Netflix has imo the worst recommendations on the entire internet. They are a Java shop that’s 5-10 years behind the industry in AI, and now they’re trying to use the most grossly inefficient and overcomplicated method out there to try to be trendy instead of just using a DLRM like everyone else was in 2020. What could go wrong?
Netflix had an algorithm contest, when was it, like 20 years ago? A team made a highly effective content recommendation system and Netflix never used it. Why? It killed revenue.
Separately, there simply isn't enough content to recommend. It's not like you need help finding the perfect Wikipedia article. At any one time, there are ~50 things a user would probably watch.
Feels more like a soft pivot to cash in on AI valuations since Hollywood is cratering. Probably worth the investment.
> A team made a highly effective content recommendation system and Netflix never used it. Why? It killed revenue.
Nope. This is an internet urban legend. The Netflix prize was only ever for marketing and recruitment. They never intended to deploy whatever winning weights that win because they overfitted on the training and validation data. The dataset they released was not their actual full dataset. It contained zero personal information, just (userid, movieid, rating, date) which is only ever going to get you so far compared to what they do internally.
And it wasn't "a team" that beat the baseline of a ranking algorithm Netflix never deployed, dozens of teams were within fractions of a point from each other. Hence the over fitting strategy.
I think the biggest change from the Netflix prize days is that their catalog went away. Back in the DVD era, they had everything. And in the very early streaming era, they still had a huge number of things to watch.
But their current catalog is badly impoverished, and they're just going to recommend the same 30 Netflix originals they always recommend to me, plus a few films or series that are rotating through on a temporary license. If I actually try to search for something specific I'd really like to rewatch, it's almost never there. They haven't quite regressed to the level of a small-town, early 90s Blockbuster, but it sure feels that way sometimes.
So honestly, how much good can the Netflix recommendation algorithm do these days, given the much smaller catalog of movies and films it apparently has to work with?
(The one that I don't get is the Kindle recommendation algorithm. If I read one really good book with a certain theme, the Kindle immediately replaces my recommendations with 40 bad knockoffs, 25% of them clearly AI written. There's apparently no signal for "actually good.")
Have you considered that a majority of the userbase regularly rewatches the same shows again and again as "comfort shows", e.g. tenth rewatch of Friends?
Here we go, the start of LLMs plugged into everything.
Do Netflix recommendations really need to use LLMs?
This post paints a very altruistic picture of how recommendations can be fed from user history, preferences, device and environment context etc. However they make no mention of needing to advertise content from paid clients, promote new releases, and increase views on certain underperforming content. I'm assuming that once their fancy LLM spits out some user recommendations, they are then run through another process to 'commercialise' the results before displaying them to the user.
Okay. Well so if the model will tend to converge on the median of a user’s interests as measured by engagement… the user would have to depart from the recommendations to increase the odds of being recommended something surprising, no?
It’s interesting to see how Netflix evaluates RecSys internally, inferring from how they are comparing GenRec to it. In both cases the premise seems to be that users mainly want to engage with more of the same.
I would be even more interested in some kind of comparison to Netflix’s much earlier system for exchanging reviews and recommendations among a user’s human social group. Seems like there could be some insights around the input of fresh signals from aligned but not strictly conforming participants.
> “Seems like there could be some insights around the input of fresh signals from aligned but not strictly conforming participants.”
Exactly. I think you nailed a good subreddit.
What is the raison d'être of the a subreddit if it’s not to share your emotional quest to “refill the vase” of your love affair with the object of shared affection?
For example, I’ve made my own, somewhat delightful, but also archaic side quests from O’Brian’s Aubrey-Maturin fiction series.
Gibbon’s “The History of the Decline and Fall of the Roman Empire”, interesting authorial voice combining fiction and historical figures, but the characters have only brief moments on the stage.
The excellent narrator Patrick Tull of the A-M series lead me to:
Childers, Erskine, “The Riddle of the Sands” (1903), the archetype of the modern spy novel—with sailing!
So you want to post your find to the community? Maybe you have retained a trace of the author’s voice? So you bring your new books to the subreddit in fellowship.
But woe to you if in your excitement to share your new find you inadvertently, or intentionally, or carelessly, or with too much subtlety, break the “catechism” or moral codes of your subReddit group and insult its saints.
Well, then you’re in trouble. But lest we forget, _fortune favors the bold_. (What are the A-M peeps going to do if I piss them off? SWAT my house?)
