Datasets:

lambdalabs

/

pokemon-blip-captions

like 258

Indeed, more information can be found at https://huggingface.co/datasets/huggingface-legal/takedown-notices/blob/main/2024/2024-02-09-Pokemon-Company-International.md

Hi,

We have received your notice and this repository is scheduled to be taken down March 17th at 8:00 am PST.

If you have any further communication, please email legal@lambdal.com .

Best regards,

Jordan Uggla, SWE at Lambda

This is not legal advice, but note that several ways of making sure this dataset does not infringe on anyone's rights, while still being kept around for ML research and academic reproducibility, were discussed by the community, for instance in this thread: https://twitter.com/Buntworthy/status/1767486492178301121

Thanks for the suggestions. We we have temporarily made the dataset non-public until a resolution is achieved.

More details can be found here: https://huggingface.co/datasets/lambdalabs/pokemon-blip-captions#notice-of-dmca-takedown-action

In the meantime, gated user access is enabled to keep this discussion alive.

Since there were Policy scholars on social media pointing to this case as a novel one, I think it's necessary to emphasize a few points about the procedure HuggingFace took here:

• The expectation under the DMCA is to "responds expeditiously to remove, or disable access to, the material". Based on de-facto industry standards, "expeditiously" does not mean waiting for a response within 7 days from the uploader.
• The expectation under the DMCA is that the platform takes action "to remove, or disable access to, the material" — not pushing the responsibility onto the uploader's decision and schedule. (This action is required by the platform to avoid additional liability.)
• It's unusual for the CTO to jump in directly addressing the uploader, but it's worth noting the CTO's awareness of the facts and circumstances around this dataset — as it is indeed well-known. (The DMCA specifies what steps to take when the platform has such knowledge.)

Thus, the evidence from the exchange above and the repository history suggests this takedown was not handled in a compliant way. It's great that HuggingFace is conducting these procedures in the open, which invites reviews like these, but that does not absolve the company from acting in a compliant manner. (On the contrary, the bar is higher.)

I'm sure you'll agree it does not help anyone if HuggingFace ends up in years-long litigation as the Internet Archive digital librarians suggested as a possibility. As you know, my opinion of the legal teams of HuggingFace is that it desperately needs an upgrade — based on the many questionable decisions being made creating, funding or hosting certain repositories (including one that was known for years by AI Ethics communities to contain CSAM). I would recommend either hiring experienced lawyers, or conducting better reviews of actions like these in order to ensure compliance. Please take the necessary steps to reach the bar you set for yourselves — not least for DMCA compliance.

I have requested to have access to your data. Can you please accept it so I can use this for my project? Thank you so much.