I tried to use Pocket a while back. First it put me in an HN cluster where it would recommend all the articles posted on HN that I didn't save to Pocket because I wasn't interested.

Then I intentionally avoided saving articles from HN and just saved any decent blog posts related to programming. One of those must have mentioned Merkle trees or something because it started recommending "5 ways Bitcoin will make your farts smell better" etc. Then I gave up.

I think there might be a point where recommendation algorithms kind of 'overfit'/become a filter bubble. I remember reading a article about the youtube recommendation algorithm led to increasingly more radical videos until you end up on 'that weird side' of youtube (https://www.nytimes.com/2018/03/10/opinion/sunday/youtube-po...). Depends entirely on what metric you are trying to opimize, I guess. With youtube is probably something like "time spent on the site". If you want non-sensational content instead you should probably read papers on arxiv or something.

For some reason YouTube assumes I really like Mariya Takeuchi - Plastic Love and recommends it from almost every video I watch. A look at the comments suggests I'm not the only one this happens to.

https://www.youtube.com/watch?v=3bNITQR4Uso

Or maybe someone at YouTube just has great taste ;)

> If you want non-sensational content instead you should probably read papers on arxiv

I do the YouTube equivalent and watch mostly educational channels: minutephysics, numberphile, 60 symbols, cgp grey, 3 blue 1 brown, spacetime, veritasium, codyslab, physics girl, etc. computerphile for checking my layman's terms explanations of computer concepts.

> It's lots of outrage & negativity, and I imagine it's because that's the kind of article people are actually reading & sharing, so that's what feeds the recommendation algorithms

I would think a ML model could fairly accurately judge the level to which articles are written in an intentionally inflammatory way, I would like to see tech companies start to make a sincere effort to allow users set an adjustable filter on this "inflammatory-ranking" dimension of the content in their feed.

> pseudo-personalized clickbait that is exactly what I'm trying to avoid.

I'm not sure this is possible anymore. Even legitimately well written articles and sources now use clickbait titles to just be competitive. And you can't really blame them, they're just trying to stay in business using the systems people like us make. I'm really concerned that we're conditioning new generations to expect clickbait titles. It makes me wonder how long until this is just the norm for all titles. And the most abrasive thought, how long until we've reached a point where avoiding clickbait titles actively stunts your ability to get information and knowledge?

At this point, I've stopped worrying about the titles and just focus on the sources.

But even if it is tailored to your liking, you'll also suffer from a worse bubble than we all already suffer from.