I never understood why they don’t ask you what you like about a movie or tv show. I do not use streaming platform rating system because it’s so dumb: if I say I like “pacific rim” I get recommended all the dumbest sci-fi action contents I already know I will never watch. I love pacific rim for its esthetic, lighting, photography, and music. I couldn’t care less about mecha and generally hate modern sci-fi.
Like, just ask me instead of trying to guess incorrectly
When the Pandora music service launched, I remember reading about how it worked (predicting what you might like next) and it explained how they don't need to know anything about the songs at all (such as genre, tempo, instrumentation, place of origin, year of release, language, etc.) to predict what else you'll like. They simply identify other users with a similar history to you, and deliver to you some of the stuff that those other users had listened to that you hadn't listened to yet. If you abandon a song (suggesting that you don't like it much) then you become a bit less associated with the users who play that song a lot.
You don't think they have a genuine interest in providing good recommendations? A good algorithm will hook viewers more than a bad one. For the algorithm to work, they need to know what you like, and perhaps why you like it.
They can get more accurate results for "what keeps this person watching for longer with more ads" from your exact watching patterns than by asking you. HBO even had a very successful show along these lines.
How many survey participants would truthfully respond with "I liked when they showed nudity"? A video host might instead just see that the nudity scenes are where people gravitate towards with their pauses, plays, fast forwards, and rewinds.
honestly, who cares. they could make the best recommendation engine in the world, but their catalog is abysmal. It's the Facebook of film/entertainment. My wife tends to watch some ai written slop on there as a guilty pleasure, but otherwise it's nonexistent to me and i don't bother opening up the app anymore
Authors and "AI platform and serving" section are taken over by immigrants.
I wonder when the enshittification of the tech industry started and whether Netflix should perhaps just make better productions that exceed the average YouTuber in quality.
Yeah, I think that’s the cynical and correct answer, they don’t have the explosive growth from last decade and need something to boost their share price
In this setting, they are essentially using it as a feature extractor.
As for moving to this versus your bespoke feature extractor, _given_ that your existing features and their compositions are not antithetical to their language representation, an LLM will be an equal or better feature extractor. But for example, considering the basic feature genre, if your collaborative filter has learnt features that smartly recommending a show with one text feature "sci-fi" to people whose preferences have the text feature "comedy" because of learned behaviour despite the text, then you have to verbalise this feature "scifi,laugh track" or maybe providing samples of the subtitles of the show or add a "frequently co-watched with" section (which contains comedy shows) to the prompt, to effectively get an LLM to do the same thing (or many other ways to induce a hybrid embedding)
In many cases, people have almost entirely verbalizable features and feature compositions in their existing systems even if it may not exactly be optimal. So it's a good idea to try out LLMs there.
Composition mentioned everywhere above is crucial. Provided you can verbalise your important features, LLMs can perform very strong deductions and compositions out of the box above and beyond our own feature interactions that we use with say xgboost setups. And it's dynamic in the sense that it gives you a foundation model you don't have to retrain to use new (verbalizable) interactions.
Spotify seems to use an LLM for their new DJ feature, and it honestly is way better than any other music recommendation service I've ever tried. It will give you music that actually sounds like the music you ask for.
What I'm excited for is Bumble's AI matchmaker. LLM categorization may actually be the key to a decent dating app.
I just checked. The AI DJ showed me a suggested request button with the prompt "late afternoon reggae rock with Lorna Shore". That's like suggesting pickled fish with strawberry ice cream just because you happened to order both a few times.
I've not seen any difference between the spotify "AI DJ" and their daily playlists, they literally just pick a song you like and run similar.
The only difference is that every so often the DJ yaps at you and rotates to a different playlist.
It's still absolutely nothing like a DJ, and spotify daily recommendations are still hopeless compared to what Google Music ( rip ) used to do, which actually was able to deliver diverse playlists based on recommendations.
Just chopping between different homogenous blocs is not a good experience.
Netflix has imo the worst recommendations on the entire internet. They are a Java shop that’s 5-10 years behind the industry in AI, and now they’re trying to use the most grossly inefficient and overcomplicated method out there to try to be trendy instead of just using a DLRM like everyone else was in 2020. What could go wrong?
Netflix was torchbearer of projects that defined thr bigdata and cloud era. They were used as an showcase example for microservice architecture.
Their recommendation system was seen as industry standard best in class.
I wouldnt potray them in negative light.
Netflix had an algorithm contest, when was it, like 20 years ago? A team made a highly effective content recommendation system and Netflix never used it. Why? It killed revenue.
