Opt-out form?
Not sure if it's legally required or not but I feel like it would be nice for people to be able to remove their posts from this dataset if they want to. Would you be open to adding an opt-out form and edit the dataset based on them @alpindale ?
Hmm
I guess he could.
I agree that the best way to solve the issue is to allow users to opt out, It's just the ethical way to do it, despite the fact that such data collection is legal outside the EU (or even inside the EU, I don't know about that).
@SpacehogSutra if you have nothing productive to add, shut up and stay out of it. You have made your argument abundantly clear by now, and have posted that stupid image too many times. We get it, you don't understand nor care about how the law works.
FYI, Clem is the founder and CEO of Hugging Face.
Here is Clem's post about the other Bluesky dataset that was deleted from Hugging face. https://bsky.app/profile/clem.hf.co/post/3lbvlyphqd22r
FYI, Clem is the founder and CEO of Hugging Face.
Here is Clem's post about the other Bluesky dataset that was deleted from Hugging face. https://bsky.app/profile/clem.hf.co/post/3lbvlyphqd22r
Not sure if it's legally required or not but I feel like it would be nice for people to be able to remove their posts from this dataset if they want to. Would you be open to adding an opt-out form and edit the dataset based on them @alpindale ?
dont post ur life online ;p no post no data
Not sure if it's legally required or not but I feel like it would be nice for people to be able to remove their posts from this dataset if they want to. Would you be open to adding an opt-out form and edit the dataset based on them @alpindale ?
dont post ur life online ;p no post no data
true
I'm not against it. If a user wants their data removed, they can send a PR with the changes they want (as long as they can confirm the data belongs to them). Since I'm not legally obligated to do it, I won't spend all day manually removing every offended user - but I am willing to review PRs if they contain valid proof that the posts they're removing belongs to them. Cheers!
I'm not against it. If a user wants their data removed, they can send a PR with the changes they want (as long as they can confirm the data belongs to them). Since I'm not legally obligated to do it, I won't spend all day manually removing every offended user - but I am willing to review PRs if they contain valid proof that the posts they're removing belongs to them. Cheers!
Based, you dont have to do it, but if you feel like it you can.
Just dont placate the people that have been harassing you.
Based, you dont have to do it, but if you feel like it you can.
Just dont placate the people that have been harassing you.
All in good sport! I didn't feel like I was harassed at all. People are so energetic these days. Did something good happen to them?
Based, you dont have to do it, but if you feel like it you can.
Just dont placate the people that have been harassing you.
All in good sport! I didn't feel like I was harassed at all. People are so energetic these days. Did something good happened to them?
Hey, it's good, i was just worried because of how they treated Daniel van Strien.
I mean, he doesn't have to do it.
That is not how it works when it comes to EU law
I'm not against it. If a user wants their data removed, they can send a PR with the changes they want (as long as they can confirm the data belongs to them). Since I'm not legally obligated to do it, I won't spend all day manually removing every offended user - but I am willing to review PRs if they contain valid proof that the posts they're removing belongs to them. Cheers!
Sorry, but this method doesn't fit the bill under the Blue Sky Developer Guidelines. Sending a PR is arcane to the vast majority of the population, and dependency on your discretion is not deletion.
I mean, he doesn't have to do it.
That is not how it works when it comes to EU law
Uhm, who are you replying to?
I'm not against it. If a user wants their data removed, they can send a PR with the changes they want (as long as they can confirm the data belongs to them). Since I'm not legally obligated to do it, I won't spend all day manually removing every offended user - but I am willing to review PRs if they contain valid proof that the posts they're removing belongs to them. Cheers!
Sorry, but this method doesn't fit the bill under the Blue Sky Developer Guidelines. Sending a PR is arcane to the vast majority of the population, and dependency on your discretion is not deletion.
Really rude tbh
Not just rude, but its kinda catty.
What makes people think that the Bluesky Developer Guidelines are relevant? Those only apply to apps and services that are federated on Bluesky, e.g., connections to other fediverse servers and what not.
What makes people think that the Bluesky Developer Guidelines are relevant? Those only apply to apps and services that are federated on Bluesky, e.g., connections to other fediverse servers and what not.
true
People post public data to the internet, others read it, others save it; screenshot it. These are not private communications.
