• 0 Posts
  • 114 Comments
Joined 2 years ago
cake
Cake day: February 15th, 2024

help-circle

  • So you need SpaceX to increase its launch capacity to somewhere between 20 to 80 times its current rate, completely reimagine the economics of space-based radiators and solar panels, accept that all that shiny new power and cooling CapEx is just as disposable as the GPUs, and you have to adjust the expected use cases and SLAs to allow for nothing being repairable and latency making them unsuitable for many AI tasks.

    Yes. “Possible.” I mean, I guess, but we’ve got some powerful Hyperloop energy here. I cannot imagine it ever being more economical than bullying and bribing rural areas with access to a power grid.









  • In small but statistically and financially significant numbers, yes, absolutely.

    The bigger paradigm shift will be moving corporate users to per-token pricing, and that’s the one that will really kill Anthropic and OpenAI (and maybe Oracle) if it doesn’t go how they need, but there are way more actual people spending money on AI than one might think… or hope.


  • wjrii@lemmy.worldtoTechnology@lemmy.world*Permanently Deleted*
    link
    fedilink
    English
    arrow-up
    20
    ·
    2 months ago

    I like listening to Ed Zitron stuff, though admittedly part of it is AI Doomer comfort food, but one of the drums he beats is that the spend is absolutely obscene, and they’re going to have to start dramatically dialing up the prices, and soon, to have any chance at all of converting to profitability.

    From what I’ve seen, even people who like AI won’t pay for anywhere near as much as they’re using now while it’s free or flat rate.




  • Not the main issue of course, but I found this fun:

    Both YouTubers said the exposed database records indicated significantly lower customer interest than previously reported. According to Coffeezilla, internal order identifiers suggested roughly 30,000 total orders associated with around 10,000 unique customers, far below earlier public estimates claiming nearly 600,000 reservations.

    So this grift will only net them ~$3M, not sixty. My heart weeps for them.






  • They say they used a paid actor. Of course, even if that’s true, it’s not particularly hard to find someone with a similar pitch, accent, and timbre, and then finish fixing it to make sure it’s as confidently soothing as the NPR voice you wanted to steal in the first place. I suppose in one sense it’s not utterly different from hiring a soundalike, but now the soundalike is damn near perfect (the clips in the article are VERY similar and feel more like the difference in recording equipment than anything else) and doesn’t need to actually be available to perform for new impressions. Yet another example of “withstand motion for summary judgment, string it out, lobby against future guiderails” as the totality of Silicon Valley’s legal philosophy.