sloppy (algorthimic feed customization)

Every algorithmic feed carefully balances what you want to see, what platforms want you to see, and what someone's paid for you to see. These aren't always in tension (revealed user preferences often beat stated ones), but many users are looking for more control. Platforms are, of course, reluctant to give it. In the attention economy, power accrues to whoever owns the means of algorithmic production.

Recently, X in particular has struggled to get this right. Users increasingly complain about slop, content and ad quality, and more. That made me wonder how you could provide more user control, particularly on a platform like X with a notoriously expensive API.

So I hacked together a chrome extension called "Sloppy" (lol) that lets you retroactively remove content from your feed. It has some helpful attention-related utilities: you can toggle ad removal, blur images and videos, and turn on feed-only mode. But what I was really excited about was using a small LLM to enable users to semantically describe what they do and don't want to see.

This worked pretty well! Given the short textual nature of most X content, feeding a tiny model tweets and then taking an action based on user guidance is actually performant. You can imagine all sorts of optimizations from this basic approach. There are obvious limitations, namely that without the API inserting tweets into your feed is difficult. But there are clever ways around that. If people are interested, I'll publish this to the extension store.