GPTGuardian

Hate speech detection coverage comparison, n=100
Hate speech detection accuracy comparison, n=100

Inspiration

All of us use discord daily, and we all wanted to experiment with the OpenAI API. This seemed like a great opportunity to do so.

What it does

GptGuardian is a content moderation tool for discord that monitors for hate speech, harmful content & misinformation (or a user defined level of censorship) and fact-checks text, images, audio, video and hyperlinks. The application allows for varying degrees of customization for administrators monitoring their servers.

How we built it

GptGuardian is build with python, and uses the discord.py API wrapper, the openAI API (in a multitude of ways), and several other hooks that work in parallel, to provide a nuanced opinion on the content in question.

Challenges we ran into

The whole of the audio/video features were decided upon late into the development of the project, and as such, had little planning going into the competition.

Accomplishments that we're proud of

Surprisingly, our fact-check algorithm proved extremely powerful. By combining GPT-4 and Bing Search we created a chain of requests that uses both to their full power, with high degrees of accuracy.