Post
Another day at Bluesky, and it looks like growth is not slowing down! Welcome!
We’d like to take a moment to share our stance on AI and user data:
A number of artists and creators have made their home on Bluesky, and we hear their concerns with other platforms training on their data. We do not use any of your content to train generative AI, and have no intention of doing so.
Bluesky
@bsky.app
Bluesky uses AI internally to assist in content moderation, which helps us triage posts and shield human moderators from harmful content. We also use AI in the Discover algorithmic feed to serve you posts that we think you’d like.
None of these are Gen AI systems trained on user content.
November 16, 2024 at 2:17 AM
1.8K reposts
217 quotes
21.7K likes
Thank you for joining Bluesky, we’re happy you’re here!
Our full Terms of Service and other policy documents are linked below:
Unfortunately most people are overlooking the fact that BSky's tos doesn't have any restriction against them selling our data nor restrictions against 3rd party data collection. So while on the surface they appear anti-AI & wont do the scraping themselves, however they allow others to do it freely.
Tough talk lil bro..until the future updates saying you had to use AI for various reasons lol
Please understand that 'AI' is an incredibly broad term that applies to any computer program designed to do the job of a human and is not a new concept or term. Machine Learning is not a synonym, neither is image generation.
Can you clarify that the 3rd parties you are using for content moderation and suggestions (like hive) also won't use art shared here to train their AI?
Thank you for working to protect your team members from harmful content!
I assume your not against AI generative content or the sharing of it?
If there's content you'd think we would like, and those suggestions are based on AI, how does that happen without training the AI making suggestions on the data of users they're making suggestions to?
Alright, now just integrate glaze preprocessing or some form of protection against scrapers for uploaded data, and you'll win over the entire artist community on this platform!
"Translate" is very important, but now it's broken,fix it please.
can you do us a solid and make it so that you have to have an account in order to see posts, It should help in subtracting the use of AI stealing data and art and as it stands other AIs can still steal shit from this website as long as they can have easy access to it.
But is it not still being fed to AI? I'm still a bit confused and concerned tbh
Thank you for clarifying the difference between AI in behind the scenes operations and genAI because I've seen a lot of people who don't know the difference
I appreciate your transparency. Thanks for taking the time to let us know.
Everyone is asking the following question: who is/are the owners of Bluesky?
We will appreciate tranparancy.
Who is/are supporting Bluesky financially?
Again; transparancy is the keyword.
Would be nice to mention that the federated protocol can't prevent others from using the data to train, just like docs published on the web. Does the protocol include a 'robots' or other means to allow users to express intent about their content ( unenforceable - but still useful ) ?
Can you stop sending me Taylor Swift content? Please!
Can you use a different word other than AI
Machine learning systems and recommender systems and collaborative filtering are not things people talk about when they use the word AI
They are almost exclusively using it to refer to generative AI
I truly hope that if someone appeals a content moderation decision that was made by AI, a human looks at that appeal, rather than AI looking at it again, which seems to be the standard at other crappy social media sites. Or, even worse, them telling you they don't have time to review it.
I cannot understate the value of contextual info like this being communicated, not only at all, but in the language of the platform.
Thank you for sounding like straightforward humans.
What’s to stop a manufacturer of deadly electric cars from hosting his own instance and training his AI on the bsky firehouse?
The only good use of AI in social media! Screw you Gen AI!!!
Your moderation AI is shit, sorry for the bluntness but that's it
I would really like to see a "real ID verification system" to propell real people who put their reputations on the line, amplified over anonymous accounts
i was about to ask about the harms of what the last paragraph answers. bsky really looks like a legit good place on the internet.
dont let your algos and policies ban and suspend accounts talking about Gaza and genocide!
Thank you for not allowing our content to be scraped for a scam.
And also for letting us know why the Discover feed is so bad.
bsky.app/profile/bsky...
Bluesky uses AI internally to assist in content moderation, which helps us triage posts and shield human moderators from harmful content. We also use AI in the Discover algorithmic feed to serve you posts that we think you’d like.
None of these are Gen AI systems trained on user content.
There was a whole discussion about this some time back. That's the perfect use for AI - to protect the mods. Thanks!
Can you filter out people I follow from my Discover feed? I mean, I kinda already discovered those people, y'know?
Serious question. Why is my Discover feed mostly filled with people I already follow, instead of people I might want to follow? Right now it’s not very different from my Following feed and basically redundant.
Thank you. Gen AI is trying to murder human creativity rn and is hand-in-hand with the facists.
How are you dealing with armies of trolls and fully autonomous AI accounts?
This sounds great but I'm disappointed in your Discover algorithmic feed. I've done no interaction with any political posts other then blocking and muting posts/words to get it off my feed. Yet I'm still inundated with political posts and X screenshots even though I only interact positively with art
Well, you sure have a lot of hate filled, antisemitic content on your platform. So, I don;t think those AI regulators are doing their job very well
Well it simply isn't working. In fact it is garbage.
