An Arc Boost that removes bad tweets using AI. See the original post.
I wrote this in an hour so don't get too excited, but some people asked me to share the code. I hope that we can make this boost better together. Or that this will at least teach you how to create your own LLM-based boost! Contributions are very welcome.
![arc-boost](https://private-user-images.githubusercontent.com/15680527/240443426-ab9355b9-f828-44d8-b4d4-49e4de584998.png?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.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.ceBS3uRbgUKU9AVLwEpWmLT6vIdO6OnSn5-KR_TjqPc)
- Create a new Arc Boost using the "Replace" template
- Swap out the template's content.js for this repo's content.js
- Add your own OpenAPI key
- GPT turbo isn't always smart enough and frequently misclassifies tweets (e.g. labels any tweet about AI as 'bad')
- An easy solution is using GPT-4 but that could require modifications to the prompt for the best results
- It doesn't know about a tweet's images, author, etc.
- I haven't done the math on how expensive this would be to run constantly :)