Someone finally explained what I was missing about how ATProto works and I'm definitely a lot less salty about it now. I'm a little pupset that it took this long to have someone explain it in a way that actually gets at what I was complaining about with it, but hey, I'm glad to be clued in at last.

I still don't like the did:plc system though. And don't come at me about did:web, that's not a suitable alternative to did:plc.

#ATProto #Bluesky

@Rusty I learned more about ATProto yesterday and I got even saltier. What made you less salty?

@scout I was under the impression that the interactions weren't stored on the PDS and were somewhere in the AppService stack so I was confused how you'd get the interactions without being on a given relay.

@scout Wait so why do you hate AT Proto?

I'm super curious what made it click for you!

@knowtheory.net The logbook for PDS's made it click. Making it so all the interactions are stored on and shared from the PDS makes a lot of sense. It's the missing piece I've been trying to get my head around.

@Rusty Because they keep pretending they're decentralized and federated and all those buzzwords. But when you look at it more, it really isn't. Take AppView for example. Who's gonna self host that? It takes a lot of resources. So you use theirs. They also have only one frontend/client app. I cannot take people seriously who seriously claim bluesky is federated to any degree. I'm not too knowledgeable about this stuff, but based on what a friend of mine who is told me, it's pretty bad lol.

@scout The thing I like about ATProto over ActivityPub is that it uses websockets to directly access media content. ActivityPub basically mirrors every content it sees which is incredibly space inefficient since every server has to effectively copy and host the data. That said, having a ton of people accessing a single websocket address if something goes viral seems like it'd light the person's PDS on fire, so idk

@Rusty @scout it’s an implementation detail aimed at increasing reliability against unreliable nodes on the network

you can absolutely have the client fetch things to some extent but it would mean that the instances themselves need to handle a lot more traffic

in principle instances just need to cache remote posts, in practice they create a copy and mastodon even copies the media for some bizarre reason

@Rusty @scout some AP implementations don't cache or at least can be configured not to