THANK YOU for doing this, seriously. This is exactly what AI should be used to do. Anyone against this can come talk to me & I’ll explain how destructive moderation can be to mental health.
in the dos days bulletin boards usually forbade flaming, that is "you are a blah blah woof woof!" which nobody wants to see. but it is hardly hate speech to say you hate a billionaire bandit, a war criminal, or a hypocritical politician who hurts the people. yt and fb and *itter all censor that.
Thank you for telling us; knowing this, I have unpinned the Discover feed as AI uses massive amounts of energy such that casual use is environmentally reckless. Strive to use this tool for important matters please and thank you! :)
different kind of ai. this is standard "ai" (algorithm) that has powered everything on the web for decades, not wasteful new generative ai
Fantastic clarification, however one thing I'd like for y'all to work on is the aforementioned Discover feed. It's got the same problem all feeds do in that it shows in large part content from accounts I already follow. I've already Discovered them, any way to filter that to make it all new?
the problem is discover can only know you've seen something when you see it /in discover/. there's no real way for it to know you saw something on following or any other feed. it's a really fundamental structural limitation in how feeds work from what i gather.
On basically every website that's got a feed? YouTube does it, bird app does it, insta yadda yadda. You'd figure it could flag accounts already followed but I dunno enough about algorithms to say.
i'm saying not on here. within bsky/atproto's design. not just,,,,in general on the internet
By sharing User Content through Bluesky Social, you grant us permission to:
Use User Content to develop, provide, and improve Bluesky Social, the AT Protocol, and any of our future offerings.
“any of our future offerings.”
And of course ToS can and frequently do change w little notice regardless.
Bluesky is focused on growth. The answer to virtually every user related question now will be very favorable toward users. It was the same with Twitter. By using a platform like this you implicitly agree to essentially whatever.
and as soon as you disagree you're free to take all your data to another app that's the whole foundation here! bluesky is the main atproto app but inevitably someone will make another.
The Discover feed has been getting so good recently! It's where I hang out the most on the app!
Solid use of the technology, and kudos for not training on user content.
Thanks for describing AI'S role as a tool for content moderation
When will this AI give me more chance to appear in Bluesky? 🥹 it avoided me. I have been here for more than 1 month and posted daily but nobody views or cares 😭 super smol artist.
"the ai" doesn't control who's posts show up when for the most part. there's no secret, no algorithm to please.
try asking to be included in the various art feeds! just message their creators, they're usually more than happy to add someone.
Give me a short and easy to understand primer on blue sky that I can share with my CANADIAN friends on Facebook. Many of us have quit using X and Facebook won’t allow us to share news.
I appreciate the use of AI in the triage. But in general we would very much like the following step to be juried moderation -- that is, a jury of your peers. If someone writes, "voting that way is slitting your own throat". AI tends to read this as a threat. Human beings can recognize a metaphor.
What if a community of users wanted to use their content to train a GenAI that they owned or leased to, for example, drive a custom feed they collectively managed and used?
We are excited to be on the platform.
Being responsible with user content is diametrically opposed to the new policy at X. A huge differentiator 👏
Can you include a way to appeal content that has been falsely flagged by your auto-moderation? I've seen people's icons and cartoon drawings get flagged and when I check, it's just two character's heads looking at each other, nothing even remotely suggestive or sexual in nature.
the poster can appeal labels placed on their post already
Anyway! If the auto-moderation gets it wrong on one of your own images, you can tap the "1 label has been placed on this content" box at the top of the post to get the appeals screen:
And if you're not the person who posted it, you can report it under Other and include in the notes that you're not reporting the post but it's labeled incorrectly. The auto-classifier is great 99% of the time and terrible the other 1%, and with this many images a day posted, 1% is a lot of wrong.
Absolutely everyone in the field hates automated image classifier systems, but not having them is even worse; it's a "there is no good answer" situation.
Is this going to be yet another site where I follow people but I'm not shown their posts and, instead, am fed a bunch of crap a machine thinks I want? Can't you just show me everything by everyone I follow and leave me alone to find what else I want to see???
that's exactly what the following feed is. the only default algorithmic feed here is discover.
I don't want an algorithm to feed me random crap!
The "Following" feed, and other non-algorithmic feeds are just for you then!
Interesting, so cleary and lets going on to watch how works.
Can you work on getting a function for users we want immediate notifications from ?
Thank you. Many of my friends are artists and have left the other site for fear that their work would be stolen and used.
Question: What pre-emptive measures do you have against bot accounts (Aside from user blocking)? How will you better moderate than other soc media etc.? Happy to read about this anywhere if publically available.
