The lack of nuance is killing all these sites for me, not just Netflix, which I gave up on managing. It's when sites carefully note what I look at, and start making drastic changes based on that.
It's like if I go for a walk, metaphorically, and I see a giant pile of shit, and that's a genuinely interesting thing to see, because Jesus Christ that is huge. Then the algorithm notices I stopped to look at that, and the next time I go for a walk everything is filled with shit. That's what my Instagram feed is right now, metaphorically speaking. A never-ending field of shit.
It doesn't even have to be a pile of shit; it could be something that was interesting by itself, but that doesn't mean I want a deep dive.
This happened to me with youtube and feral hog trapping. A two-minute video of that showed up in my feed, and yeah, it was interesting to see how and why it was done. But my feed was full of that for a long time, and even now, years later, one will pop up just desperately trying to get me to re-engage.
It was interesting as a one off, not as a way of life.
This made me think of how dystopic would be a virtual reality with an algorithm that behaves like this to give the user more of what they pay attention to
To me, the "best of the best" is its own thing. I do not watch romance movies in general, but I've had good experiences watching what people call the best of the best of the genre. Similarly for a lot of genres, media types, etc. Even just the "best of the best" of "cute animal video" can be fun every once in a while.
But I have to keep a heavy hand on the YouTube watch history and remove many things that I may have enjoyed as an exception, but don't want to see endlessly offered up forever. Some of the strange attractors in the algorithm are very, very powerful, like the aforementioned "cute animal videos". Another problem I hit is situations like, I watched the video because, say, a parrot was doing a very good impression of Captain Picard (just making this up, sorry) which was given due to general sci fi interest, but the algorithm sees "a ha! another hapless human who likes cute animal videos! Après cette vidéo, le déluge!"
At least YouTube has that knob, and does generally seem to honor it. The algorithms that don't are very hard to keep from degenerating into the lowest common denominator, because the slightest hint that you like some extremely popular thing or have an interest in a very lucrative ad keyword almost immediately swamps my actual interests.
I didn't mind naive rating algorithms before statistics and recommendations got so gamified.
YouTube and Amazon are some of the very worst. I regularly watch a video or song on YouTube (/Music) and explicitly say "I enjoyed that but don't want YouTube to focus on recommending me things like that so I'm not going to hit like".
Amazon, there are entire classes of links I won't click or products I won't search/buy because I don't want them filling my recommendations forever.
And no, having to manually dig through settings and find a list of my entire history to manually delete entries isn't a good workaround. Too much effort, easier to just not give the platforms the data in the first place.
I wonder if most people like/don’t care about recommendations and don’t want to spend time giving companies more feedback.
I work in streaming recommendations and basically no one voluntarily gives extra feedback, and most initiatives that aren't machine learning based on true behavior fail on A/B tests.
I do agree a "I'm watching this but I regretted it" would be a useful button, if the companies respected it. But my default presumption is that the already know this, and don't care, and that baiting engagement / addiction is at the very least "views" and possibly a deliberate strategy.
I think we need to start building intermediary feeds for all of our services (presumably via AI or something) that lets us customize how much we get of each type of content (e.g. no more than 5% about bad the economy is, no more than 5% about a new AI trick, 0% about 'drama of the day', 40% educational)
If you weren't aware, YouTube kind of has a version of this signal, which is to go into your history and remove the video you just watched. Supposedly this removes its influence on further recommendations. And then if that video or one like it appears again in recommendations, you can tap the ... next to the video and suppress recommendations for it by picking 'not interested'.
In my experience at least cultivating recommendations this way has worked really well for me on YT, I get consistently recommended a mix of stuff I want to watch and stuff that is at least theoretically interesting to me even if I don't want to watch it.
I wish other social services had the same mechanism (and wish it was more ergonomic on YT).
I understand the frustration with current recommender systems, but what are the alternatives? Adding a bunch of arbitrary buttons or sliders isn't likely to actually improve the experience.
"People who liked this also liked this" is already a pretty good way of discovering things. The problem is when the catalogue is crap. I get plenty of good recommendations on Prime Video and Criterion Channel, which both actually have a substantial amount of great films available to watch.
I don't want "recommender systems", I just want easily-browsable catalogues that start out being sorted by their actual genre (looking at you, Netflix homepage with a bunch of bullshit 'categories'). I want to be able to look through things at my own pace and in my own way. I want to browse through movies/shows the same way I browse for records in a record store.
I'm happy to take recommendations from friends and family members who know what my nuanced tastes are, I don't want recommendations from some unknown algorithm built by someone for the purpose of keeping me on the platform.