What were they expecting? It's naive to assume this is the only person who scraped it. At least natural sounding AI is more benign than real life data mining being done on your real identities.
@clem quand un utilisateur demande à faire supprimer ses données, si le propriétaire du repo ne le fait pas c’est à l’hébergeur des données illégales d’intervenir sans délai et de suspendre le compte qui enfreint la loi. Vous êtes partenaire privilégié CNIL, et soumis à la RGPD. Ça serait vraiment bien que vous preniez les lois applicables (et votre propre règlement !) au sérieux plutôt que de négocier avec le proprio du repos qui de toute évidence n’en a rien à faire et se moque ouvertement de demandes légitimes.
@clem quand un utilisateur demande à faire supprimer ses données, si le propriétaire du repo ne le fait pas c’est à l’hébergeur des données illégales d’intervenir sans délai et de suspendre le compte qui enfreint la loi. Vous êtes partenaire privilégié CNIL, et soumis à la RGPD. Ça serait vraiment bien que vous preniez les lois applicables (et votre propre règlement !) au sérieux plutôt que de négocier avec le proprio du repos qui de toute évidence n’en a rien à faire et se moque ouvertement de demandes légitimes.
Pierre Laval?
I'm not against it. If a user wants their data removed, they can send a PR with the changes they want (as long as they can confirm the data belongs to them). Since I'm not legally obligated to do it, I won't spend all day manually removing every offended user - but I am willing to review PRs if they contain valid proof that the posts they're removing belongs to them. Cheers!
Yeah - that's NOT how copyright law works. You do not have a license to distribute those posts and comments, so you are actively violating the copyrights of every user you scraped. And now that you've been informed that you're violating those copyrights, continued distribution of these posts constitutes a willful infringement, which opens you (and Hugging Face) up to elevated penalties per infringement.
FWIW, I've already filed a DMCA takedown request with Hugging Face to have my copyrighted posts (author = did:plc:r3hl5udsbhenyneix6nnydod) removed from this data set.
so you are actively violating the copyrights of every user you scraped.
You can't copyright shitposting on social media.
I've already filed a DMCA takedown request with Hugging Face to have my copyrighted posts removed from this data set.
You don't have copyright to any posts in this dataset.
I'm not against it. If a user wants their data removed, they can send a PR with the changes they want (as long as they can confirm the data belongs to them). Since I'm not legally obligated to do it, I won't spend all day manually removing every offended user - but I am willing to review PRs if they contain valid proof that the posts they're removing belongs to them. Cheers!
Yeah - that's NOT how copyright law works. You do not have a license to distribute those posts and comments, so you are actively violating the copyrights of every user you scraped. And now that you've been informed that you're violating those copyrights, continued distribution of these posts constitutes a willful infringement, which opens you (and Hugging Face) up to elevated penalties per infringement.
FWIW, I've already filed a DMCA takedown request with Hugging Face to have my copyrighted posts (author = did:plc:r3hl5udsbhenyneix6nnydod) removed from this data set.
There's no copyright implications here -- you've already given permission to share your posts publicly via AT Protocol in section 2C of the Bluesky Terms of Service. If you don't want future postings included in public datasets, you should either not use Bluesky / AT Protocol for them, or see if they have a mechanism to exclude your posts from their distribution.
You can't copyright shitposting on social media.
You may want to review copyright law (US at least) so that you don't end up making an expensive mistake in the future due to your ignorance. A copyright is created upon creation of the content - no registration is required (but a registration is helpful if you take someone to court).
You don't have copyright to any posts in this dataset.
I found several in there that are being distributed without my permission.
You can't copyright shitposting on social media.
You may want to review copyright law (US at least) so that you don't end up making an expensive mistake in the future due to your ignorance. A copyright is created upon creation of the content - no registration is required (but a registration is helpful if you take someone to court).
You don't have copyright to any posts in this dataset.
I found several in there that are being distributed without my permission.
You gave permission in section 2C of the Bluesky terms of service.
There's no copyright implications here -- you've already given permission to share your posts publicly via AT Protocol in section 2C of the Bluesky Terms of Service. If you don't want future postings included in public datasets, you should either not use Bluesky / AT Protocol for them, or see if they have a mechanism to exclude your posts from their distribution.
