Once there is enough adoption, expect token costs to skyrocket and advertising enshitifaction to accelerate. We are still in the “first hit is free” stage of ai drug addiction.
Once there is enough adoption, expect token costs to skyrocket and advertising enshitifaction to accelerate
Models on local hardware that give adequate performance and the availability of non-US models mean that path is doomed to failure. If they crank up the price then people will look elsewhere. While they might not get frontier-level performance out of the alternatives, there’s plenty of stuff out there that’s good enough for 1/4 to 1/10th of the cost. Or just the cost of power, if you’re willing to pay for some hardware and are operating on a small scale.
The only play the AI companies have got now is twofold :
- Regulatory capture, hence all the press releases of basic and frankly embarrassing sandbox failures and “we need to slow down” talk to try and get that rolling, and
- Banning of those nasty, evil, non-US models in the name of national security.
And then what? The rest of the world will just move on to the next most useful thing on the list. There will be ~330 million US citizens to extract wealth out of, to service trillions in data centre debt. That shit’s not gonna fly.
I run Qwen 3.8 27B, which seems pretty comparable to Sonnet 5 for coding and other tasks. Qwen 3.6 was in the neighborhood of Sonnet 4.6. Qwen 4 27B is supposed to drop in the next couple months and we’ll see how it compares to Sonnet 5.5, which benchmarks at Fable / Mythos level.
If the trend keeps up, not sure how useful those data centers will be when a gaming PC is a few months behind the frontier.
Why do you think they’re doing ethics promises now? I have a feeling that all the third party models are going to fail the ethics tests and now that nvidia owns huggingface it’ll be very easy to ban them in the states.
Download your models now
Yep, I assume that’s the plan, though HuggingFace isn’t the only game in town. Hugging Bay, ModelScope, and I assume a horde of others will fill the void if HF goes downhill.
I’m sure they will keep hardware exorbitantly expensive for as long as possible to push ppl to the subscription model. I would love to run local models, but the cost of hardware versus performance is too high compare to cloud at this point in time. I might break even in 5-10 years, but the hardware may be out of date by then. I hope things will level out in a few years. And I agree the us gov will try to ban non-us models and hardware. They may even bail out the ai companies because they will be too big to fail. Also expect ai and robot companies to get massive military contracts.
The companies that replaced employees with AI are already entering the “find out” stage.
The MBAs think they can just pull an Internet and mobile device revolution out of thin air 💀💀💀
Vibe coding a new dot-com bubble
The Internet was a slow burn starting in the 1960s and gaining momentum through one of humanity’s most productive generations into the 80s and beyond, spurred on by personal computing. Mobile was a (retrospectively) natural continuation.
LLMs have been around for a very long time but they’re just forcing the issue with a product that’s not yet compelling enough to justify its enormous cost. It’s like waiting for the iPhone and the App Store to make smartphones ubiquitous.
Ehh I don’t think mobile was just a continuation of the Internet revolution, it was its own thing that happened to change the internet a lot.
I see it as an extension of the PC combined with the Internet and hardware advances. It certainly changed the Internet and our society. But I’ll leave it to historians to fight over!
Mobile coincides with the idea of calculations and data being handled by 3rd party cloud servers instead of yourself.
That should be the time where the majority of performance shifted from being in peoples personal devices, to being in data centers owned by fewer large companies renting out access in various ways.Not sure where else you wanna move the compute now for the next “innovation”. Even the bullshit about space datacenters isn’t really any different, it’s still compute owned by a company rented out.
AI’s business model is starting to look less like a revolution and more like a $6 trillion IOU. Bain says existing AI services may cover just $1.8 trillion by 2031, leaving a $4.2 trillion hole to justify the data center frenzy. Apparently the killer app is still: »please invent enough revenue before the depreciation schedule catches us!!!«
The killer app is “you can’t build your own cloud because we’re hoarding all the compute”
Also many of the skills for building your private cloud have been forgotten or retired because Amazon and azure.
