• cinoreus@lemmy.world
    link
    fedilink
    English
    arrow-up
    0
    ·
    4 hours ago

    That would be a dangerous assumption. Ofc you can get away saying they’re distilling US models to ease the panic, but it’s an assumption, and would be catastrophic if turned wrong

    • BlaestEgnen@feddit.dk
      link
      fedilink
      English
      arrow-up
      0
      ·
      4 hours ago

      Deepseek publicly claims to have created their models, by distilling competitors and that’s why they’re able to create them really really cheap.

      There could be other motives behind such a claim, such as making competitors fear making their most powerful tools public.

      Of course they have frontier labs around as well, at the very least at various universities.

    • Feathercrown@lemmy.world
      link
      fedilink
      English
      arrow-up
      0
      ·
      4 hours ago

      True, and even if it is correct there’s no reason they couldn’t start training them without using that shortcut.