Picking up a bit on what @u2764 has been talking about...
The public timeline is what gives Mastodon value. It's what allows it to escape the chicken-and-egg trap that prevents new social networks from taking off, by giving the system discoverability.
But, as instances get bigger, the PTL becomes progressively more useless; not just because it accelerates. but because it becomes an anonymous crowd instead of a community. (Cont'd in replies)
Like, the influx of Argentine Mastodon is very illustrative of this. If you halfway understand Spanish like I do, you'll realize that in actuality most of those toots are actually profoundly banal; whether or not there is a language barrier, they are very much outside the wavelength of Mastodon's OG community. Since at this stage in the network's life, growth comes in bursts that are generated by spots of media attention, this will be a pattern with new large inflows of users.
People are going to talk about salience-sorting algorithms as a solution to this. Those people are bad, and wrong; algorithms for "relevance" sorting are, I'm pretty convinced, inherently evil.
The real solution is user self-sorting into instances that would operate as communities, which has already happened.
@brunodias The followbots did provide a useful service, at least early on, by "force populating" the public timeline, esp. for smaller instances. But you may be right now - they may have outlived their usefulness.
Also, you're absolutely right re: relevance algorithms. No algorithm is going to be worth anything unless a) it has a very powerful AI and b) it knows you better than you know yourself (a scary thought!)
@brunodias People are already switching instances. Even in Argentine Mastodon non-technical users are already figuring it out really fast.
The banal nature of the .ar toots in .social is not so much that it's less of a community — on the contrary! people already know each other's names and even chat in the PTL without @ mentions — but that it's a different kind of community.
@brunodias The thing with .social is that it is pretty much a mixed bag. There's the early adopters, now a minority, and several groups of users from the various moments that @Gargron opened subscriptions, each incoming group from a different surge and with a distinct feel.
Some other instances have much more consistent constitutions, I suppose.
@hisham_hm yeah, pretty much my point. I kind of wonder what the point/future of .social is, given that it's always going to be this weird mixed bag
@brunodias .social will lose relevance over time; it will still be "the original dev's instance" but other instances will have more relevant content for most people. If Mastodon continues to crack into the mainstream, eventually some groups might even move away from .social to hang out in their country-specific instances (e.g. the one where the celebrities are, etc.)
@hisham_hm Not even country-specific, interest/community specific.
@brunodias yes yes. I mention country because language is an obvious barrier/filter.
@hisham_hm It is but like, witches.town is a successful bilingual instance. I think people are probably less annoyed by posts they don't understand than by posts they understand and don't care for (see: me)
@brunodias but witches.town is tiny. I'm thinking what will happen when we get to 1M user instances (which is only 12x bigger than the largest ones now... and the Mastoverse was 12x smaller in the beginning of this month).
@hisham_hm The PTLs on a 1M user instance would be completely useless, though, so they'd probably not even be available to look at. So if your instance was that large, it wouldn't really matter whether it was all people who spoke your language because only your home timeline would matter to you.
@brunodias I would think that GNU Social-style !groups would address filtering and gathering around common interests, but @Gargron finds them redundant with hashtags. So maybe it's a matter of coming up with a proper UI for (federated?) hashtags-as-timelines.
@hisham_hm Being able to pull out a search into a column (like you can on tweetdeck) would be awesome, though I still think small instances are particularly desirable anyway; again, it's the "oh I didn't know I wanted this" factor of being in that space that adds value.
@brunodias Oh, don't get me wrong. I think small instances are here to stay and will be a core element of Mastodon, whether it goes fully mainstream or not.
@hisham_hm But also, like, I think that if Mastodon goes fully mainstream, not everyone is going to want the same experience and I can imagine some users choosing 1M+ user, no PTL instances because they only want to follow their preexisting friends and celebrities or (god forbid) even brands. But the PTL is a core part of the "traditional values" of Mastodon.
@brunodias I believe 1M-instances will still have a local timeline. The federated timeline is already the same kind of crazy noise that you'd have in a 1M-instance. It's a "random sample" of the network. In a language-specific instance it would even make sense in times of major public events.
@hisham_hm I think at 1M users it would just go too fast, and it would probably also have really bad performance implications for the instance itself.
ESPECIALLY in a single-language instance, because then everyone would be on the same time zone and it would just be going way, way too fast at peak activity hours.
@brunodias that's a UI thing. The web interface currently auto-refreshes (when not scrolled), but Tusky for example doesn't, you need to pull for more toots.
@hisham_hm True, I guess disabling the auto-refresh would keep the PTL interesting in very large instances.
@brunodias The problem with a third party deciding what is relevant for me is that this is a lot of power.
Newspaper had this power too, but I could decide to read another paper, if I wasn't happy.
Having a curated timeline for the casual user isn't inherently bad. It could be what sets different instances apart.
@peter Different instances will naturally be set apart by different user bases curating a different federated/local timeline, though. I'm just very skeptical of the possibility of algorithms being able to actually improve user experience; it certainly hasn't worked for the various proprietary social networks.
@brunodias I guess "improved" is very subjective. But I can think of very simple filtering that might make a timelime more relevant.
E.g. I could imagine an instance where the public timeline only (or with higher preference) shows Chinese toots.
You could have an instance that shows toots about tech stuff, while filtering out the food pictures.
This might not be an "improvement" to you, but it might be for the people who choose to use that instance.
The problems with users splitting into the "right" communities for them, though:
1. Switching instances has way too much friction for anyone with an account that is even a little bit established.
2. Followbots were a mistake, as they make federated TLS indiscriminate rather than a curated reflection of the interests of local users.