And herein lies the issue with filtering/tailoring based on what you've watched/liked before: it can never 'escape' and give you something different that you might actually really like.

I don't know how many good titles I discovered browsing random titles in video stores back in the day.

And it would be so simple to solve. Prepare a tailored list. Instead of using it on all available space, fill only 90% of the space with items from the list. Fill remaining 10% with random items that do not belong to that list.

Yep this is how you’re supposed to escape from a local maximum. But maybe Netflix’s usual customer doesn’t stick around long enough to have this problem so it’s not worth fixing. I can’t imagine none of their machine learning engineers know how to fix this.

Yeah, I bet they know. Question is, do their managers want them to fix this?

There is this hypothesis that I've seen floating on HN at times, that Netflix does this to hide the fact that their selection is very small. I'm not totally convinced, but a lot of their UI choices feel like good evidence in favour of this hypothesis.

Thankfully there are alternative interfaces to browse the Netflix catalog:

https://www.coollector.com/index.html#netflix

I think that could be solved by having a button to mark something as watched, removing it from recommendations and the catalog other than the list of things you watched.

You could hide episodes you've seen or entire seasons up to entire shows. Would fix a lot of the issues I'm having.