• 4 Posts
  • 70 Comments
Joined 3 months ago
cake
Cake day: April 6th, 2026

help-circle




  • DevDave@piefed.socialtoFuck AI@lemmy.worldLet AI Burn
    link
    fedilink
    English
    arrow-up
    3
    ·
    10 days ago

    “The too big to fail” part might be claims that private equity is using “leveraged debt” to finance the data centers using the AI claims of crazy projected income.

    The money they’re using is your savings and checking account balances being loaned out. The banks don’t give a fuck if/when this blows up as its not actually their money and their socket puppets in the government won’t actually punish any of them.






  • Reminded me. Human creators of procedurally generated content often “cheat”. Best example is Valheim which most people believe is 100% procedurally generated landscapes/geography. Reality is Valheim has a hidden super or master map with the gaming copying a small circle of it and then uses the procedural generator to decorate that.

    If not for all the modders its unlikely anyone would know.


  • They’re certainly trying their damnest to outspend each other to be the last man standing.

    The open models vs the leviathans (Gemini, Claude, OpenGPT, etc) share a lot of the same architecture with some major differences:

    the first difference is parameter count. Very rough example: we know the difference between right vs right or green vs green. That distinction or concept is basically what a parameter is. These tiny little details may be an entire parameter or just a marginal shift between 0 to 1. Each parameter is anywhere from one bit to two bytes with the size further influencing the nuances of various patterns. In application, a model with 300 billion parameters needs 600+ gigabytes of fragile and difficult to mass memory chips.

    Second difference is the ability to handle a large context window. I understand this enough that it gives me a headache but not enough to explain fairly well enough. Models you can run on consumer hardware is unlikely to a have a large context window.

    Third difference is speed. There is a scifi book series about space faring clones of a guy named Bob. In one of the books, two Bob clones are slowly losing their minds because an alien librarian will only answer one question at a time when it reaches the front of a massive queue of other questions. Every time I have worked with a model, its usually a rapid fire process to zero in one the desired result. That’s not happening for most normal people who don’t have a micro cluster of computers in their home.

    On the tail of that last one, they are seriously trying to hook these things up to nuclear power plants to make them financially viable. At home, having a computer running full tilt can translate to a dollar or more of extra electrical costs. $100 added to the power bill at the end of the month on top of not being able to do anything else with an expensive computer is dead on arrival.


  • DevDave@piefed.socialOPtoFuck AI@lemmy.worldAI slop memes and the plot to make AI legit
    link
    fedilink
    English
    arrow-up
    17
    arrow-down
    1
    ·
    edit-2
    13 days ago

    edit: please don’t downvote people because you don’t like the question, think its too obvious, or whatever. Discussion and debate is a good thing! If we shun people for asking questions, we end up with a bunch of stupid people. Also I don’t think this was a bait or leading question.

    Three issues I see:

    I actually don’t mind some parts of “AI” technology but I start heading into Neo Luddite when you factor in these models used massive amounts of stolen material to make them. Furthermore AI is only going to benefit a tiny minority of people. They stole the public library and now want to charge you money to access it.

    Completely procedural generated content is generally kind of crap. Humans are so good at spotting patterns that our brains create patterns where they don’t exist. As AI generated content floods the market with “slop”, it will (has already in some cases) drown out the real creative work of individuals. The quality is guaranteed to never improve because “AI” isn’t actually intelligent and is only designed to make a pattern of data that is acceptable to humans. Eventually humans detect that pattern. This is one driving force behind our ever evolving cultures and is something “AI” can’t do. Feed slop back to AI and it breaks. Feed existing culture to a human and it evolves to create something new (eg Mozart synthesized into dubstep)

    Two parts to this: The end goal of the current “AI” war by the tech houses is to obliterate the competition. Our reality is recursive. In nature when there is no competition you end up with an extreme collapse of the ecosystem. The winning tech house will lead to a similar dystopian nightmare. The other part is that in the process of getting to that point, the winning tech house will ratchet the price up as high as it can. The coding tech houses have been forced to raise prices but its still not high enough to be profitable. Economic pressure of competition is preventing any of them from raising their prices higher.