Our AI-generated future is going to be fantastic.

Archive link, so you don’t have to visit Substack: https://archive.is/hJIWk

  • jarfil@beehaw.org
    link
    fedilink
    arrow-up
    11
    ·
    11 months ago

    How much time and power per page?

    Then, it seems like Meta trained Llama with copyright protected books without permission, so the model might stop being free at any moment.

      • jarfil@beehaw.org
        link
        fedilink
        arrow-up
        4
        ·
        edit-2
        11 months ago

        Heh, we’ll see.

        For starters, keep that copy safe, in case Meta has to pull down all the “illegally opensourced” copies of LLaMA. If the “authors” (publisher corporations) have their way, it will become illegal to run the model without paying them royalties, so Meta will only be able to offer it as a paid service. You might still be able pirate it though, if people are willing to share.

        For the future, when your next computer comes with neuromorphic RAM capable of running those models a billion times faster… just hope it doesn’t also come with DRM checks built in to stop illegal models from being loaded (“RAM access error: unauthorized content detected”… doesn’t that sound like every author’s dream? 🙄)