• 0 Posts
  • 239 Comments
Joined 3 months ago
cake
Cake day: March 22nd, 2025

help-circle






  • I’m increasingly convinced that while AI in general had a big leap forward a few years ago, since then there’s some mass hysteria type effect where everyone seems convinced that it’s been getting steadily better at the same rate - when I’ve not been able to see any of that.

    I keep reading all these benchmarks that are supposedly destroyed every three months, but they never seem to turn into any noticeable improvements when I use the damn things. It’s now at the stage where if someone tries to tell me otherwise I just assume they’re either a bot themselves, or falling for some kind of hysteria/confirmation bias type effect.

    So, it wouldn’t surprise me in the least if autopilot literally was getting worse as they try forcing more data into it and getting nothing but overfit. It was already as good as it was going to be based on its fundamentals, which is to say, not good enough.








  • skisnow@lemmy.catomemes@lemmy.worldSome things don't change
    link
    fedilink
    English
    arrow-up
    4
    ·
    edit-2
    6 days ago

    The fact that there’s no buckets means that you can’t then usefully draw any further conclusions about the ratio of buckets to things. In your first two examples we can take the results and use them to work out further things like how much might the buckets weigh, what happens if we add more buckets or more things, etc.

    In the divide by zero answer, we know nothing about the buckets, and the number of things becomes meaningless. But worse of all is that it’s easy to hide this from the unwary, which is why you occasionally see “proofs” online that 1=2, which rely on hiding divide-by-zero operations behind some sneaky algebra.

    When we say we “can’t” divide by zero, we mean ok you can divide by zero, but you’ll get a useless answer that leaves you at a mathematical dead end. Infinity isn’t reversible, or even strictly equal to itself.