• loathsome dongeater@lemmygrad.ml
    link
    fedilink
    English
    arrow-up
    22
    ·
    4 days ago

    One thing that is not reported at all from tech news outlet is what the spending in China looks like this “AI race for (allegedly) national security”. From a glance China has been using non-LLM AIs for a long time pretty well. With LLMs they have Alibaba, Z ai, Deepsewk, Moonshot etc. which are decently established as vendors and research outlets and people seem to be using LLMs as well. But I never hear China planning or wanting to build 500 gigawatts of data centres to win this supposedly existential race. I wonder what things look like in China.

    • ☆ Yσɠƚԋσʂ ☆@lemmygrad.mlOP
      link
      fedilink
      arrow-up
      8
      ·
      4 days ago

      I did a bit of a dive into how AI is being applied in China here, and highly recommend this interview with Alibaba cloud founder on the way he sees this tech being developed going forward.

      You’re right that LLMs are just part of a mix there, and the primary focus is on actually making useful things. Another huge advantage China has that’s not often discussed is that they have far more training data by virtue of having a much bigger population than the west.

    • CriticalResist8@lemmygrad.ml
      link
      fedilink
      arrow-up
      7
      ·
      4 days ago

      It was revealed by the financial Times that in China at any time electricity production outpaces demand by 2-3x, + they are under trade embargo for nvidia chips so they have to be creative with what they have. For them and global South countries ai is potentially a way to nullify sanctions and provide, it’s existential. The race is already over, the west can’t compete with that.

        • CriticalResist8@lemmygrad.ml
          link
          fedilink
          arrow-up
          7
          ·
          4 days ago

          It’s a vast answer that I’m not 100% finished with myself, but the premise imo is the same way many of these countries jumped through the adoption of the personal computer straight to smartphones. They didn’t have the ‘development’ (infrastructure, budget, industry etc) to support personal computers but once smartphones came around they modernized with those directly, and 4G too without even going through cable internet – 4G is super popular in Africa even in poorer areas and they’re investing in coverage.

          In the same way they see AI as something they can adopt to help with their national challenges, for example healthcare to name just one – which is a very complex problem with brain drain, lack of infrastructure for people to get to the hospital, etc – so if they find a way to provide healthcare with AI somewhere in the process, they could treat more people more easily. Other industries are food, construction, education, electrification, etc.

          From an economic perspective it reduces cost of production when you integrate it into the process and therefore can help countries under sanctions and embargos get more mileage out of what they do have available. H100 gpus are forbidden from being exported to China, they can only get H20 which have 20% of the capabilities, so they are developing their own alternative - probably using AI in the process to develop them faster (maybe not yet in chips directly but I know they’re using AI in other industries already). It’s helping stretch what they can access to get the most out of it. In the meantime, they are stretching these H20s like with alibaba’s new cloud algorithm which was posted on the grad some time ago, that reduced the resources load by 82% and therefore fewer GPUs needed to support their center.

          Iran has recently published guidelines for AI usage in academia, and they now allow it provided you note the model you used, time used, and that you can prove you understand the topic. All of these countries are also very interested in open-source AI since they can develop on it and avoid one-sided proprietary deals. They have a need to “catch up” as fast as possible and see AI as a way to accelerate their development and close the gap with the imperial core.

          And of course Cuba announced not so long ago it would make its own LLM, though I’m not sure where that is at currently.

          We are still in the premises of it all of course, but that’s the trend I’m seeing. It’s difficult to find info from these countries about how they are using or plan to use AI right now, but I did find this news that Malaysia, Rwanda and the UAE have signed a strategic partnership to boost AI adoption in the global south: https://www.bernama.com/en/news.php?id=2451825

    • Darkcommie@lemmygrad.ml
      link
      fedilink
      arrow-up
      3
      ·
      4 days ago

      When they run out of money or they have to decide to cut it or in the very unlikely case that they get a miracle

  • ComradeRandy@lemmygrad.ml
    link
    fedilink
    arrow-up
    7
    ·
    4 days ago

    Considering the environmental impact AI is generating in the West, I wonder if AI is meant to be the final squeeze of the proletariat before they expect full environmental collapse.

    • ☆ Yσɠƚԋσʂ ☆@lemmygrad.mlOP
      link
      fedilink
      arrow-up
      14
      ·
      4 days ago

      Honestly, I can’t see how the AI bubble can last long enough to really make a huge difference in terms of the environment. I think it’s rather a last desperate push for the west to try and magic their way out of being completely deindustrialized.

      • Imreallybored@lemmygrad.ml
        link
        fedilink
        arrow-up
        3
        ·
        3 days ago

        I can forsee the bubble popping and something like Palantir buying up all the data centers for pennies. Could be the point of tripling down.

      • ComradeRandy@lemmygrad.ml
        link
        fedilink
        arrow-up
        6
        ·
        4 days ago

        I agree, I’m just musing on the possibility that there is a growing consensus among the capitalist class that environmental collapse is likely unavoidable at this point.