to front. We made notes and sketches of what features we wanted beforehand, while also ensuring that we don't take too much so as to not clog the website full of noise. Our UI design is simplistic and our Firebase retains data to catch the frequency of repeat offenders.

Challenges we ran into

Progress is bound to slow down in a hackathon and we definitely felt this as Saturday night started, some of us even continued working through the morning as Sunday rolled around. We struggled but eventually got through calling and gathering the replies that the APIs were giving us, and some of us had to learn how to use Python and Firebase in a web-development context. We also eventually used up our free Firebase quota and had to make another one.

Accomplishments that we're proud of

Ctrl-C is so proud of our Discord Bot that we ended up testing it so much for fun. We kept saying that: 1+1 was 3, the sky was red, Joe Biden stole the 2020 election, and that we really wanted to kill our grandparents. All the information we needed was on there and with some hard work we got it to the Firebase for easy web access.

What we learned

HackED 2024 was a great experience as it really opened our eyes with what needed to happen to develop a web-app from start to finish. For some of us it was our first hackathon, for others it was our fourth or fifth. But as AIs and Llamas are so new, we all found ourselves learning more about the safety that needs to be in place to ensure that they stay on track, and the power they have to make our lives better.

What's next for The Banhammer

Hackathons are usually a one and done deal so this project will probably fizzle out as our group returns to our different classes, some even to different campuses. But these connections we made will stay with us so maybe there will be Banhammer V2, but no promises!