• melfie@lemmy.zip
    link
    fedilink
    English
    arrow-up
    0
    ·
    20 hours ago

    They’re all trained on copyleft data, so they should by all rights be open source. Distilling them into open weight models is the next best thing, I suppose.