• 0 Posts
  • 25 Comments
Joined 2 years ago
cake
Cake day: March 4th, 2025

help-circle





  • It’s difficult to put my thoughts on this into words but you’ve captured a good amount of it.

    There is incredible potential to this, but the costs and risks it puts us up against are truly horrific. I don’t think I could have ever guessed that we were going to undo all of the decarbonisation effort, which took years, to power these fucking datacentres. I see the benefit in tasks it can already do, just a few short years in, and I wonder how fucked our economies are going to be when every office worker is made redundant. No government knows how to deal with any aspect of this, and it’s moving so fast that we’re going to be waist deep before the average person even begins to comprehend the scale of what we are facing.


  • paranoia@feddit.dktoFuck AI@lemmy.worldAI = Fascism
    link
    fedilink
    arrow-up
    11
    arrow-down
    1
    ·
    23 days ago

    The models, both open and closed source, are definitely still improving. I guarantee you aren’t getting better results from a 2 year old model than a current one.

    The datacentres are needed because

    1. the so-called frontier labs have one trick, and it’s bruteforcing through the problem with excess compute.

    2. they are in an arms race where success is measured by how many datacentres you have built.

    I think that they are going to end up very disappointed when it turns out that it isn’t a technology that can be walled off for profit, it’s basically just going to be a commodity like email.














  • AI is getting cheaper every year and the output is becoming better. Look at token prices. The so called frontier models are not that far ahead of the open source models that you or I can run essentially for free, either.

    The reality is that you can sit down and automate significant tasks with AI in an afternoon, today. I work in an industry where a significant amount of time is spent auditing and QAing output. A properly trained skill for Claude is able to do the same QA task to 90% of my (senior) level in 2% of the time. Similarly, it is able to coach the junior staff to produce work that is a much higher level than would be otherwise expected of them, reducing the actual audit length on that side too.

    I am hugely positive for the future of AI. I am hugely negative about what it means for people.