4 ms·
I think the problem is they're using machine learning on something where it's hard to know when to reward the AI. Do you reward it for max minutes the users wat
by alan_n 8y ago
I think the problem is they're using machine learning on something where it's hard to know when to reward the AI. Do you reward it for max minutes the users watch, max clicks, etc? None of them imo really measure whether the users are satisfied with the recommendations which even if that means less watch time for them, it's good for youtube as a brand.
I just ignore them completely now except for when browsing music. They often don't even help when watching serial things (e.g. part 1, 2, 3), I'll often make the effort to look up a playlist instead. I have also always ignored the front page, so can't say anything about it.
I do wish youtube would let you pick areas of interest and whether you wanted to be recommended based on those areas. If a video fits in an excluded area, you would just get random recs, otherwise you would get recs from around that area. Preferably a mix of recent + old + really old videos that still have good ratings / watch time.
Also they would have less to sort through, etc, if they allowed creators to "archive" a video (because its outdated and maybe some other newer one replaced it). This happens to me all the time. And as a user it would be nice if for example, when some video's audio was bad and then a newer fixed one uploaded, if I was prompted to go to the new one. Similar thing with serial content. It would be nice if creators could establish one video follows another or is related.