Login
You're viewing the discuss.systems public feed.
  • Mark NewtonNewtonMark@eigenmagic.net
    Jun 20, 2025, 11:00 AM

    @cstross @davidgerard I keep wondering about what’ll happen when this one pops.

    The other bubbles you mentioned left something tangible behind:

    The original Dotcom bubble resulted in a bunch of local haul fiber optic transmission systems being sold at distressed asset prices, yielding (eg) Level3 Communications as a global connectivity wholesaler.

    Property bubbles leave the property behind, maybe with some people who couldn’t afford homes finding themselves in the right place at the right time to buy one at bankruptcy auction prices.

    When someone loses their shirt, it isn’t usually set on fire; They’ve lost it because they’ve been forced to sell it below-cost, which means the counterparty is buying it below-cost. Cheap shirt!

    The LLM bubble seems different, though. The only thing that makes the products of AI companies look feasible is that VC inflows mean they can be sold cheaper than the replacement cost of GPUs and the electricity cost needed to power the datacentres the LLMs live in. If the VC money retreats none of this will be viable.

    Who’d buy an AI coding assistant or a personal agent or whatever if its API cost $100 per day? Or $1000 per day? If there isn’t a profit stream to keep GPT-4 or Claude running, what happens to applications using the GPT or Claude APIs?

    There’s a pretty good chance that when this bubble pops, there’ll be nothing left. At all. Quite different from other bubbles!

    💬 3🔄 1⭐ 0

Replies

  • 💬 0🔄 0⭐ 0
  • Mark NewtonNewtonMark@eigenmagic.net
    Jun 20, 2025, 11:17 AM

    @benji_w @cstross @davidgerard Sure, enormous building full of power hungry single-purpose GPUs that are too expensive to run. I guess if it sells cheaply enough at the bankruptcy auction you could bulldoze it and redevelop the land for housing, right?

    💬 0🔄 0⭐ 0
  • George Bgbargoud@masto.nyc
    Jun 20, 2025, 1:46 PM

    @NewtonMark @benji_w @cstross @davidgerard

    They're not really single purpose though, or at least not that narrowly single purpose. ML has existed before LLMs and there will still be useful ML models after the LLM hype wave passes and the cheap GPU capacity will probably lead to a lot of development of more focused and proven useful models. Sure it's expensive to run all the GPUs 24/7 but they don't need to, they can just have high burst capacity.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0
  • Charlie Strosscstross@wandering.shop
    Jun 20, 2025, 11:34 AM

    @NewtonMark @davidgerard There's another type of bubble that leaves nothing behind: a Ponzi scheme (and other scams).

    It can be hard to tell a scam from a bubble, until you look at the underlying assets. The AI bubble, however, smells about 80% of scam and 18% of wishful thinking (with 2% "breakthrough in assisted diagnostic imaging" on the side). So it'll crash harder and further than most of the current century's previous bubbles.

    💬 1🔄 0⭐ 0
  • Mark NewtonNewtonMark@eigenmagic.net
    Jun 20, 2025, 11:38 AM

    @cstross @davidgerard It’s so hard to reliably tell the difference between malfeasance and stupidity. The fact that so many of the LLM people used to be crypto people suggests scamminess, but my spidey sense also says Altman isn’t really as clever as he’s made out to be and he might actually be genuinely stupid.

    💬 0🔄 0⭐ 0
  • Charlie Strosscstross@wandering.shop
    Jun 20, 2025, 11:43 AM

    @NewtonMark @davidgerard Criminology tells us that many—probably most—deliberate (non-impulse) criminals think they're smarter than they actually are: they do crimes because they think they've discovered a shortcut to wealth, having overlooked the drawbacks. Stealing cash from the till without noticing the CCTV cameras watching it.

    Altman strikes me as being of that ilk.

    💬 1🔄 0⭐ 0
  • Mark NewtonNewtonMark@eigenmagic.net
    Jun 20, 2025, 11:51 AM

    @cstross @davidgerard Hmm. That tracks. Like a corrupt politician who’s seemingly unaware that that bribe they thought they’d got away with will actually be investigated, even if it takes ten years.

    💬 0🔄 0⭐ 0
  • David Gerarddavidgerard@circumstances.run
    Jun 20, 2025, 4:08 PM

    @NewtonMark @cstross Altman is smart in particular ways. Important to note that as far as I can tell he *hasn't lied to his investors* yet:

    * we have no path to revenue
    * we have no plan to revenue
    * treat this as a donation
    * also we might build AI GOD, but that's a forward-looking statement you understand

    💬 0🔄 0⭐ 0
  • C. Scott Ananian (he/him)cscott@kolektiva.social
    Jun 20, 2025, 12:56 PM

    @NewtonMark @cstross @davidgerard I suspect the 'damage' done to copyright will be a lasting effect. If the bubble can push through laws that weaken copyright to allow training of LLMs on piles of pirated books and academic papers, perhaps actually public goods like the Internet Archive and arxiv and scihub can benefit.

    Of course, it's still possible for us to end up with /worse/ copyright regimes that only let the biggest AI players benefit from the knowledge commons while keeping it locked away from the rest of us, but I can dream at least.

    💬 0🔄 0⭐ 0
  • Gracious :loading: Anthraciteanthracite@dragon.style
    Jun 20, 2025, 2:25 PM

    @NewtonMark @cstross @davidgerard

    if the constant stream of "here's my profitable new startup that I vibe coded in a week" posts is any indicator there will be *some* useful software services whose codebases are giant labyrinths of generated code, and some of them will be making enough money to hire a human to go in and start hacking away; there's probably going to also be a lot of generated code in any huge, ongoing projects like Adobe's stuff - I would be *very* surprised to learn their push towards AI features in the past few years hasn't also involved a bunch of vibe coding.

    💬 0🔄 0⭐ 0
  • 💬 0🔄 0⭐ 0