• 2 Posts
  • 143 Comments
Joined 3 years ago
cake
Cake day: June 14th, 2023

help-circle
  • One huge issue is that LLMs do weird and stupid things differently than how humans do them.

    If you’ve developed an eye for reading human-made changes, you’re not necessarily going to recognize new and surprising failure modes as easily. It’s literally harder than regular code review.

    Humans with modern tooling, for example, rarely hallucinate field/class/method/object names because non-spicy autocomplete keeps them on the rails. LLMs seem much more willing to decide the menu bar is .menuBar and not .topMenu, probably because their training corpus is full of the former.



  • Trying to sell consumers on “scaling solves everything” is going to be a hard sell.

    If we look at general purpose computation, which had decades of actual scaling-solves-everything growth, you had two influences that made the message resonate with customers:

    • Clear existing applications where more power made the experience straightforward better. Your spreadsheet took an hour to recalculate at 8MHz and 20 minutes at 25MHz. A lot of the “bigger model” stuff is plateauing with marginal or spotty gains. If I feed another 5 Internets of data to ChatGPT, will that summarized email be that much better?

    • New applications that could be demoed on specialised low capacity hardware and scaled down to consumers as more power became available. Think of early CGI on hardware costing tens of millions, and now you can run Blender on a $149 laptop. Since most commercial AI plays are hosted services, there’s not much opportunity to tease that way anymore.




  • The difference was that Amazon knew how to make a profit, but was reinvesting into infrastructure plays and bigger fish.

    If they had to, they could have been a modestly profitable bookshop in 2002. AWS and monster logistics might not have developed to put them in the 13-digit club though.

    Does any AI-centric play have that fundamental fallback? The services that seem to be most effective at direct monetization, the coding tools, are typically running at huge losses. If they raised costs to cover, precious few firms will pay basically the salary of a senior dev for an emulation of an enthusiastic junior dev with an affinity for footguns.

    The less enterprise-focused products-- parasocial toys, image and video gen, will likely try to dip into consumer subs and advertising, but can that generate the cash volumes these platforms demand?



  • The “default” mode for a USB keyboard allows submitting 6 keys + modifiers. Some boards define nontraditional input descriptors that allow more, but that mode is not guaranteed to work in places like the BIOS menu or naive KVM switches.

    To avoid phantom keypresses when you hit three keys in a “square” on the matrix, a diode can be placed in series with each switch so current can’t go through an “indirect” route.




  • HakFoo@lemmy.sdf.orgtoMicroblog Memes@lemmy.worldyea
    link
    fedilink
    English
    arrow-up
    5
    ·
    1 month ago

    Wouldn’t the vision of immortality make the problem even more immediate for them?

    “Oh, Holden Bloodfeast XIII, born 2259 might die at 16 from cancer from the envitonmental catastrophe I’m sowing… Meh.” rings different from “my own lust for immortality will be turned into scavenging a wasteland I wrought?”





  • It creates a clear heirarchy of information too. The system owns the title bar, so any operations there are system operations.

    At one point browsers did something similar for security awareness-- real permission prompts, etc. were set a few pixels over into the main UI to establist that they were “real” and not part of the page content.

    Most of the time, we’re not so starved for pixels that we have tp be stealing from the title bar.

    Hell, we lived thtough 640x480 desktops without even the cheat of hamburger menus.





  • It smells more like Facebook than Steam to me. they can print money for now because they have established scale and customer base, but it feels a bit slimy to where it might not be that appealing to new users. Dating services in general have a bad vibe-- bot problems, low quality matches, dark patterns, so authenticity is a big selling point, something AI drives a huge stake into.

    I’d expect that thr gay community, after decades of being a target for abuse, tends to be a bit more sensitive of red flags and looking for truly safe spaces. The Facebook comparison breaks down there, as it has 700 million Aunt Martha users whose most politically sensitive post is in defence of Miracle Whip on salads.