I mean, you say "is this it?", but the multi-language support of literally all other Android keyboards I've used are basically unusable. Take Google's default keyboard, it requires you manually switch languages for swiping and predictive text to work. But that's completely unworkable for the way multi-lingual people end up actually typing texting. I swap back and forth across languages, even if just for 1 or 2 words and having to explicitly toggle languages is a MAJOR speedbump in typing.
Swiftkey just lets me auto-complete/swipe in multiple languages seamlessly. Want to swipe 1 English word inside a completely Dutch sentence? No problem. The reverse? No problem. Using Google's keyboard the same way is endlessly frustrating.
What do you mean when you say "Google's keyboard"? I use Gboard and frequently swipe sentences which mix Spanish and English just fine. You enable "Multilingual typing" on the specific keyboard and then tick off languages from other installed keyboards to enable them. Maybe you haven't tried gboard for a long time?
There's some internal matrix of "Supported" clean movement, mostly between languages with an extensive shared vocabulary. You can enable Multilingual Typing theoretically anywhere, but it only really "works" among languages that share a layout and character dictionary.
But if you need to loanword a non-common word from English into another language (e.g. Hebrew), Gboard just can't do it. If you change character sets, you change entire dictionaries. SwiftKey has a unified dictionary among all languages and makes a best guess approximation based on the current language selected and then tries the other languages you have configured.
This means that if you're going along in, say, French, but then need to reference a word in Hebrew, you can swipe over to Hebrew, bang out a word, then swap back to French, type words in English (which get autocorrected in english) and then continue in French. This isn't really feasible in Gboard without a lot of back and forth.
Maybe Google updated the Gboard since the last time you used it, but it definitely predicts from multiple languages in the same sentence for the last couple of months at least.
The default one on Samsung will let you select multiple languages under "Suggest text predictions". Not sure if that's a clone of the default Google one, but I had so far assumed so. Seems to switch decently between Dutch and English, for me.
Yeah. Though decently is not the same as excellently.
Written on a SwiftKey keyboard used to write English, Swedish, Danish and Icelandic with minimal issues.
Though to be fair, I think the stock keyboard is probably almost as good these days and that the difference is that my SwiftKey keyboard has been trained on my writing for several years.
Though I have no idea more than that I quickly get annoyed when I try to type on the stock one.
I use GBoard with both English and Spanish installed, and I don't have to manually switch anything to swipe in both languages, or get predictive text in both languages.
> Happens to me a lot on Youtube too - you watch just one thing - and now your recommendations are full of things like that one thing.
Spotify's recommendation system is much better than youtube's (at least for me). I frequently "joke" that it is one of the few applications of ML that actively make my life as a consumer better.
Having been following Spotify's "Discover Weekly" for several years now, I'm actually really impressed how it manages to blend my long-term taste with recent moods. If I've been listening to one type of music for 1 or 2 weeks, there will be a noticeable uptick of it in the recommendations, while still mixing in less recent tastes.
As a father of a young child, Spotify's algo's are completely useless to me. If their ML is so smart it should be able to determine that Row row row your boat doesn't mix well with Anthrax. Wish I'd be able to toggle 'don't recommend kids music' somewhere.
I used Spotify to play music at a children's party. I am still getting recommendations for "Happy Birthday" months later. According to Spotify, it goes very well with Sheryl Crow and R.E.M.
I would argue for the vast majority of people language is not the primary indicator of what they consider good or bad music. I would assume it doesn't go into the algorithm at all and I certainly would not want it too. I have discovered some really cool music that I'm not even sure what language they are in this way.
Well i think the statistics are very strongly in my favour because English pop music is very popular all over the world and there are plenty of countries where the majority of people don't speak English (and still listen to English music).
When I was young we (in all the kids my age) listened to plenty of of English music even though we didn't understand anything. And let's not even talk about the plenty of examples of songs in English where even native speakers don't understand the majority of the lyrics.
Are you American by chance? Feel free not to answer, but living in the US I was surprised to learn friends in other countries commonly listen to music they can’t understand (most commonly English, sometimes German).
Put things in perspective for me - we really live in a bubble here. I guess Spotify could help reinforce that if that’s what the market wants.
I think it's likely you're right, although I haven't looked for any hard numbers.
