This project was deeply inspired by Tim Ferriss' idea of information diets: we all consume too much news, and I wanted only "real news" to be able to reach me.

The news is >90% noise, ragebait, and engagement farming. Most headlines are just trying to make you feel something.

Under the Rock News ingests RSS feeds from sources like the BBC, CNA, MIT Tech Review and Bloomberg and runs each article through a scoring system.

The metrics were:

1. Actionablity: Can you make a decision or take any actions as a result of learning this?
2. Long-term impact: Will this matter?
3. Uniqueness: Is this just a retelling of something on-going?

So yes, I tried to make it very Tim-coded.

Articles are summarized with BART and scored by GPT-4, and only those that exceeded a threshold reached your personal news feed.

...

Unfortunately, I've sunset this (for now, perhaps) because the $40/month API cost wasn't justified when the better alternative turned out to be not consuming any news at all. I also couldn't work out a solution to the core problem of encoding my own biases into the filtering criteria, or structuring it in a better way, which means it had limited use in being shared.

It was, however, my first time working with a UIUX designer directly. I learnt a lot! And you should definitely check her out if you're reading this.