I've given Bluesky permission to distribute those posts on their site and through the AT protocol. This dataset neither belongs to Bluesky nor is it being distributed through the AT protocol. What you think is happening is that the copyrights to those posts vanish upon publishing through Bluesky, when that is incorrect. Otherwise, why would section 2D reiterate that users continue to own their User Content, and is granting Bluesky (not @alpindale or Hugging Face) the next set of permissions in 2D sections i through iv?
I've given Bluesky permission to distribute those posts on their site and through the AT protocol. This dataset neither belongs to Bluesky nor is it being distributed through the AT protocol. What you think is happening is that the copyrights to those posts vanish upon publishing through Bluesky, when that is incorrect. Otherwise, why would section 2D reiterate that users continue to own their User Content, and is granting Bluesky (not @alpindale or Hugging Face) the next set of permissions in 2D sections i through iv?
You gave Bluesky permission to publicly distribute those posts via the AT Protocol, which does not have a concept of non-public content. Once publicly posted, courts have consistently held that copyright doesn't apply; publicly posted content may be scraped and used for whatever purpose the scraper wants, which is what @alpindale has done.
With regard to your other question, yes, you retain copyright in your posts, which allows you to make them available via other channels, sell them to a magazine if you wanted, etc. Has nothing to do with Bluesky's public distribution and subsequent uses following from fair use of those public postings.
@jimscard - whatever purposes does not include the unauthorized redistribution of my content. Courts haven't ruled that I can simply copy the entire website of CNN and share it elsewhere, just because it's publicly accessible. You're correct that it would be unlikely that I would succeed in a case where @alpindale trained a model that used that content. (Indeed, how would I ever know?) But that's not what is happening here.
In addition to selling my posts to other channels, my copyright also protects my right to sell them to folks building corpus for purposes like training LLMs. @alpindale including my posts in this training set adversely affects my ability to do that, and I expect a court would agree.
As for fair use, that's a defense that @alpindale or Hugging Face can claim for redistributing the content, it is not a positive right that grants them license to do so. It's not @alpindale and Hugging Face's decision whether this redistribution falls under fair use, that's up to a court to decide based on the four factor test, which this likely fails three of the four tests. (The fact that the work's been published already works in the redistributors' favor under the "The Nature of the Copyrighted Work" test.)
If @alpindale wants to do the ethical thing, it would be pretty simple to build a small web interface where folks who consent to participate can add their Bluesky handles and his code filters the data file accordingly. The fact that there's so much resistance to doing this suggests that this is less about responsible research (revisit the Belmont Report, esp. the parts about respect for persons and consent) and more about fulfilling other ulterior motives.
Sure thing bro, all social medias are selling your data left and right (while it's free here) and this is publicly available information not some sort of DM collection. Anyway stop wasting alpindale's time with your nitpicking, it's obviously more precious than yours.
German courts have recently decided that the collection and compilation of publicly available data in order to make them openly available to the wider research community is lawful and exempt from GDPR under the research exemption and exempt from copyright as long as it is not used for financial gain. So there could be copyright implications for downstream models, but not for the dataset.
Of course, law could be different in other countries, but i doubt it since GDPR is european and copyright is one of the most standardized legislation in the world.
See e.g. here: https://www.lausen.com/en/lg-hamburg-kneschke-v-laion-e-v/
German courts have recently decided that the collection and compilation of publicly available data in order to make them openly available to the wider research community is lawful and exempt from GDPR under the research exemption and exempt from copyright as long as it is not used for financial gain. So there could be copyright implications for downstream models, but not for the dataset.
Of course, law could be different in other countries, but i doubt it since GDPR is european and copyright is one of the most standardized legislation in the world.
See e.g. here: https://www.lausen.com/en/lg-hamburg-kneschke-v-laion-e-v/
The decision doesn’t cover personally identifiable information which is very clearly protected by the EU RGPD and includes, if nothing else, an absolute right to request removal of one’s PII from any electronic system, public or not. German laws are very strict when it comes to PII. We are not talking about a few anonymous images here.
But we are talking about mostly pseudonymous posters on a decentralised open network whose sole selling point is that the data is not controlled by a singular institution.
Additionally, there are still exemptions under Article 89 of the GDPR. So data collection is for legitimate research purposes it is not flat out illegal to collect even PII. And i have shared a first (perhaps surprising ?, idk) precedent.