Separately, there simply isn't enough content to recommend. It's not like you need help finding the perfect Wikipedia article. At any one time, there are ~50 things a user would probably watch.
Feels more like a soft pivot to cash in on AI valuations since Hollywood is cratering. Probably worth the investment.
> A team made a highly effective content recommendation system and Netflix never used it. Why? It killed revenue.
Nope. This is an internet urban legend. The Netflix prize was only ever for marketing and recruitment. They never intended to deploy whatever winning weights that win because they overfitted on the training and validation data. The dataset they released was not their actual full dataset. It contained zero personal information, just (userid, movieid, rating, date) which is only ever going to get you so far compared to what they do internally.
And it wasn't "a team" that beat the baseline of a ranking algorithm Netflix never deployed, dozens of teams were within fractions of a point from each other. Hence the over fitting strategy.
I think the biggest change from the Netflix prize days is that their catalog went away. Back in the DVD era, they had everything. And in the very early streaming era, they still had a huge number of things to watch.
But their current catalog is badly impoverished, and they're just going to recommend the same 30 Netflix originals they always recommend to me, plus a few films or series that are rotating through on a temporary license. If I actually try to search for something specific I'd really like to rewatch, it's almost never there. They haven't quite regressed to the level of a small-town, early 90s Blockbuster, but it sure feels that way sometimes.
So honestly, how much good can the Netflix recommendation algorithm do these days, given the much smaller catalog of movies and films it apparently has to work with?
(The one that I don't get is the Kindle recommendation algorithm. If I read one really good book with a certain theme, the Kindle immediately replaces my recommendations with 40 bad knockoffs, 25% of them clearly AI written. There's apparently no signal for "actually good.")
How much advanced machine intelligence does it take to not put stuff on the recommended list that they know I've already watched?
Have you considered that a majority of the userbase regularly rewatches the same shows again and again as "comfort shows", e.g. tenth rewatch of Friends?
Here we go, the start of LLMs plugged into everything. Do Netflix recommendations really need to use LLMs?
This post paints a very altruistic picture of how recommendations can be fed from user history, preferences, device and environment context etc. However they make no mention of needing to advertise content from paid clients, promote new releases, and increase views on certain underperforming content. I'm assuming that once their fancy LLM spits out some user recommendations, they are then run through another process to 'commercialise' the results before displaying them to the user.
Okay. Well so if the model will tend to converge on the median of a user’s interests as measured by engagement… the user would have to depart from the recommendations to increase the odds of being recommended something surprising, no?
It’s interesting to see how Netflix evaluates RecSys internally, inferring from how they are comparing GenRec to it. In both cases the premise seems to be that users mainly want to engage with more of the same.
I would be even more interested in some kind of comparison to Netflix’s much earlier system for exchanging reviews and recommendations among a user’s human social group. Seems like there could be some insights around the input of fresh signals from aligned but not strictly conforming participants.
> “Seems like there could be some insights around the input of fresh signals from aligned but not strictly conforming participants.”
Exactly. I think you nailed a good subreddit.
What is the raison d'être of the a subreddit if it’s not to share your emotional quest to “refill the vase” of your love affair with the object of shared affection?
For example, I’ve made my own, somewhat delightful, but also archaic side quests from O’Brian’s Aubrey-Maturin fiction series.
Gibbon’s “The History of the Decline and Fall of the Roman Empire”, interesting authorial voice combining fiction and historical figures, but the characters have only brief moments on the stage.
The excellent narrator Patrick Tull of the A-M series lead me to:
Childers, Erskine, “The Riddle of the Sands” (1903), the archetype of the modern spy novel—with sailing!
So you want to post your find to the community? Maybe you have retained a trace of the author’s voice? So you bring your new books to the subreddit in fellowship.
But woe to you if in your excitement to share your new find you inadvertently, or intentionally, or carelessly, or with too much subtlety, break the “catechism” or moral codes of your subReddit group and insult its saints.
Well, then you’re in trouble. But lest we forget, _fortune favors the bold_. (What are the A-M peeps going to do if I piss them off? SWAT my house?)
https://www.reddit.com/r/AubreyMaturinSeries/
It feels icky to me that they feed a log of what you’ve watched, when, and on what device, over to an LLM just to get back some recommendations.
Even though It’s probably exactly the type of thing I would do on a locally hosted LLM
This sounds like data they've already been collecting, they're just using it in a different way.
I never understood why they don’t ask you what you like about a movie or tv show. I do not use streaming platform rating system because it’s so dumb: if I say I like “pacific rim” I get recommended all the dumbest sci-fi action contents I already know I will never watch. I love pacific rim for its esthetic, lighting, photography, and music. I couldn’t care less about mecha and generally hate modern sci-fi.
