• zwerg@feddit.org
    link
    fedilink
    English
    arrow-up
    1
    arrow-down
    7
    ·
    12 hours ago

    I get what you are saying, but its not just data egress, its how the model was trained. I’m thinking of models trained to inject security weaknesses, for example.

    • ApeNo1@lemmy.world
      link
      fedilink
      English
      arrow-up
      7
      ·
      8 hours ago

      I see no reason US based AI companies would not carry the same risk especially with the well documented cases of groups like the NSA openly asking US technology companies to provide back doors.

      • zwerg@feddit.org
        link
        fedilink
        English
        arrow-up
        2
        ·
        8 hours ago

        For sure, that’s why I dont trust any of them. Its interesting because unlike open source, open weight models don’t tell you anything about what was included or excluded.