The amount of fake "legal issue" spam is ridiculous
Come on, the posts were public.
Why tf bluesky policies will apply to huggingface or random model finetuner. They apply only to the app itself.
Big companies are training on your data for a long time already, but now you're attacking this dataset made for fun.
Data was collected in legal way, because it's publicly available already.
Also alpindale here uploads this dataset from his name, not from Pygmalion. So Pygmalion's policies doesn't apply to him.
This data has not been anonymized AT ALL. I've said it over there, I'll say it again here
I do not care how large or small the dataset is. I did not consent to my data being used in this way.
These reports are not spam, they are legitimate users of Bluesky making lawful requests to have their data deleted. They created the posts, and simply publishing them on the AT Protocol network does not give people the right to train on them unless they have explicitly agreed to that. Public network ≠ public domain. Regardless of the ethical quandary, the dataset creator scraped posts made by citizens of the EU without their consent, meaning the dataset is very likely in violation of the GDPR. I hope HuggingFace takes it down before the EU knocks on their door.
This data has not been anonymized AT ALL. I've said it over there, I'll say it again here
I do not care how large or small the dataset is. I did not consent to my data being used in this way.
This is literally just spam
It is not spam. Take any DID ledger credential from the "author" tab, and insert it into the following link at the position specified by <here>.
https://bsky.app/profile/<here>
example: https://bsky.app/profile/did:plc:5ug6fzthlj6yyvftj3alekpj
It is not spam. Take any DID ledger credential from the "author" tab, and insert it into the following link at the position specified by
<here>.
https://bsky.app/profile/<here>
example: https://bsky.app/profile/did:plc:5ug6fzthlj6yyvftj3alekpj
Still harassment
Though I can agree that my initial statement that these legal requests are "fake", isn't very true because huggingface operates in eu and they probably will remove this data.
But it will just be re-uploaded on the website that isn't subjected to eu laws.
(If it's really illegal in eu, because data collected there is public)
I hope you downloaded the model in case they bully alpindale in removing it.
Permission was given by the post authors in Section 2C of the Bluesky Terms of Service.
Permission was given by the post authors in Section 2C of the Bluesky Terms of Service.
True
Though I can agree that my initial statement that these legal requests are "fake", isn't very true because huggingface operates in eu and they probably will remove this data.
But it will just be re-uploaded on the website that isn't subjected to eu laws.
(If it's really illegal in eu, because data collected there is public)
I hope you downloaded the model in case they bully alpindale in removing it.
already cloned, why not
The irony of the anti-AI backlash is that by fighting open AI, people are actually empowering the very companies they claim to oppose. When individuals protest AI training on public data, they seem to forget that tech giants have been cataloging and owning internet content for years. Corporate AI will train regardless - the only difference is whether the process is transparent (open AI) or opaque (closed corporate systems).
The real choice isn't between AI that uses public data or not-that ship has long sailed. The choice is between collaborative, potentially democratized AI development and closed, profit-driven AI controlled exclusively by Big Tech. Attacking open AI doesn't protect your data; it just concentrates the power of AI development in fewer hands.
Will we look back on today’s anguish over open content being mined to train neural networks and marvel at how blind we were to the importance of huge, open datasets for democratic, transparent AI? Will we say that the mining of open content is vital if we want models that are less biased than those trained on smaller proprietary datasets? Will we think that open datasets are also vital if we care about barriers to entry in the new world of the data-rich and data-poor? (Remember, that may be the single most important generative inequality in the market of the next fifty years, the one from which an entirely skewed new political economy flows. And you want to make it harder to have open datasets?)
https://openfuture.eu/paradox-of-open-responses/misunderestimating-openness/
At least please be honest. As I said elsewhere, I don't care either way. But this dataset is non compliant. It could become so with a script worth 20 lines of coding, but i guess it's more important to stick it to those "anti" than being remotely fair. That's definitely one way to shoot oneself in the foot.