[Johnny555]

today's AI is pretty good

It's pretty good at suggesting exactly the same type of movies you've watched before, but not so good at those "but how about something completely different, but something that I'll still like" suggestions, that's where a real human movie buff can help.

[ryangittins]

I've long pondered building exactly this, an anti-recommendation engine. You'd go through and mark some favorite films or genres, and it'd come back with something great but totally unlike your usual picks.

Foreign flicks not your thing? Try Parasite.

You don't remember what it's like to be an awkward pre-teen? Eighth Grade will remind you.

Not a big action movie person? Maybe you need to watch Die Hard.

Kids movies are just for kids? Spirited Away!

You can't connect with female protagonists? You've got to see The Invisible Man.

Don't find food interesting? Try Tampopo.

Sick of movies which try to make a statement and fall flat? Promising Young Woman.

Most dramas feel too contrived? Marriage Story.

Musicals and plays aren't really your thing? Hamilton.

[zw123456]

Agree. Too much here is more of the same. Ugh.

I like cowbell.

Here is an endless loop of cowbell ringing and nothing else.

It's like torture.

[ahartmetz]

Just noticed this again today. Spotify vs Hype Machine (browse, then "recommendations" by checking out people who also liked $song). Spotify is somehow stuck on dark, aggressive electronic music for me. It's not the only thing that I like, not even my favorite really.

[vidarh]

Music recommendations are even worse than movies, in that they need to take into account what you've just listened to (if I'm listening to a specific genre, I'm unlikely to want something in a very different tempo and style), time of day (at bed time I'll often put on some soft classical music; that doesn't mean I want that every other time because I play it a lot), mood (if I suddenly play 1980's pop, it's probably nostalgia, and suddenly playing new music is not what I want), and at the same time I have preferences, but those are mediated by what I'm currently listening to.

Whenever I try to get a service or other to recommend music they are remarkably bad at this.

[ahartmetz]

last.fm had my favorite auto-recommendation system, but it seems like they removed the personal and artist radio features.

[vidarh]

Yeah, they're the closest I've been to being happy about recommendations by far, though they too did get into very weird "pockets" of recommendations for me at times (not bad, just very obsessively giving me music from a very small niche of my music interests).

[cbanek]

Gotta have more of that cowbell!

[sundvor]

And rainbows. :)

Love GGP's anti-recommendation engine.

[Akronymus]

Think indie movies aren't worth watching? Primer.

Tbh, almost all of the recommendations I follow are word of mouth, so I'd really appreciate such an engine.

[actusual]

This isn't completely true. What you're talking about are called "echo chambers" or "filter bubbles", and there are ways to make sure they don't negatively impact users.

Additionally, studies have found that (when A/B testing recommender system vs no recommender system) users create their own, more localized "echo chambers" in absence of a recommender. This is measured by the "diversity" of content consumed, which decreases if a user is their own recommender.

[h0l0cube]

> This is measured by the "diversity" of content consumed, which decreases if a user is their own recommender.

The recommender in the GP and OP is a video store clerk, that’s probably self selected into that kind of job by their special interest in movies

[dreamcompiler]

Some of you will remember the "Pepsi Challenge" [0] when Pepsi would make commercials about random people tasting unlabeled soft drinks. People usually chose Pepsi in taste tests because it was sweeter than Coke. What the commercials didn't show was that most people actually bought Coke because Pepsi is too sweet for many people when they drink a whole bottle.

Movie and music recommendation engines seem to work like this. Human tastes are much more complex and multi-dimensional than "This person prefers genre X" but AI systems don't even come close to grasping the subtlety.

[0] https://en.wikipedia.org/wiki/Pepsi_Challenge

[jrochkind1]

Remember when Netflix had the context for the best recommendation algorithm? I don't think they ended up using anything from it.

[usrn]

I haven't thought about it but most of the movies I've really enjoyed were recommended to me by my dad, girlfriend, or my brother in law. I can't think of anything I really enjoyed from a recommendation algorithm. The last movie I tried I certainly didn't enjoy.

[jvalencia]

The reality is likely that the algorithms are trying to increase viewership, not quality of views. So they are likely to feed you things that get you to watch another thing, rather than things that are worth watching, leave you satisfied, and off doing something else.

[E4YomzYIN5YEBKe]

Do Netflix et al inject "noise" into their recommendation engines? Like 1/10 titles are random rather than based on your preferences in order to break out of "local minimums".