Datasets:

Languages:
English
Multilinguality:
monolingual
Size Categories:
100K<n<1M
Language Creators:
found
Annotations Creators:
no-annotation
Source Datasets:
original
ArXiv:
License:

Reconsider sharing

#7
by mmitchell - opened

The recent FTC round table (https://www.ftc.gov/media/creative-economy-generative-ai-discussion-october-4-2023) highlighted for me how Books3 distribution in particular could be a net harm, and there was discussion about whether the content was legal to distribute.
I wonder if this is a dataset that should not be further shared.

For books3 alternative:

  • Install IPFS and launch it.
  • pip3 install stc-geck && geck - documents
  • You downloading tons of books and papers
Defunct Datasets org

Thank you, @mmitchell .

Indeed the source data of this dataset is no longer accessible since some months ago. See comment here: https://huggingface.co/datasets/the_pile_books3/discussions/5#64be8b19cf4f379eebc4dba8

We thought about just keeping the dataset metadata card for historical purposes.

We are adding a warning to the dataset card and also will raise a specific error.

Please note, also for the record, that the pirate website hosting the content itself (the-eye.eu) removed the compressed archive that this script relied on because of a DMCA / EUCD takedown request. As far as I can tell, this dataset was not functional from July 13th onwards.

FWIW, the content of this dataset was never legal to redistribute like this; even its original creator admitted this. HuggingFace: please be more careful with the datasets that are hosted on your platform — especially these top-level datasets (maintained directly by HF) and other third-party datasets that have had serious legal issues & reports on them. This is not the only one with severe problems.

Defunct Datasets org

We also raise a DefunctDatasetError.

See: #9

albertvillanova changed discussion status to closed

Sign up or log in to comment