I miss when you could rate how much you liked something after a watch and they could incorporate that. Now it's just kinda "how long did you watch it and who else watched it" which seems less nuanced?
that's very true for reviews of products, the classical five stars. Even with comments, they are seldom more than a data point, and stars themselves are given with very different meanings by different people.
So when reviewing reviews, care is needed. Simply picking the product that maximizes stars and number of reviews is a bad metric for the quality of a product, that is, for what your perceived quality of that product will be.
That's why I look at two things:
1) distribution of reviews; if frequency of stars decreases monotonically with star number, it is probably a good product, and I look at 3/4 star reviews for pro and cons;
2) if the distribution has a spike in the 1 or 2 stars, I further investigate to see what the problems are, regardless of number of reviews or how many five star the product has.
What's being referenced here is maybe the tip of the iceberg in terms of inadequacies of current recommendation systems.
I suspect there's a range of indices related to content interaction that companies use poorly or even maybe nefariously — for example, are they motivated to present what you want, or what will keep you engaged with the site? Are their assumptions about why, say, you're spending a lot of time on a video correct? This post is focused sort of on options to communicate with the recommendation system, but there's a lot that could be said in terms of mismatched system-user goals in the system, and poor assumptions being made by the system in general.
I wondered too as I was reading it whether it's worth making the distinction between "different features of user experience with the content" and "metaresponse". That is, I can feel positively and negatively about the same video, or like it for one reason but not another; at the same time I can want to provide a response explaining a response. There's a difference between providing information about how I feel about some content, and information about how I want that information to be used.
>There's no way to indicate, “I’m engaging with this, but I hate myself for doing it.”
that would be a thumbs down if it exists. The system already knows you're engaging with it, they checked that you stopped scrolling and did all sorts of stuff around the thing you are engaging with.
We've already forgotten the hype train that Netflix orchestrated when they held a paid contest for who can build the best recommendation engine. What was it - a million dollar prize?
It was shit back then, and it is still shit today. It just got worse over time.
The lack of nuance is killing all these sites for me, not just Netflix, which I gave up on managing. It's when sites carefully note what I look at, and start making drastic changes based on that.
It's like if I go for a walk, metaphorically, and I see a giant pile of shit, and that's a genuinely interesting thing to see, because Jesus Christ that is huge. Then the algorithm notices I stopped to look at that, and the next time I go for a walk everything is filled with shit. That's what my Instagram feed is right now, metaphorically speaking. A never-ending field of shit.
It doesn't even have to be a pile of shit; it could be something that was interesting by itself, but that doesn't mean I want a deep dive.
This happened to me with youtube and feral hog trapping. A two-minute video of that showed up in my feed, and yeah, it was interesting to see how and why it was done. But my feed was full of that for a long time, and even now, years later, one will pop up just desperately trying to get me to re-engage.
It was interesting as a one off, not as a way of life.
This made me think of how dystopic would be a virtual reality with an algorithm that behaves like this to give the user more of what they pay attention to
I'm astonished that the tactics are still so dumb. I buy a dining room table and forever after the site is recommending dining room tables.
To me, the "best of the best" is its own thing. I do not watch romance movies in general, but I've had good experiences watching what people call the best of the best of the genre. Similarly for a lot of genres, media types, etc. Even just the "best of the best" of "cute animal video" can be fun every once in a while.
But I have to keep a heavy hand on the YouTube watch history and remove many things that I may have enjoyed as an exception, but don't want to see endlessly offered up forever. Some of the strange attractors in the algorithm are very, very powerful, like the aforementioned "cute animal videos". Another problem I hit is situations like, I watched the video because, say, a parrot was doing a very good impression of Captain Picard (just making this up, sorry) which was given due to general sci fi interest, but the algorithm sees "a ha! another hapless human who likes cute animal videos! Après cette vidéo, le déluge!"
At least YouTube has that knob, and does generally seem to honor it. The algorithms that don't are very hard to keep from degenerating into the lowest common denominator, because the slightest hint that you like some extremely popular thing or have an interest in a very lucrative ad keyword almost immediately swamps my actual interests.
>Some of the strange attractors in the algorithm are very, very powerful
[Watches video on WWII]
Youtube: Congratulations on becoming a Nazi, here's instructions on how to join a supremacy group in your area.
I didn't mind naive rating algorithms before statistics and recommendations got so gamified.
YouTube and Amazon are some of the very worst. I regularly watch a video or song on YouTube (/Music) and explicitly say "I enjoyed that but don't want YouTube to focus on recommending me things like that so I'm not going to hit like".
Amazon, there are entire classes of links I won't click or products I won't search/buy because I don't want them filling my recommendations forever.