But anecdotally, I only speak English and I listen to lots of Finnish, Swedish, German, and Norwegian music.
I don't know if Spotify just started recommending this stuff based on the few metal bands I knew from those places, but it started recommending other genres too and I really like lots of it. Not sure how I'd find this music otherwise, actually.
That's exactly why I enjoy listening to foreign pop music: I cannot hear how vapid the lyrics are.
But besides that, great music does not need to spell it out to convey meaning and emotion.
If the appreciation of a foreign song keeps growing, I will eventually look up a few translations. Their mixed interpretation usually only enhances an already enjoyable experience.
Can be helpful to turn off watch history during such sessions, or to remove the videos from your history afterwards. Agreed that there should be an automated solution, but you can avoid it today will a small amount of work.
Are we talking about the same Spotify? Because the only reference I found to removing songs from my recommendations is a thread from 2020 that says it is not possible.
I set up the browser session we use as a TV with profiles entirely to prevent any more of my spouse's yoga videos from leaking into my carefully curated feed of machine shop footage and construction videos.
Spotify does have an incognito mode to prevent this, but you have to carefully enable it beforehand. Why you can't just delete stuff from your history is baffling.
I agree. There are some playlists and songs I only listen to in the gym, and similar songs constantly float to the top of Spotify's recommendations for me due to gym being 3x week.
I did consider creating another "gym-only" user on our Family plan, but Spotify should really have a way to create named contexts for the one user. e.g. commute, gym, run, working from home, on a plane, etc.
I actually watched someone give a presentation at spotify on exactly this when I worked there. This was probably about 5 years or so ago. I have no idea what happened to that project. Probably the biggest issue would be getting people to give you location permissions but they also talked about working around that. It's a big org though they kill off ideas all the time.
There's a lot I like about Pandora and a lot I don't. The "stations" you mention can be pretty good. The problem I have is with the app itself. It's buggy and learning to use it well can be frustrating. A simple example would be when using the back arrow to return to where I was in the app. If I try to go back to a previous screen, it shrinks the app and when I resize it, Pandora restarts from scratch. This seems to take forever and it happens just about every time I use it. The only reason I haven't deleted it and switched to another app, is that I can't be bothered with all that while I'm working or driving (the only time I use it). Of course, there's other, more common issues like searching for a favourite song and only finding live sessions or worse, finding out they've stopped carrying a certain artist on my playlist.
Seems to be an unsolved problem to train the algorithms for recognizing different situations. One recommendation for all roles the user has. Though, thinking about, it's probably unsolvable as long as the interfaces remain simple and focused on satisfying only the one user, not the different roles, which would complicated the interfaces.
I think this is fairly well solved from a mathematics point of view (high-school level k-means clustering). The unsolved bit is simply how we get the Spotify et al. product managers to care.
All they need to do is filter out content with genre=kids from recs. They could create a separate kids recommendation item if people actually want that.
This wont happen whilst they have family and kids’ account types. Charge parents more for the premium option of not filling their carefully curated genre recommendations with nursery rhymes and Ed Sheeran.
It's fairly trivial to keep the recommendations consistent depending on the latest request, not just the logged user. Pandora does a good job at this. We have it on our Alexa and at dinner, we take turns with the kids requesting songs. If we stop requesting, it keeps playing an internally consistent series based on the latest song we asked for. So if it's the kids' choice, it's a never ending row of Kids Learning Tube :-)
From a technical point yes, but profiles are too rough and hard to use for this. What I mean is more some way to automatically maintain a kind of sub-profile, for each different aspect of a user, but exposed user-friendly and effortless. Most people won't maintain a separate user profile just for recommendations, as it's too much work for too little benefit.
It should be noted that in a Tesla, changing the Driver Profile (stores settings like seat and mirror positions) also changes the logged in Spotify account.
Both Spotify and Apple Music algos have no idea about music mood vs time of the day vs activity. Like late evening before sleep being perhaps not a good moment for heavy metal. And running not a right time for slow classical music.
Which is weird in case of Apple Music because Apple knows exactly if I am sleeping, running or driving a car - just from reading my watch.
If they're working on it then the point still stands that they don't currently have it as part of their product.
