• FishFace@piefed.social
    link
    fedilink
    English
    arrow-up
    1
    ·
    3 days ago

    Assuming:

    • 18 Mbps files
    • 60 minute long films

    That would be about 1,100 pornos. However, the 9 TB is from an unrelated case; it’s not alleged here (at least) that it’s porn. 1,100 samples would not be much for training an image or movie generating model from scratch, but 9 TB would represent a vast amount more data if it were text for books (which the other case was about), and that would probably have represented more like a significant chunk of the amount you’d need to train a language model.