And no, having to manually dig through settings and find a list of my entire history to manually delete entries isn't a good workaround. Too much effort, easier to just not give the platforms the data in the first place.
I wonder if most people like/don’t care about recommendations and don’t want to spend time giving companies more feedback.
I work in streaming recommendations and basically no one voluntarily gives extra feedback, and most initiatives that aren't machine learning based on true behavior fail on A/B tests.
I do agree a "I'm watching this but I regretted it" would be a useful button, if the companies respected it. But my default presumption is that the already know this, and don't care, and that baiting engagement / addiction is at the very least "views" and possibly a deliberate strategy.
I think we need to start building intermediary feeds for all of our services (presumably via AI or something) that lets us customize how much we get of each type of content (e.g. no more than 5% about bad the economy is, no more than 5% about a new AI trick, 0% about 'drama of the day', 40% educational)
If you weren't aware, YouTube kind of has a version of this signal, which is to go into your history and remove the video you just watched. Supposedly this removes its influence on further recommendations. And then if that video or one like it appears again in recommendations, you can tap the ... next to the video and suppress recommendations for it by picking 'not interested'.
In my experience at least cultivating recommendations this way has worked really well for me on YT, I get consistently recommended a mix of stuff I want to watch and stuff that is at least theoretically interesting to me even if I don't want to watch it.
I wish other social services had the same mechanism (and wish it was more ergonomic on YT).
I understand the frustration with current recommender systems, but what are the alternatives? Adding a bunch of arbitrary buttons or sliders isn't likely to actually improve the experience.
"People who liked this also liked this" is already a pretty good way of discovering things. The problem is when the catalogue is crap. I get plenty of good recommendations on Prime Video and Criterion Channel, which both actually have a substantial amount of great films available to watch.
>... but what are the alternatives?
I don't want "recommender systems", I just want easily-browsable catalogues that start out being sorted by their actual genre (looking at you, Netflix homepage with a bunch of bullshit 'categories'). I want to be able to look through things at my own pace and in my own way. I want to browse through movies/shows the same way I browse for records in a record store.
I'm happy to take recommendations from friends and family members who know what my nuanced tastes are, I don't want recommendations from some unknown algorithm built by someone for the purpose of keeping me on the platform.
I miss when you could rate how much you liked something after a watch and they could incorporate that. Now it's just kinda "how long did you watch it and who else watched it" which seems less nuanced?
that's very true for reviews of products, the classical five stars. Even with comments, they are seldom more than a data point, and stars themselves are given with very different meanings by different people.
So when reviewing reviews, care is needed. Simply picking the product that maximizes stars and number of reviews is a bad metric for the quality of a product, that is, for what your perceived quality of that product will be.
That's why I look at two things: 1) distribution of reviews; if frequency of stars decreases monotonically with star number, it is probably a good product, and I look at 3/4 star reviews for pro and cons; 2) if the distribution has a spike in the 1 or 2 stars, I further investigate to see what the problems are, regardless of number of reviews or how many five star the product has.
So far has worked great on Amazon.
What's being referenced here is maybe the tip of the iceberg in terms of inadequacies of current recommendation systems.
I suspect there's a range of indices related to content interaction that companies use poorly or even maybe nefariously — for example, are they motivated to present what you want, or what will keep you engaged with the site? Are their assumptions about why, say, you're spending a lot of time on a video correct? This post is focused sort of on options to communicate with the recommendation system, but there's a lot that could be said in terms of mismatched system-user goals in the system, and poor assumptions being made by the system in general.
I wondered too as I was reading it whether it's worth making the distinction between "different features of user experience with the content" and "metaresponse". That is, I can feel positively and negatively about the same video, or like it for one reason but not another; at the same time I can want to provide a response explaining a response. There's a difference between providing information about how I feel about some content, and information about how I want that information to be used.
Social media websites often let you tag things with emojis or hashtags. This is more expressive, though people will disagree on their meaning.
Why would people send the thumbs-up signal in Netflix if they didn't actually liked it? Just...do not click anything?
This is an unexpected yet welcome application of the fundamental attribution error: https://en.wikipedia.org/wiki/Fundamental_attribution_error
>There's no way to indicate, “I’m engaging with this, but I hate myself for doing it.”
that would be a thumbs down if it exists. The system already knows you're engaging with it, they checked that you stopped scrolling and did all sorts of stuff around the thing you are engaging with.
We've already forgotten the hype train that Netflix orchestrated when they held a paid contest for who can build the best recommendation engine. What was it - a million dollar prize?
It was shit back then, and it is still shit today. It just got worse over time.
I'll have to disagree here.
It was shit back then, it is shit today, but it improved a lot over time before it started to worsen.