you are viewing a single comment's thread
view the rest of the comments
[–] 1 point 7 hours ago (2 children)

Not really, it'll just mean everyone will be using Alibaba, z.ai and Deepseek models.

  • source
  • parent
  • hideshow 2 child comments
  • [–] [S] 5 points 6 hours ago (1 child)

    The laws of economics apply no matter which model it is. Right now, the costs to the end user are highly subsidized.

  • source
  • parent
  • hideshow 1 child comment
  • [–] 2 points 4 hours ago

    I can run models on my local machine that are maybe 90% as good as the latest and greatest for pennies of electricity. And with special methods can essentially run even the latest and greatest as long as I'm willing to wait a while for results or make some other special concession, still for pennies of electricity.

    Next time I upgrade my PC will be pricey, but I use my PC for more than just AI inference and will inevitably upgrade, and at that point LLM's will likely have plateaued in capability, at least for a while. And I'll essentially be able to run it all... for pennies of electricity.

    The law of economics will keep prices AI subscription prices low, because local inference will always be a cheap alternative even as RAM prices remain bloated.

    Any subsidization is for the training. Not the inference. All that will happen is the money for improving models will dry up, but the usage of existing models will remain affordable to a mass of users. LLM slop is here to stay essentially forever.

  • source
  • parent