Spotify Newsroom

You had a chance to prove my assumption wrong by writing this same exact comment without all the snark.

At the end of the day if you're just unmoved by the implications that an ML model went from a bag of tokens to a structured, explained chain of thought, and a final response on an unknown task with rewards defined in natural english (!) and intentional ambiguity most humans wouldn't even try to confront... there's not much conversation to be had.

I think the rest of us (including your colleagues) will continue to build on these models, and like most advancements there'll be a vocal crowd insisting the car isn't useful because it can't be fed with grass.

> not in recommendations

You didn't have to say that after complaining ChatGPT's web interface didn't give both us the same reply (most people in ML understand how temperature relates to LLM output)

_

By the way, if making your own personal music recommender seems like "magical thinking", maybe you're a little lost on which parts of Spotify's recommender systems are complex due to scale: if Spotify only needed to make song selection work for one person at a time, they'd have a lot more leeway in architecture.

In fact, when the problem was flipped and they were scaling humans attaching their recommendations to songs, they built on OpenAI: https://newsroom.spotify.com/2023-02-22/spotify-debuts-a-new-ai-dj-right-in-your-pocket/

Not unexpected when you're a founding organizer of the "NLP 4 Music and Audio Forum".

The replies seem to be full of low-quality guesses, which is disappointing to see on this site (assuming they were not using sarcasm). I’ll try to do slightly better and link the actual press release and include the official statement.

Hopefully someone who has access to the C suite at Spotify can leak the actual answer here (I’m sure they will get right on that). For future reference just Google “x acquires y” if you want to attempt to get any insight about “what x planned to do with y after acquiring”.

https://newsroom.spotify.com/2022-07-12/spotify-to-acquire-music-trivia-sensation-heardle/

> “We are always looking for innovative and playful ways to enhance music discovery and help artists reach new fans,” commented Jeremy Erlich, Global Head of Music, Spotify. “Heardle has proven to be a really fun way to connect millions of fans with songs they know and love and with new songs . . . and a way to compete with their friends as to who has the best musical knowledge. Since its debut, the game has quickly built a loyal following, and it aligns with our plans to deepen interactivity across the Spotify ecosystem.”

The official statement seems consistent with what was reported in this comment https://news.ycombinator.com/item?id=35570998

In other words, they tried to retool the tech toward an additional value proposition for both users and artists. The result failed to meet a quality bar, and therefore it was decided to axe the project.

The article has all the pieces but doesn't connect that Spotify's incentive is "Discovery Mode": https://newsroom.spotify.com/2020-11-02/amplifying-artist-input-in-your-personalized-recommendations/

> artists and labels can identify music that’s a priority for them... it won’t require any upfront budget. Instead, labels or rights holders agree to be paid a promotional recording royalty rate for streams in personalized listening sessions where we provided this service

"Payola" is the term for radio stations secretly charging labels to feature their songs: https://en.wikipedia.org/wiki/Payola

"Discovery Mode" is the modern structure. Instead of a station's single broadcast, it's many algorithmic playlists. Instead of the publisher paying up front because the radio and retailer are separate businesses, they just accept a lower royalty rate on plays because Spotify plays are the discovery and the sale together. Perhaps these differences and this press release is enough "disclosure" to avoid prosecution, I'm not a lawyer. I wish Spotify's internal legal analysis was public.

But all these gripes come back to: Spotify makes more money from this than when playing albums, curation, personal playlists, etc etc. There isn't a general issue of control or caring, there's a specific, nameable moneymaking "feature".

Yes, subscribing outside the app does avoid the cut. Which is why you can't subscribe to spotify in the iOS app, you have to open a browser: https://support.spotify.com/us/article/cant-subscribe-to-premium-in-the-ios-app/

Until this year, Spotify couldn't even tell you in the iOS app that you could pay for premium: https://newsroom.spotify.com/2025-05-01/following-landmark-court-ruling-spotify-submits-new-app-update-to-apple-to-benefit-u-s-consumers/

This meant in apple music, the user could open the app and it would work including paid features.

In spotify, you could open the app and it would tell you "You can't subscribe here, sorry" and couldn't even link you to a webpage you could subscribe at.

I'm certain a non-zero number of users couldn't understand what to do with that apple-approved error and gave up.

Maybe there's a reason that apple lost in court for that one.

>its desirable for a model to be based on causal rather than merely correlated variables

Ironically, your "likes_hiphop" example would appear to be an unusually clean case of a variable that is likely to exert causal influence.

What do you think the causal effect of listening to lyrics like "Prolly leave my fuckin' show in a cop car" might be, on an impressionable teenage boy say?

From one of the most-streamed hip-hop songs of all time:

https://genius.com/Post-malone-rockstar-lyrics

https://newsroom.spotify.com/2024-05-20/best-hip-hop-songs-100-top-tracks/

>A model based on "socioeconomic status" has a totally different social meaning than one based on race, even if we cannot fully disentangle the two statistically.

I see no evidence Gwern disagrees with this claim. He just seems to be arguing the "cannot fully disentangle the two statistically" part.

Spotify have announced "Spotify HiFi" (currently without an exact release date or pricing information) for 2021, which will offer lossless streaming[1].

Anecdotally though Spotify's "Very High" quality works well for me and honestly in 99% of cases it is indistinguishable. My setup sure isn't the greatest, but I have a dedicated DAC, a headphone amp, and speaker amp with some reputable headphones and bookshelf speakers. I keep CD sourced FLAC files on a local disk but I never really play them anymore.

Maybe if I had large floorstanding speakers in the 1000$+ category there'd be a lot more cases where Spotify falls short - where the HiFi offering or local FLAC would have a strong case - but I certainly recognize it is a niche case.

[1]: https://newsroom.spotify.com/2021-02-22/five-things-to-know-about-spotify-hifi/