Sidenote: It will be very interesting how the AI Acts demands for proper data bookkeeping and data version control etc. will play out with GDPRs right to be forgotten. You cannot comply with both at the same time.
Anyways, I thought that the above may be of interest because of all the regulatory insecurity that we are blessed with in europe. Personally, I think these datasets are very important for computational social science and ML research. We will see how all that plays out in court over the next few years.
But we are talking about mostly pseudonymous posters on a decentralised open network whose sole selling point is that the data is not controlled by a singular institution.
Additionally, there are still exemptions under Article 89 of the GDPR. So data collection is for legitimate research purposes it is not flat out illegal to collect even PII. And i have shared a first (perhaps surprising ?, idk) precedent.
Sidenote: It will be very interesting how the AI Acts demands for proper data bookkeeping and data version control etc. will play out with GDPRs right to be forgotten. You cannot comply with both at the same time.
Anyways, I thought that the above may be of interest because of all the regulatory insecurity that we are blessed with in europe. Personally, I think these datasets are very important for computational social science and ML research. We will see how all that plays out in court over the next few years.
You misread Article 89. Safeguards should be applied, you cannot just dump data that contains PII, especially:
- without user consent
- when users have explicitly refused consent ("data accessible to logged in users only" setting on Bluesky)
- when users have requested removal
The repo's owner has publicly stated they did it "for lolz". Which is definitely not a valid use-case listed in Article 89.
Thank you for the moderate and reasonable responses @clem and @alpindale
It seems to me that the issue is resolved.
- when users have explicitly refused consent ("data accessible to logged in users only" setting on Bluesky)
Is there such a setting? If there is, and users were included in the dataset, it would seem that they could have recourse against Bluesky, since sharing content to AT Protocol, which has no concept of non-public content at all, would seem to conflict with a setting restricting distribution. Additionally, there’s no mention of excluding some content from sharing via AT Protocol in the consent provided in the TOS.
If anything, this discussion has uncovered what appear to be serious flaws in their assumptions that people have about their content on federated services like Mastodon and Bluesky, which, like Usenet in the 1980s, are designed for the free sharing of content, with individual node owners deciding what content feeds their nodes would receive from others and which feeds they would share with the greater network.
This is the setting. Users that have it are included in the dataset. It's not a hard block but it explicitly shows whether users opted-in (or in this case: out) for their posts to be used and displayed to the world outside of the Bsky network. The owner of the repo ignored the setting.
When users agree to use the AT protocol, they still own their data, and agree for it to be used for the purpose of communicating on the network, and a few other cases. But that's not even the point.
The point is that electronic storage & handling of PII data are subject to very specific data protection laws. Collecting such data is subject to explicit consent (which users gave to Bluesky to use the network and its features, but did not give to other entities). If you don't have consent, you are not entitled to exploiting said data. And especially not when the data can be linked directly to the original user. It's really not that complicated. We can discuss for weeks whether it makes sense or not (Americans will say it doesn't, most Europeans will say it does), it won't change that it's the law.
There are even more laws that require whoever hosts said data to remove them on request by the owner of the data even when the user previously consented (consent withdrawal), and entities must comply. HuggingFace has a legal obligation to take action.
The fact that some companies do it secretly is also not an argument. It's illegal.
As a Certified Information Privacy Professional, I’m quite aware of privacy laws around the world.
Thanks for providing the setting. It clearly states that it is a Bluesky app setting only, and that it may not affect anything else, which makes sense since the AT Protocol only has a concept of public content. So this setting is irrelevant to this discussion.
The issue I have with all these arguments is that I’ve come to the conclusion that the terms of service and design of Bluesky, the AT Protocol, and other similar federated services clearly indicate that they are public platforms. Content shared on these platforms is publicly accessible and shared publicly. There’s no expectation of privacy in public spaces. Consequently, I believe the rest of these arguments fall apart like a house of cards.
What’s happened here is that a bunch of people left a commercial social media site, which even restricts access to its API, for an open, public service that is designed to, and discloses that it does, publicly share such content to anyone, which includes identifiers of the author of that content, without considering the implications of what means.
An analogy could be someone who was filmed doing something in a public park and it was subsequently shown on the news. Now, they’re attempting to erase the footage. However, since it was in public, the television station can likely disregard their request, and all the individuals who recorded the news can retain the footage as well.