Time of day aware recommendations is something YouTube seems to have had for years. It always knows what to give me up top based on if I'm sitting down for dinner or lunch or if I'm looking for an audiobook for bed etc
Or 3) they run loads of A/B experiments to optimize engagement or some target metric and they've reached a local minima and are unable to escape it without a lot of political will or Product Managers willing to stick their necks out.
Spotify has supported this for many years - it's called Spotify Premium Family[0]. You get 6 separate accounts for $16/mth. It is worth the price of admission just to keep my listening history / algo feed clean.
That's a good tip though I'm not sure that there is an option for it in the car when I most often get a request for nursery rhymes, or more recently Disney songs.
I'm trying to avoid this outcome by playing as many different genres other than "kids music" as possible to my baby. Probably won't know whether it worked for a few years but in the mean time I get to listen to real music.
That sounds like a great argument for easy profile switching, along with an option in your play history for "move this play history item to this other profile".
I keep hearing this, but it hasn't happened to me yet. I wonder if it's because my Spotify history without kids is ~10 years vs ~2 years with?
We listen to a lot of Disney music, etc in the living room on my account, but I've never had that type of thing show up in my Discover Weekly or any of the Daily Mix playlists.
I will be very displeased if/when it does happen though.... Discover Weekly is a major reason I've paid Spotify for so long.
Ha, my life exactly. Ever since my kids used my Spotify account to listen to music my discover weekly list has unbearable amounts of paw patrol, baby shark and others mixed in. Now to be fair it still puts in music that fits my taste as well (and often my kids also like that music too), and considering that this is one profile I don't know what it could do better.
A lot of the time it's not an algorithm in charge. Most of the time now recommendations are based on who paid to promote their song or podcast.
Algorithms cannot be left running totally when sponsored ads can be purchased on the fly by content creators and musicians unless a site is lying to ad buyers about ad effectiveness.
The easiest thing for them to do is simply exclude Kids music from their recs. Kids music doesn't benefit much from it anyway. Apple Music does this as do most video services.
Don't these end up in separate mixes, though? My 13 year old makes a lot of grime requests on road trips etc. and now I have a separate grime mix under my top mixes.
It does it relatively OK with the daily mixes, but daily drive and top of the year don't. That's why I don't really get it, they already know fairly well what kids music is so let me just toggle it off?
My brother who has young kids did the thing where you can make a combined playlist with me. He has indeed kids music littered all over in the shared playlist but moreover he also listened to a 100 track audio book on Spotify (which was not a great experience since it was published as an album and did not have bookmark support). The algorithm thought: oh he must REALLY like that album because he keeps on playing tracks from it, and so chapters from this book were also in our combined playlist
Can't really complain if they offer Spotify Family which gives you 6 separate premium accounts, I think it's only double the cost of a normal Spotify account.
So the solution is paying double so I can log out of my own account and into an account I created for my 2 year old so we can listen to a disney song? And when I want to listen to my own songs again then I have to log out of my kids account and back into my own?
Yes, if you want multiple accounts you pay for multiple accounts. It's a solved problem you just have to pay the $9 a month.
I have two children, myself. I just installed second account on an old phone which has no SIM card in it, it has all the kids playlists and I don't have to worry about them getting recommended any of my 90s thrash metal. :)
That unfortunately doesn't solve it. There will always be moments where kids will want to ask for songs on your Spotify, like you're renting a car and have your phone hooked up.
> Spotify's recommendation system is much better than youtube's (at least for me). I frequently "joke" that it is one of the few applications of ML that actively make my life as a consumer better.
I've had the opposite experience with Spotify - I'd say my discovery of new music has withered to almost nil since switching when Google Music shifted to YouTube.
The algorithm just churns stuff I've already listened to, or suggests artists with (consistently) two songs that feel like Spotify's own fake artists - session musicians generated to capture royalties in-house.
I have the same problem with Spotify, just the same stuff over and over and over. YouTube Music has a much more diverse recommendation approach than Spotify does in my experience.
True, but it makes for another good use: building a live playlist. Youtube would cram all random things with the couple of novel interesting clips, while Spotify will keep the mood until I decide to switch.
Spotify's algorithm works really well until you let someone else use the account. I wish there was a "child mode", "party mode" or whatever to disable updating recommendations.
Pandora gives people granular access to the songs that are being used as the basis for recommendations. You would just delete the songs used by guests.
