To anyone thinking about joining BlueSky, especially artists: everything you post is sent to a third party for AI labeling.
BlueSky uses AI to label content for moderation, and to do that they use a company called https://thehive.ai. If you look through their privacy policy, you will see that they can use content sent to them to train models for all their services, which include generative AI for both text and images.
Update: https://meow.social/@FluffyDeveloper/110652053858910840
@digslogic94 yeah :(
Nothing stops people from scraping Mastodon as well, but at least it’s not a built-in function x3
@toboearticwolf alas yeah, the code is open source so you can check it yourself: https://github.com/bluesky-social/atproto/blob/main/packages/bsky/src/labeler/hive.ts#L11
I was also curious to try it because of the “choose your own algorithm” function, but finding this out just killed it for me :(
@FluffyDeveloper oh wait what? That's... not good!
Tbh, I still don't really know what BlueSky is, but I've seen that name often.
@SteffoSpieler It's another social media network created by the same people who backed Twitter and Nostr. Instead of going with ActivityPub however, they decided to make their own VC-backed protocol called AT.
The protocol itself is not too bad, but its controlled by private companies and BlueSky has decided to basically avoid moderation and offload everything to AI labelling.
In short: it's decentralised Twitter with venture capitalists trying to control their own fediverse :/
@FluffyDeveloper
that is really Shity so instead of integrating activity pub they made their own protocol to control their own fediverse that's not really a fediverse at that point that's more of a blue sky verse lol
but since the code is open source can somebody just yank it and just integrate activity pub and make a better site?
@epicEaston197 Yep!
And instead of doing actual moderation they just use AI to label content and leave users to fend for themselves.
At that point you might as well use ActivityPub with Mastodon, Calckey, or other. Even removing the AI labeling from other instances (if they are ever allowed) may not do much because that content will eventually make its way into the original main instance.
@FluffyDeveloper
okay but I wanna say this again as a separate reply is it possible to yank thesaurus code since it's "open source" could this mean that we can just get rid of the protocol and replace it with activity pub and just make a better site for the fediverse?
Similar to how Calckey is a fork of Misskey?
@Rhube @FluffyDeveloper @SteffoSpieler their current position is that they’re exploring blocking but are concerned about the “cost” (for whatever metric they’re using as a cost function) of blocking “at scale”.
Because apparently “kick off the harassers everyone blocks” is too much to ask.
@mic @PennyOaken @Rhube @FluffyDeveloper @SteffoSpieler
no, they didn't forget.
Distracted angry plebes fighting amongst themselves is -the- primary way exploitative pieces of garbage manage to get rich.
The "Golden Rule" of all exploitative assholes from pickpockets to VC bros is simple: Distract, Divide and Conquer
@FluffyDeveloper @SteffoSpieler they saw a demand for decentralized social media services, and wanted to make their own corporate spin of it to take advantage of consumers long term?
very sickening.
@skymtf Oh I know, BlueSky for me is just a no on all fronts unless new instances pop up.
Alas, it would be really hard to ban the original instance and its AI crap, because by the time they open up the protocol (if they ever do), it will be the one with most users :/
@FluffyDeveloper Thing is, I imagine quite a few are ok with that. I also imagine that AI training will eventually work back to an algorithm for Bluesky too, which I imagine even more people would want, at least to some degree.
What they offer with regards to generative image AI seems a bit weird with regards to that, I'd guess it's just bootstrapped on top of Stable Diffusion? So they are basically training LORA's for it, or an equivalent. Curious how that might play back into Bluesky.
@southernwolf Oh definitely. People are still fine with using Facebook so AI training is not gonna bother that many people.
I would imagine they entered some kind of deal with BlueSky where they get “moderation” in exchange for data to train hive.ai models.
@FluffyDeveloper This is very worrying if true. I just got here and I'm still trying to figure it all out.
My concern is that if it's forcing AI on its users, and only people with those sort of right-leaning, libertarian values use it, then it is only learning from those people.
Boycotting companies, non-fungy bollocks and crypto bollocks might make those things go down in value, but I don't know if the same logic applies here. If it goes mainstream enough, people who don't like that stuff may just be forced underground.
I might be wrong, but I was hoping Bluesky would be the 'new Twitter' that fixes its mistakes and Mastodon would, if anything, just be a slightly clunky stopgap. This news makes me feel like the whole internet would split into two camps based on values and political ideologies.
@briiz BlueSky and it's AT protocol are definitely the new Twitter, in the sense that they are backed by venture capitalists who have a very vested interested in controlling everything that happens on the network :/
ActivityPub and Mastodon are here to stay I think: they are extremely easy to implement, and while not as groundbreaking as e-mail was when it was first introduced, AP is just too big a change to simply fade away.
@FluffyDeveloper I'm a little confused though. In some of your replies, you say that the point is to offload all responsibility for moderation onto AI and the owners of various servers running on Bluesky, thereby just automating their income and having to do very little.
But you also say they have a vested interest in controlling everything that happens on the platform? Which is it?
Also does it mean that eventually if people want to start up other servers/instances running on Bluesky, you can avoid AI that way? Will there even be tools to help you customise your experience to that extent?
@briiz They have an interest because they paid for it, so eventually they will sanitise so they can start making money out of it as with every company-backed social media thus far. They talk about "composable moderation” but it's really a fancy way to say racists are welcome :/
For the moment no other instances are allowed. The server code is open source though, so the AI functions could be removes, but if the main instance is big enough then posts and images will still make their way there.
@FluffyDeveloper Hmm. A lot to get my head around. But I think if everyone is using the main instance, doesn't that sort of defeat the point of a federal network in the first place?
I hope the media will eventually ask the right questions about this.
@briiz It definitely does, and I can't not be suspicious about the fact that they aren't allowing other instances yet.
Honestly, BlueSky is just too small to be interesting for the media beyond a few articles here and there. I have seen more than article about it though, and basically all of them criticise the hands-off approach to moderation.
@sirlan_ff00ff @briiz @FluffyDeveloper embrace, extend and extinguish pattern I believe you are referring to and...
Yep! I mean, it's like if you want good internet platforms it's better off being owned by the workers themselves instead of a megacorp.
@murm Here you go :)
Look under "How We Use Information We Obtain”, 4th bullet point.
@murm No need to, here is the actual code calling on hive.ai API.
https://github.com/bluesky-social/atproto/blob/main/packages/bsky/src/labeler/hive.ts#L11
HalDe 🧢🦕
Fluffy Developer Fox
Diglsy
19
Emmet O'Neill
ToboeArticWolf🏳️🌈
Steffo 
Charlotte 
/Cinny 
pollinator stan
JocelynAyodele
epicEaston197
Dr Ro Smith
Ms. Penny Oaken
Micky
CJ Paloma
Q.H. Stone
JeanineFinn
JellyOsaurus
Dick Smiths Fair Go Supporters
Shirley Eugest
Skylar Caulfield 
Southern Wolf 🐧🦀
TheNokMachine 
𝘽𝙧𝙞𝙞𝙯 ~
sirlan → @
trip
Murm is gone.