It could also be argued that consent is irrelevant to the discussion, considering the public nature of the content and the likely legitimate interest or public interest as the basis for processing, rather than consent (which is the weakest basis anyway).
As a Certified Information Privacy Professional, I’m quite aware of privacy laws around the world.
Thanks for providing the setting. It clearly states that it is a Bluesky app setting only, and that it may not affect anything else, which makes sense since the AT Protocol only has a concept of public content. So this setting is irrelevant to this discussion.
The issue I have with all these arguments is that I’ve come to the conclusion that the terms of service and design of Bluesky, the AT Protocol, and other similar federated services clearly indicate that they are public platforms. Content shared on these platforms is publicly accessible and shared publicly. There’s no expectation of privacy in public spaces. Consequently, I believe the rest of these arguments fall apart like a house of cards.
What’s happened here is that a bunch of people left a commercial social media site, which even restricts access to its API, for an open, public service that is designed to, and discloses that it does, publicly share such content to anyone, which includes identifiers of the author of that content, without considering the implications of what means.
An analogy could be someone who was filmed doing something in a public park and it was subsequently shown on the news. Now, they’re attempting to erase the footage. However, since it was in public, the television station can likely disregard their request, and all the individuals who recorded the news can retain the footage as well.
It could also be argued that consent is irrelevant to the discussion, considering the public nature of the content and the likely legitimate interest or public interest as the basis for processing, rather than consent (which is the weakest basis anyway).
I am based in the EU, consent is not optional. Very much -explicitly- the opposite. In any case, and that's the most urgent issue here, removal requests for PII must be honoured. I invite you to have another look at the GDPR as your comment makes me think you're approaching this discussion from a non-EU perspective. The US for ex. generally have a much more liberal perspective on data privacy and consent and I think Americans often fail to understand some commonly accepted US practices are absolutely illegal in other countries. EU laws consider personal data & information privacy a fundamental right.
Time to submit that PR then. That deletes all your data. Too much work?
I'm not against it. If a user wants their data removed, they can send a PR with the changes they want (as long as they can confirm the data belongs to them). Since I'm not legally obligated to do it, I won't spend all day manually removing every offended user - but I am willing to review PRs if they contain valid proof that the posts they're removing belongs to them. Cheers!
Actually, you are obligated to do it, @alpindale by virtue of the Bluesky Developer Guidelines, HF Content Guidelines (and y'know, the GDPR). And no, it doesn't matter that you're doing this for lulz, or to correct an injustice, or anything else. The Bsky Guidelines are pretty clear:
- All services must have a method for deleting content a user has requested to be deleted.
- Developers must have a system for appropriately responding to all user reports of violations within their apps.
- Developers must maintain updated contact information accessible to the public including an email address that is regularly monitored.
- Developers should maintain reasonable security measures to protect against unauthorized access or disclosure of any end-user information or app data.
- Failure to respond appropriately to known violations of Bluesky Social’s Terms of Service, Privacy Policy or Community Guidelines may result in suspension of access to services or features run on Bluesky’s infrastructure. Developers should keep records of all reports of violations and their responses, and Bluesky may at any time request such data to ensure compliance with its policies.
The law is also very clear. You lack a lawful basis for using data, at least of EU data subjects. You also do not have a valid purpose for using this data, nor did consent to post on Bluesky translate to your use of that data -- users didn't consent to you snarfing up their information to make some sort of political statement. The purposes we agreed to are in Bluesky's Privacy Notice. Training an LLM model is notably absent.
And you haven't obtained consent, or provided notice to users whose data you used. Since you're a controller of this data set, you're infringing on the GDPR by not providing transparency, having a lawful basis, adhering to fairness, securing the data at all ... I could go on. You're also violating many, many other laws.
Finally, you're also violating HF's own Content Policy (https://huggingface.co/content-guidelines) which states
"We do not tolerate the following Content on our Platform:
...
- Content published without the explicit consent of the people represented;
- Content that infringes or violates any rights of a third party or an applicable License;
- Content that violates the privacy of a third party;
- Content that violates any applicable law or regulation;"
...
"Some Content is deemed broadly inappropriate for the Platform, and will be removed and may lead to further consequences for the Users depending on severity. Such Content is covered in the 🙅♂️ Restricted Content section and will usually be addressed directly by the Hugging Face Team."
cc: @clem