Accountability

• 

  Example of a prompt and response

The problem: misinformation and spam content producing bots are not fought hard enough, and they currently lead to most Americans feeling some sort of misinformation in their lives.

The solution: accountability, our app allows users to provide a youtube link (with potentially easy access/support for other platforms) which is then downloaded, transcribed, and processed to be displayed for the user. It rates the candidate on a scale from 1-5 on how truthful they were in the given clip, provides a summary, confidence score, and their most “deceptive” (and often funniest) quote from the clip. It provides sources at the bottom and considers many sources from varying outlets to avoid bias due to how we prompt it.

The tech: The entire app is built on a python backend, django frontend, and aws to host the project so that anyone can use it. Important libraries and api include python library yt-dlp, deepgram for transcription, and perplexity for querying and analysis. We switched the python library youtube-transcript for faster analysis and aws compatibility, though. The pipeline is constructed such that you can put in a link, do nothing else, and get back a complete, accurate, and sourced analysis of the video content you entered. We are doing a localhost demo because as you may know, Amazon and Google don’t exactly get along (querying a youtube video from an aws server causes problems) so for now its local, however we have a multitude of options like precomputing a ton of transcripts and constantly updating a database which can then be queried with a search and those results can be returned.

NOTE: due to google denying requests from aws servers (shocker), the online webapp may not work properly. If you get an error, this is likely why. It works occasionally, though.

Impact: Many people, especially people who aren’t well versed in technology or AI can’t see the obvious problems with AI generated content such as deepfakes, or even simply clips taken out of context. Implementation of this technology into a site like facebook, twitter (x), or any platform would allow the user to easily see a confidently rated analysis of the media they’re consuming which greatly increases the health of the platform. Misinformation is a disease, and complacency with it will only further pollute the internet and stir social tensions. This tech can be used to label and analyze any video technically, however its focus is politics and specifically misinformation.

Built With

• and-aws-to-host-the-project-so-that-anyone-can-use-it.-important-libraries-and-api-include-python-library-yt-dlp
• and-perplexity-for-querying-and-analysis.-we-switched-the-python-library-youtube-transcript-for-faster-analysis-and-aws-compatibility
• deepgram-for-transcription
• django-frontend
• the-entire-app-is-built-on-a-python-backend