My impression is that's a problem with Google Shorts.
Google has been trying to get me to watch Shorts since they introduced them with little effect.
One day they offer me a video of a Chinese girl transforming into a nine-tailed fox on Shorts which was a good choice for me. After that they want to show me endless AI slop videos of Chinese girls transforming into all sorts of things on America's Got Talent always with the same music, the same reaction shots, etc.
I tend to look at the recommendation problem as a classification problem [1] but one thing that challenges that is that the answer to "do you want to see more content like this?" is different for content that is 5% of your feed (you relish it) to the very same content if it is 95% of your feed (you're disgusted.) My own answer to the diversity problem [2] is to k-Means the content into 20 clusters and rank the cluster independently so I am always operating at the 5% point.
With ordinary YouTube content I frequently get introduced to something like Techmoan or Jay-Z videos that I can binge with relish the same way I can binge episodes of Shangri-La Frontier season 2. But shorts don't do that for me.
If I was developing a recommendation-based content site based on short content in 2025 I would take a cue from Yostar games and make it so people are actually discouraged from sitting in really long sessions but rather you get them to keep coming back frequently to graze. I'm amazed at how a mechanism like oil in Azur Lane can rescue you from a grindy task right when you are starting to get sick of it forcing you to either engage with some other part of the game or real life for a while. There are a few of us who will spend the holiday break playing Dynasty Warriors 9 or Asgard's Wrath 2 and realize we spent a few work weeks worth of time playing a game, but even then you burn out, I think mobile games are more successful at getting more people to spend more time with games, often with content that is thinner.
[1] ... actually every problem, Hot Dog/Not Hot Dog is not a joke in my pod
[2] which I haven't seen in the literature. I used to think that I didn't understand or believe many ideas in the recommender literature such as "negative sampling", now I think the recommender literature is frequently wrong