It’s worth noting that he also fired many of the staff who know how to ensure that they’re actually safe, as well as the staff who would approve financing.

  • @GissaMittJobb@lemmy.ml
    link
    fedilink
    25 days ago

    Why would you need 24/7 delivery for AI? You can vary the load for training as much as you want, while inference can be throttled to meet energy availability. There’s nothing inherent to AI that warrants that type of power profile.

    • @nuko147@lemm.ee
      link
      fedilink
      English
      15 days ago

      You can not stop training that easily, it takes weeks or months of planning. Also idling hardware is bad usage of capital. Companies do not like these. They want to eliminate energy bottlenecks. Its not that AI needs it, its economics.