• somename [she/her]@hexbear.net
      link
      fedilink
      English
      arrow-up
      15
      ·
      2 days ago

      Yeah, people don’t really get the point of the massive data centers. While they do need a decent chunk of infrastructure to run their models for customers, a large amount of it is just for training, data processing and collection, and other R&D tasks.

      The models themselves, once trained, are much more reasonable in terms of computation.