> "Spotify says it’s using large language models (LLMs) to understand the user’s intent. Then, Spotify uses its personalization technology — the information it has about the listener’s history and preferences — to fulfill the prompt and create a personalized AI-generated playlist for the user."
So "AI" to understand the request but I wonder how much of the second playlist generation step is machine learning based or if it is just their normal playlist algorithm fed with keywords. One of the reasons I don't like streaming music services is that the automatic playlist generation produces minor variations of the same playlist based on listening history & popular songs that it's hard to break out of. I'd rather they used ML to work out that if I take a break from my normal mix of music to play one pop song for my kids, I don't want to then hear 90% pop forever more!
hopefully you can disable this and not be bothered by it. the recent-ish "smart shuffle" feature shuffles songs in a playlist and also brings in songs outside the playlist, which i can see as potentially being useful, but there's many times i want to swap from shuffle to not-shuffle while listening to a playlist (eg, two or more songs that flow together) but because smart shuffle is controlled by the shuffle button it only appears in my user experience when i'm trying to do something entirely different, which only makes me resent it
Smart Shuffle is bugged for me. It always plays the same songs in succession whether or not they have been added to the playlist. The recommendations are quite poor too, much like radio it seems to match genre rather than the actual music itself
Anyone aware of a similar feature for foobar2000? I have an extensive library mostly tagged from Discogs, including release IDs. In theory, this should be sufficient to cluster music by genres, pull similar releases from Discogs "similar" feature and correlate data from https://everynoise.com. Obviously, in case of album mixed genres things will mix up, but I'm not sure there's a model that can correlate existing tracks with a genres/styles latent space which you can query for a playlist.
Can someone confirm if Plexamp + Tidal can also achieve this?
Considering the prompts in the story, you could do most of that without the first half of the sentence in Pandora for decades at this point. Some of my stations can drive, a few are close to being able to drink.
I shazamed a song at the airport yesterday, and opened spotify once I'd taken my seat. Realising it would be good to listen to this during the flight, I scrambled to download it. But nup. Spotify wasn't having it. No 'download' button anywhere. Running out of time until take off, I googled, scanned the top 4 articles, mostly guff, and still nothing (google mostly thought I wanted to illegally download music out of spotify). Anyway, in desperation, tried chatGPT, it said go to the three dots and select download. I tried but there's no download option.
So long story short I couldn't download it. Why are basic things so hard.
I believe you can only make playlists available offline. So, you would have to create a playlist containing that single song and then there should be a download button (didn't test it).
It's wild that users have to be psychic enough to figure this out.
It's a super daggy solution, but I've found being humble and putting a little message in the app to let users know of oddities goes a long way. E.g. "To download this song, you need to make it part of a playlist" would have been enough. Inconvenient for users wishing to download a song for offline listening, daggy for spotify's UI, but much better than the status quo of frustrating users.
Yeah, but it's maybe not so obvious that this is a bad solution, because allowing single downloaded songs would then also require a way to find those songs while offline. So the best UX would probably end up putting the song into a playlist called "Downloaded Songs", which would always be downloaded. Net result, a playlist that has downloaded songs, with a slightly better UI for putting songs into that playlist.
In Apple music songs can be in three^ states: Null, In Your Library, or Downloaded, with downloaded being a subset of “In your Library”. When you are offline you can look through all your library and see those songs which you’ve added but not downloaded as greyed out, versus the download ones as black. There’s also a view that filters the library so it only shows downloaded. If you click a black Downloaded song, it plays, if you click a grey In Your Library one, it says “Connect to Data to play this song” or similar.
^ I lied, there’s actually a fourth hidden state “Cached” that the UI presents as “In your Library” rather than “Downloaded”, and will give the same “Connect to data…” error if you directly click on it to play while offline. However, if you click Play on the album/playlist/etc itself, you can then navigate via the “Next Button” and “Up Next” queue to pick the song and if you played it recently, it’ll work. Great if find yourself in a plane and there are some songs you’ve been listening to lately but forgot to download.
I’ve been meaning to write a blog on “The Hidden Cache Apple Music Doesn’t Want You To Know About” for some time now lol. But knowing Apple I’d half-expect them to “patch the bug” by preventing the hidden cache from being played offline.
Apple Music has a "Downloaded" browsing section (along with Artists, Albums, Composers, Genres, etc) that you can browse and search separately. Works fine.
But isn't that the whole point of UX engineering? Liked songs is also "just a playlist" but by adding a unique icon for it now you can add it with one tap, filter artists that you like the most, and when you go to an artist's page it can show all the songs by that artist that you've liked.
I think you can create such a playlist. I did that (I called it "on the road" for exactly that purpose) and every time I add a song it's downloaded immediately. Unless they changed something.
But yeah in general I agree that Spotify UX is less than optimal.
Take for example their recent addition to the desktop app called "Now playing view" which opens every time you play a song (unless you disable it), squishes your current view into a three-pane window, and displays valuable information such as the song's album cover, which is already on the screen in the lower-left corner!
Possibly the UX for songs downloads has been A/B tested into oblivion. This is one of the downsides of 'data driven design' - any feature that's immensely useful in limited circumstances looks unimportant because it appears to be something users don't go to very often. In this case if 99.9% of users are happy being online the entire time, the 'Download' button won't show up in the data very often, so designers will reduce its prominence more and more. It sucks for times when you need it but really that's a reflection of reality - most people just don't need it.
Their new UX for liking a track and adding it to playlists is also awful. It's so hilariously bad that I wonder if they ever did user testing.
Say you want to like a track and then add it to a playlist. It used to be a heart icon and a plus icon. Now they combined everything into a plus, but depending on your current context it either adds it to an existing playlist or to your liked music. Also, if a track is already liked I would expect to "unlike" it when clicking the now green plus icon. But no they had to make it show a context menu because they combined liking and adding to playlists. It's just so confusing, and the previous UX worked fine and was clear for everyone.
I think you're right no one tested the plus icon. Because if someone did, then they would notice that in the bottom bar, there's [-] [+] but in the playlist, it's reversed, [+] [-]. What a stupid idea, why it's not consistent?
Shazam has a feature to automatically add shazammed songs to a "My Shazam" playlist in your Spotify account if you consent to, and then you could set that playlist to be always downloaded offline. Easy peasy.
I'm not aware of any complex software out there that wouldn't have the same complaint. I don't feel that your frustration is a good reason to not build cool new features for users.
So "AI" to understand the request but I wonder how much of the second playlist generation step is machine learning based or if it is just their normal playlist algorithm fed with keywords. One of the reasons I don't like streaming music services is that the automatic playlist generation produces minor variations of the same playlist based on listening history & popular songs that it's hard to break out of. I'd rather they used ML to work out that if I take a break from my normal mix of music to play one pop song for my kids, I don't want to then hear 90% pop forever more!