Like, just ask me instead of trying to guess incorrectly
When the Pandora music service launched, I remember reading about how it worked (predicting what you might like next) and it explained how they don't need to know anything about the songs at all (such as genre, tempo, instrumentation, place of origin, year of release, language, etc.) to predict what else you'll like. They simply identify other users with a similar history to you, and deliver to you some of the stuff that those other users had listened to that you hadn't listened to yet. If you abandon a song (suggesting that you don't like it much) then you become a bit less associated with the users who play that song a lot.
I dont think they care about what your opinion is.
You don't think they have a genuine interest in providing good recommendations? A good algorithm will hook viewers more than a bad one. For the algorithm to work, they need to know what you like, and perhaps why you like it.
They can get more accurate results for "what keeps this person watching for longer with more ads" from your exact watching patterns than by asking you. HBO even had a very successful show along these lines.
How many survey participants would truthfully respond with "I liked when they showed nudity"? A video host might instead just see that the nudity scenes are where people gravitate towards with their pauses, plays, fast forwards, and rewinds.
honestly, who cares. they could make the best recommendation engine in the world, but their catalog is abysmal. It's the Facebook of film/entertainment. My wife tends to watch some ai written slop on there as a guilty pleasure, but otherwise it's nonexistent to me and i don't bother opening up the app anymore
1670 is a great show on Netflix. Come to think, it's currently the one great show, which seems to be about as many as they can do at once.
yeah i'll admit they'll release one good series maybe once a year
> My wife tends to watch some ai written slop on there as a guilty pleasure
Are they label as such?
Authors and "AI platform and serving" section are taken over by immigrants.
I wonder when the enshittification of the tech industry started and whether Netflix should perhaps just make better productions that exceed the average YouTuber in quality.
I’m curious as to what readers think Large Language Models will bring to the recommendations table that our current machine learning doesn’t.
It will bring the ability to tell investors that Netflix is using Large Language Models.
Yeah, I think that’s the cynical and correct answer, they don’t have the explosive growth from last decade and need something to boost their share price
I’m curious too, as ML is pretty amazing; LLMs are one application of ML; LLMs have a much narrower good set of uses than broad ML.
You can use LLMs for things they are a bad fit for. I know someone who uses it like a spreadsheet to add sums of numbers, etc.
In this setting, they are essentially using it as a feature extractor.
As for moving to this versus your bespoke feature extractor, _given_ that your existing features and their compositions are not antithetical to their language representation, an LLM will be an equal or better feature extractor. But for example, considering the basic feature genre, if your collaborative filter has learnt features that smartly recommending a show with one text feature "sci-fi" to people whose preferences have the text feature "comedy" because of learned behaviour despite the text, then you have to verbalise this feature "scifi,laugh track" or maybe providing samples of the subtitles of the show or add a "frequently co-watched with" section (which contains comedy shows) to the prompt, to effectively get an LLM to do the same thing (or many other ways to induce a hybrid embedding)
In many cases, people have almost entirely verbalizable features and feature compositions in their existing systems even if it may not exactly be optimal. So it's a good idea to try out LLMs there.
Composition mentioned everywhere above is crucial. Provided you can verbalise your important features, LLMs can perform very strong deductions and compositions out of the box above and beyond our own feature interactions that we use with say xgboost setups. And it's dynamic in the sense that it gives you a foundation model you don't have to retrain to use new (verbalizable) interactions.
Netflix recommendations are bad. So, I expect them to be bad differently.
Spotify seems to use an LLM for their new DJ feature, and it honestly is way better than any other music recommendation service I've ever tried. It will give you music that actually sounds like the music you ask for.
What I'm excited for is Bumble's AI matchmaker. LLM categorization may actually be the key to a decent dating app.
I just checked. The AI DJ showed me a suggested request button with the prompt "late afternoon reggae rock with Lorna Shore". That's like suggesting pickled fish with strawberry ice cream just because you happened to order both a few times.
I've not seen any difference between the spotify "AI DJ" and their daily playlists, they literally just pick a song you like and run similar.
The only difference is that every so often the DJ yaps at you and rotates to a different playlist.
It's still absolutely nothing like a DJ, and spotify daily recommendations are still hopeless compared to what Google Music ( rip ) used to do, which actually was able to deliver diverse playlists based on recommendations.
Just chopping between different homogenous blocs is not a good experience.