Hugging Face is the most ethical AI company ever!
https://huggingface.co/DreamAngel/Hana_Bunny_v1.0
original archive: https://archive.is/stbxz
this is a model created to generate fake AI revenge porn nudes of a specific influencer
they switched off access only just now, because Bluesky was hammering the HF staff with “the fuck you’re an ethical company”
(CONTENT WARNING: pic of AI boobs)
404 added an update that the dataset was removed:
Update: Following the publication of this article on Tuesday evening, van Strien removed the dataset. “I’ve removed the Bluesky data from the repo,” he wrote on Bluesky. “While I wanted to support tool development for the platform, I recognize this approach violated principles of transparency and consent in data collection. I apologize for this mistake.”
So they named the product sucking the data after the Facehugger? At least they know that they are in the abomination business. Will they be releasing an AI named Bursting Chest?
The company was named after the U+1F917 🤗 HUGGING FACE emoji.
HF is more of a platform for publishing this sort of thing, as well as the neural networks themselves and a specialized cloud service to train and deploy them, I think. They are not primarily a tool vendor, and they were around well before the LLM hype cycle.
to be honest, they give me a lot of mtgox vibes:
- extremely stupid name
- technically predates the worst excesses of the AI bubble
- very eager to enable the worst excesses of the AI bubble
they also went full DARVO complaining how mean everyone had been to them over them abusing personal data and then telling us we needed to get with the program
I hope Blue sky sues them.
This is a bit of a nuanced issue though. The person merely published a dataset made from publicly available data than anyone can re-create themselves using the Bluesky Firehose API. Could it be used to train a model? Yes, but that isn’t the only use case and the person who posted it has no control over what other people use it for. If someone does train a model using it then that’s their legal issue to work out, not the publisher’s.
It’s the same argument billionaires were using to justify silencing people who posted the movement of their private jets. The billionaires argued that this data could be used to harass them, but the posters argued the data is public and they aren’t responsible for what other people do with it.
The legal system is the perfect place for working out nuanced issues like this.
If I were a lawyer and making this lawsuit I would argue that “publicly available” does not mean “public domain”, and that without acquiring usage rights for the data then you don’t have the right to use the data.
If the courts rule against a decision like this then that would mean that any website that hosts any materials that can be accessed without an account must then provide that material to any person who accesses it free of charge which is a gigantic consequence to this nuanced issue.
I’d just bury them in GDPR notices, which quite a few people did
My point is that you can’t talk about usage rights of a dataset without talking about a specific use case. The suggested use case was to provide a static test dataset for systems developed to use the firehose API, but the dataset could be used for literally anything from making funny memes (fair use) to training a LLM model (arguably not fair use). Does the existence of an illegal use case automatically mean the dataset itself should be illegal though?
As a collorary, a photocopier can be used to create unauthorized reproductions of copyrighted works. Should making and disturbing photocopiers be illegal because they are capable of and used in the process of violating copyright law, or should we accept the photocopier absent of a use case isn’t breaking any laws and go after the people who use them to illegally create unauthorized reproductions?
A data set isn’t like a photocopier in any meaningful way.
It’s not a tool, it’s information, and some of it counts as personal data under EU data protection laws.