I turned off youtube recommendations years ago so I can't comment on that, but spotify's have pretty much always been terrible in my experience.
Anything suggested on the front page is either songs i already have in my liked songs or completely out of place. And the songs that i've already liked are from a few specific genres and artists, which seem completely random.
A few examples:
* Ratatat and Röyksopp both are in my top 5 of all time according to Spotify's own stats, yet I never got any suggestion about E.VAX, Kunzite, or Röyksopp's releases (their Lost Tapes playlist and their latest album, released about 4 months ago).
* the 'recommended for today' section is regularly filled with random synthwave when it's a genre i barely ever listen to, and a lot of 'electronica/trance/organica/deep house/whatever you want to call it', which i don't really listen to either (at least this kind of electro music).
* still, the worst offender has to be podcasts. I have zero interest in podcasts, have never clicked on any and likely never will, yet it's always the first thing on the app homepage, just below 'recently played'.
I've tried to use the browsing categories but these are most of the time just as poor, they only contain a few playlists and the latest big releases of the genre.
Highly subjective, but i'm tired of personalized suggestions & feeds. Just because I watched or listened to something does not mean I want more of it. And imho 'just use a no history session' or 'click not interested' do not solve the problem, especially since the argument in favor of recommendations seems to be that it is 'more convenient for the user'.
Why can't I just browse instead? Spotify has a very extensive way of categorizing songs based on multiple characteristics (https://developer.spotify.com/documentation/web-api/referenc...), so why can't I just use that when searching for new music, directly in the normal app?
All these recommendation engine problems people aren't discussing here aren't bugs or problems, they're features. They're either pushing the content that makes people engage with the platform the longest, or that promote another feature that the platform is trying to push on you. With YouTube, it's MrBeast style videos, influencer bait, and shorts. With Spotify, it's for sure podcasts.
Podcasts are so heavily pushed by Spotify because they're trying to make themselves the centralized one stop shop for your audio consumption. They didn't pay Joe Rogan millions for no reason.
As a man under thirty, I see the JRE logo on the front page of my Spotify at least five or six times a month. I have no interest in listening to podcasts on Spotify. I very much dislike Joe Rogan. I hit the option to not request it to me again, and yet there it is.
They're not recommendations based on your taste. They're a mix of just enough of your taste so that you trust and buy-in, mixed with whatever flavor of the week the recommendation engine would love to sell you.
> Spotify's recommendation system is much better than youtube's (at least for me). I frequently "joke" that it is one of the few applications of ML that actively make my life as a consumer better.
I hear this all the time but Spotify just plays stuff I've listened to. It's not a discovery service for me, it just plays the hits.
I'm wondering if that's just what it is and everyone likes it because it's playing stuff they already like for twelve tracks and then one new song.
std::map (as opposed to std::unordered_map) uses a balanced binary search tree approach and thus guarantees O(log n) lookup for anything that has a well-defined total order (i.e. comparison function), without worrying about hash functions, etc.
> "the /usr split is there for a reason!". No, it's just an historical quirk.
It's a historical quirk on linux, where there is no clear separation between "base OS packages" and "3rd party packages".
On FreeBSD the split is very real, anything in /bin/ ships with my OS and is maintained and updated by the FreeBSD team. Anything in /usr/bin/ comes from ports and is thus a 3rd party package I installed and can be safely nuked and I need to maintain/update it.
> It's a historical quirk on linux, where there is no clear separation between "base OS packages" and "3rd party packages".
It was a historical quirk to start with. At Bell Labs, back in the early 1970s, Unix was being developed on PDP-11s with RK05 hard disks (with removable disk packs), which had an amazingly generous capacity of 2.5MB each. The Unix operating system had grown too big to fit on a single RK05 disk volume so they had to split it across two. Other operating systems of the period faced similar issues, but dealt with them in (arguably) more elegant ways – on IBM mainframes, OS/360 maintained a database ("catalog") mapping file paths (dataset names, to use the proper terminology) to volume names, so you could move a file to another disk without changing its path. True to Unix's penchant for simplicity, its authors decided instead to just split the OS into / and /usr. And the split survived long after they'd upgraded to more spacious disks.
