James Dreben
@jdreben.omg.lol
350 followers 770 following 350 posts
Check out my home page (jdreben.omg.lol) or talk with me to get to know me 👋 I’m more active on https://mastodon.world/@jdreben Reposts ≠ endorsement
Posts Media Videos Starter Packs
Reposted by James Dreben
quadraticink.bsky.social
My first thought was that this might be an over-reaction. Of course, bsky uses automatic labeling as part of their moderation of images and videos. It is impossible not to.

But looking at Hive's ToS and other documentation, there is no limitation on Hive using submitted content to train their genAI
jdreben.omg.lol
What’s up with this… Is 1) The Hive really being used, and 2) using images here for generative AI? Pretty bad look if true
To anyone thinking about joining BlueSky, especially artists: everything you post is used to train generative Al models.
BlueSky uses Al to label content for moderation, and to do that they use a company called https://t hehive.ai/. If you look through their privacy policy, you will see that they all content sent to them to train models for all their services, which include generative Al for both text and images.
It's a built in «feature" and cannot be turned off.
jdreben.omg.lol
Very interesting. Yeah in general I assume companies can and are incentivized to train on their customer data unless they explicitly cannot through license or agreement.

The main thing I don’t like is the lack of choice. I want artists to be able to opt out of a 3rd party training
jdreben.omg.lol
Definitely can all be scraped. And probably are by multiple sources.

The only question I was raising was whether The Hive specifically who they’re feeding all images to is doing it.
jdreben.omg.lol
Yeah I’m sorry. This definitely escalated beyond my control.
jdreben.omg.lol
Exactly. Honestly this was a really old post of mine, I think at the time the privacy policy did say they retained that right but I don’t see it any more.

I apologize if I have spread misinformation. I do NOT know whether or not they are using it for training gen ai. I just was worried 2 years ago
Reposted by James Dreben
gootarts.bsky.social
important note is that while hive does have genai image stuff, it's limited to stock stable diffusion and flux. they are not making new models and are not using bsky images to train them, they're doing software as a service for open source image generators.
Reposted by James Dreben
julietshen.bsky.social
seeing this screenshot and a lot of ₊˚⊹♡ discourse ♡⊹˚₊

1) this is from 2 years ago, privacy policies change
2) pretty sure hive deletes customer data after like 2 weeks

hive is NOT generative AI, they use AI models and traditional ML for spam and abuse and safety purposes
jdreben.omg.lol
What’s up with this… Is 1) The Hive really being used, and 2) using images here for generative AI? Pretty bad look if true
To anyone thinking about joining BlueSky, especially artists: everything you post is used to train generative Al models.
BlueSky uses Al to label content for moderation, and to do that they use a company called https://t hehive.ai/. If you look through their privacy policy, you will see that they all content sent to them to train models for all their services, which include generative Al for both text and images.
It's a built in «feature" and cannot be turned off.
jdreben.omg.lol
I would still like to see something in their policy that says they will not train on customer data. Has anyone seen anything about that?
jdreben.omg.lol
That is really good. I feel terrible if this old post of mine is spreading incorrect information 😥 I will keep looking into this and I guess delete so as to not spread misinformation if so.
jdreben.omg.lol
You’re right I can’t either any more. I would note I originally posted this 2 years ago, and the privacy policy was last updated in May of this year.
jdreben.omg.lol
I don’t disagree per se. The issue I have is primarily if the labeling service they use also trains on those images.
jdreben.omg.lol
I think the reason this post of mine is being revisited is that someone from Bluesky today confirmed they use The Hive for auto labeling images.

So it’s really a question of does Bluesky’s deal with Hive *prevent* The Hive from training on images? Because their normal policy gives them that right
jdreben.omg.lol
Indeed :(

I hope they remedy this but I am not expecting them to at this time
Reposted by James Dreben
faineg.bsky.social
You don’t need to become an expert on AT Protocol, you don’t need to learn to code, but I really do want to impress upon people that “federated social media” is a vitally important concept in a time of global authoritarian and technofeudalist consolidation.
jdreben.omg.lol
Yes I agree with you completely
jdreben.omg.lol
You raise good points. You definitely have got me feeling a bit of a smarmy shit and wanting to correct myself.

for what it’s worth, I am willing to change. :( I want Bluesky to succeed in decentralizing. I fear it being bought. it has been easier for me to understand Mastodon’s decentralization
jdreben.omg.lol
I can’t edit. So I will go ahead and delete to not further misinform. Thank you again
jdreben.omg.lol
Woah! Thank you for corrected me. I will edit and probably delete the original message. This is great to hear
jdreben.omg.lol
A 2 way check sounds good. I don’t see why that goal had to be accomplished by building in a way for “trusted authorities” to circumvent a 2 way check
jdreben.omg.lol
Sure, you need to have that domain have an A tag pointing to your account. Sounds like you already understand this. Is your question why i think circumventing this check is a bad thing?
jdreben.omg.lol
This approach is “mocking” 2 way verification, ie reproducing the results without the true checks.
jdreben.omg.lol
When we are referring to “self verification”, what we are really referring to is 2 way verification. I say I am associated with a website. And the website says I am associated with the website. Check.

What BS is rolling out is 1 way verification. BS or one of their appointees says if you get check