Is there a way to get notifications of posts by one’s favored follows?
just for fun have a "chaos mode" that just serve people things at complete random
that's the firehose. literally just browse posts as they come in.
Can we disable the AI from the Discover feature? We should be able to discover on our own. If not...when?
Sorry if this was already cleared up somewhere else, but is there or will there be a feature that enables you to reset your AI weights should you decide the Discover feed no longer serves you?
bluesky team ily ihope you find a hundred dollars in the floor today
This is a use of AI I actually kind of support (provided I'm understanding it correctly!), as with a large platform it's hard to have every post looked at by a person, but having an AI flag things for review is definitely helpful!
Why every tech company is using AI?
looks like there is no place to be safe from AI
"ai" in this context is just shorthand for algorithms. it's likely not the kind of ai you have recent concerns about. it's the stuff that's been powering suggestions and searches across the web for decades.
pls just fix the UI it looks cheap and isn’t smooth at all
At some point, malicious army of bots will try to activate here. Some clever AI-based software could help to keep that crap away.
What is bluesky's stance on sharing user data with 3rd parties? This is a concern that artist have with other websites they used to post on.
Just don't stop me from seeing the posts of people that I
intentionally follow, please.
Because that shit is annoying when other social networkls do it. Thanks.
I do like to read various points of view - for balance.
Honestly that’s fair sometimes stuff can slip through a humans eye , just as long as you don’t go full musk and add a shitty chatgpt knockoff and have it train off real artists I say it’s fine
I just want the same NSFW content I had from X lol. Show me the spam javelin!
Bluesky might not be working on GenAI but can we be sure the company will never enter into agreement with a 3rd party who will use the data for generative AI training ala Reddit?
Will your company be a leader in the industry and also provide transparency into your direct hire and contract content moderation workforce? For example, what contractors does Bluesky rely on for human content moderation?
What’s the point in this app and why am I here?
Thank you for specifying that not all @¡ is gen @¡, to this day I still see a lot of people getting that confused
Now this is more like it for how AI should be used!
Although, Is there's at least human fellows to sort out things out incase the content moderation AI screws up in some way?
It wasn't harmful, but it was something I'd definitely rather not see again. Hairy dude in a dog mask and dom harness with a pink nipple peeking through the chest hair. 😞
Yeah. Those are "pups". Adding "puphood" "pup" and "pupplay" to you muted words helps a little. A LOT of them post untagged dick pics/porn. Bro and I mute/block them on sight now. You may also want to add "abdl" since diaper sniffers also spam untagged porn here.
I’m offering myself here to be blocked cuz yeahhhhhh I’m definitely nsfw lol.
Do these AI's censor conservative speech like so many other platforms do?
As long as Nick Fuentes, Stew Peters, Andrew Tate and people like them are not allowed to infect this place we good.
My point is I hope Bluesky doesn't let raging white supremists, nazis and women degrading scumbags turn this place into a twitter cesspool.
I understand your concerns and indeed, social platforms should provide a respectful and inclusive space. I hope Bluesky can build a healthy environment where people can share different views. Everyone should work together to create a friendlier online space.”
Finally, a company using AI for good instead of evil.
You are right, art must remain created by a human and not by an AI robot.
It's almost as if you have some sort of guidebook on how not to do things.
X
Can you please add an option to turn off reskeets, or reposts?
Thank you, cool, I can still have the other options as well....
How can I tell the Gen Ai that I ONLY want timely factual things from competent trustworthy real individuals? And emergent local information about fires, gunshots and traffic closures?
If I could make one recommendation in terms of an effective use of AI.
Could you make it so that for any image added to your page, that the site will scan it and attempt to write alt text for the image?
This would be for the purposes of accessibility for the blind.
I personally think having the option to have others write the alt text for you is probably a better route to go down. It can even have a system in place in order for it to be non-exploitable.
Having it be generative might make artists uncomfortable, given that there has to be a database to store the information in order to make the description accurate to the image.
You may be the first organisation to prioritise using AI to do the jobs we don’t want to, rather than to do the jobs we want to. Very refreshing!
Thank you for the continued clarification bluesky. As you might guess we are a concerned bunch due to the existing platforms following the path of "deal with it" when it comes to TOS changes. Very much hope bluesky remains awesome and transparent :) in the future
We wish for your further development, we want you to cross the score of 10 million daily.
And they’re actually doing something to protect it, too! A+++ in my book
the main concern people have about AI usage for art is people training models to create art (generative ai, abbreviated as gen ai here).
bluesky is saying that they do not do this.
what they do do is use ai models to assist in moderation and removal of inappropriate content
so e.g. perhaps there is a model that can detect if an image contains gore, trained on a set of images that someone has gathered and classified into 'gore/not gore'
Also, the Discover tab is more or less the same type of algorithm that almost every page that recommends content to you uses
