673
Basically this (infosec.pub)
submitted 4 months ago by to c/funny@sh.itjust.works
38 comments fedilink hide all child comments
 
you are viewing a single comment's thread
view the rest of the comments
[–] 21 points 4 months ago (8 children)

This is what really gets me with AI and why its 1000% a bubble.

Putting aside all the ethical problems.

Putting aside the environment problems.

Putting aside how basically everyone hates it.

Lets say its popular and will take off etc.

Its all going to become more efficient enough you can run it all locally. Why are we trying to piss away zillions of dollars on data centers and these stupid companies, when 99.9% of uses will work with a model running locally on the GPU/NPU?

  • permalink
  • fedilink
  • source
  • hideshow 8 child comments
    •
  • [–] 6 points 4 months ago (3 children)

    Running is one thing, training is another though. That doesn't come cheap and is not really feasible at home.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 3 child comments
    •
  • [–] 1 point 4 months ago*

    Its all going to become more efficient enough you can run it all locally. Why are we trying to piss away zillions of dollars on data centers

    The idea that the hardware we currently have is good enough for future applications hasn't been true of software in general. As better hardware becomes available, software has improved to take advantage of it. Also, the improvements in software have allowed the development of even better hardware, in a virtuous cycle.

    Also, regarding local vs cloud computing in general, datacenters provide an economy of scale that can't be matched by local compute. Shared datacenters are also more efficient than everyone having expensive hardware that they only use a fraction of the time. This is why datacenters were expanding rapidly even before the recent advances in large language models.

  • permalink
  • fedilink
  • source
  • parent