YouTube Reals

Youtube Reals

Inspiration

AI-generated scripts are taking over the internet: you click a video, and two minutes in you realize the narration is robotic, the facts are vague, and nobody actually made this. The only way to find out is to spend your time on it. We wanted a way to know before you press play, and to protect the real creators whose work gets buried under it.

What it does

YouTube Reals is a Chrome extension that reads a video's transcript and gives you an AI-script score: an estimate of how likely the script was written by AI. With one click on any YouTube video, it returns the result in a few seconds and informs you how likely the script was AI-generated.

How we built it

Chrome extension: grabs the transcript from the YouTube page and shows the score in a popup.

Backend: a small Python server that receives the transcript from the extension and passes it along.
Superhuman Go agent: the core of the project. We built a custom agent in Superhuman Go that analyzes the script for signs of AI writing, such as uniform sentence structure, formulaic transitions, generic claims with no specifics, and a lack of personal detail, and replies with a score.
Slack: connects the backend to the Go agent and carries its reply back, which the server returns to the extension.

Challenges we ran into

For our project, we struggled with understanding the problem space in the context of what Superhuman Go and Grammarly would be able to solve. We spent a lot of time in the ideation phase, trying to scope out a problem that is relevant, meaningful, and timely.

Accomplishments that we're proud of

We are proud that this is a product that we can imagine ourselves using, as it is able to predict the probability of AI-generated scripts quite well. It is able to help viewers more accurately understand if the content they are watching is genuine, authentic and helpful.