Any other explanation for the split is essentially a retcon. Some of those retcons (even if, as other commenters have pointed out, not your own) may actually have become true – some of them may have been approximately true to begin with, and they influenced people's decisions, thereby making themselves more true over time. But its ultimate origins will forever remain this quirk of computing history.
Funny aside: yours is an excellent comment, and yet proof that you didn't read the article, as the first part is almost word-for-word identical to the post.
I don't mean to shame you, I sometimes comment without reading TFA, and in your case you add a few more details that were not present in the article. I just found it interesting.
A much better separation is achieved in a few Linux distributions where every package is installed in a separate directory.
All the files that might be expected by others to be in certain standard locations are sym-linked to those locations, e.g. the executables to /usr/bin,/usr/sbin,/bin or /sbin, in order to appear in PATH.
In this case you no longer need any kind of database to know which files may be safely nuked to delete any package.
Moreover, in FreeBSD there is no such separation between the "base OS packages" and "3rd party packages", implemented as a difference between root and /usr. You might have misremembered /usr/local, which is indeed a place for "3rd party packages" in all UNIX-derived operating systems.
There are many "base OS packages" that are installed in /usr/bin or in /usr/sbin.
In any FreeBSD system, you can see their source files in /usr/src/usr.bin and in /usr/src/usr.sbin.
I have been using FreeBSD for a quarter of century, since FreeBSD 2.0, and there has never been such a separation between root and /usr.
The separation between /bin and /usr/bin and the other similar pairs was made only to allow /usr to be unmounted, when it is on another device than the root device, but still have in the root file system the minimal set of tools needed for diagnosing and repairing any broken file system or network connection.
In ancient FreeBSD installations it was always recommended to have a separate small root partition, e.g. of a few hundred megabytes, and some large partitions for usr and var.
This original use has become completely obsolete, because now, for diagnosing and repairing problems, it is preferable to boot from an USB stick or from the network (using a ramdisk as root file system), and then run diagnostics or repair programs without touching even the root file system unless modifying it is intentional.
In FreBSD it might still be possible to put /usr on a different partition or device and then unmount /usr, but in many Linux distributions this traditional usage is broken, because some of the programs installed in the root directories need components installed in /usr, so when /usr is unmounted they stop working.
The split is even stronger on NetBSD, where /usr is the base OS and /usr/pkg what's installed by the user through pkgin (binary packages) or pkgsrc (ports).
Likewise, the system configuration goes to /etc while the userland configuration goes to /usr/pkg/etc.
All it takes to factory reset a NetBSD system is an rm -Rf /usr/pkg.
Business outside the EU, interacting with users in the EU are bound by the GDPR. There might not really be a way (currently) to impose penalties on those businesses for violations, but they are certainly bound by them.
The maximum fine allowed by GDPR is "10 million or 2% of global revenue, whichever is higher". The goal is to ensure the GDPR "has teeth" even against companies for who 10 million is a drop in the bucket.
Keep in mind that large parts of the GDPR were already law in many EU countries, meaning there's years worth of enforcement activity that you can lookup to see how similar laws were enforced.
And mostly that has not been "handing out the biggest fines possible" and more "fines scaled to how grossly you violate the regulation". Companies who try their best to follow the law, have good processes and respond promptly, get a slap on the wrist or even just a warning if they remedy the issue fast. Companies that blatantly violate the law and stonewalling regulators get the harsh fines.
The real reason isn't that the system would be GPLv3. The real reason is that GPLv3 includes a patent grant clause, which many companies (Apple included) are very worried about applying to them if they ship GPLv3 code, therefore they don't.
> ...does seem to suggest that the "garbage multiplier" effect of immutability is an ill fit for applications that also create a lot of garbage naturally
That actually depends on how your GC is implemented. For example, due to laziness+immutability, Haskell produces a lot of garbage and a lot of allocations. This is not a problem with the GHC compiler, as the GC design makes allocation cheap (effectively a bump pointer allocator) and GC cost scales with the amount of non-garbage (this is, like all GC design, is a trade-off that can get you into trouble with some workloads).
Swiftkey just lets me auto-complete/swipe in multiple languages seamlessly. Want to swipe 1 English word inside a completely Dutch sentence? No problem. The reverse? No problem. Using Google's keyboard the same way is endlessly frustrating.