I see a lot of anti-AI-rhetoric here, but very little pragmatic ideas on how to change things. The answer here is „don’t touch it“, but that will not change the world, except that it makes you feel good and in a few years you can say „told you so“.
It is not that AI does not work, as often said here, it does work for many applications. LLMs will stay and people will use it.
So, are there any ideas that make an actual difference? Nationalize it? Regulate it? Build your own better one?
30
Replies
·
0
Boosts
@hfalcke For goodness sake: Let's continue the uncompromising anti-AI-rhetoric.
The AI lobby is so big, has such an enormous amount of money to spend. They run the operating system on most phones, the operating system on most laptops and PCs, the browsers that surf the web, the biggest social networks.
It's the pragmatic ideas that will not change our world...
@djoerd Not a very scientific approach, I am afraid, and one that will not lead to actual politics that will change the world for real 🤷🏼♂️
@hfalcke
Right. And yet, the economic forecast currently is uncertain. Will there be high margin models? Or a few low performing low margin, high volume ones? Will it be worth investing in higher performance models in the long term? Nobody knows.
A lot of actual politics depends on the outcome of these questions (add plenty of detail, which I omitted here). Then there are policies we could implement now (carbon taxing, data and copyright sovereignty, etc.) to shape some of this future.
For this, it's important to have an alternative narrative beyond the Fediverse as currently the most prevalent narratives seem to be overly and uncritically optimistic, IMO. Consequently, politicians dismiss and ignore concerns.
@hfalcke
NB when the bubble bursts, most assets will be in the US. We don't know what the current administration will do, but I doubt potential bailouts will be for a better world.
But perhaps the "rest of the west" should prepare for its own bailout conditions. At least such that cloud centres can actually benefit local companies. It might affect only the dwarfs, but could set the stage for a better market.
(Though, I fear our politicians will make up everything as they go, when the time comes. And given the authoritarian trends in Europe, all ideas on digital sovereignty lead a marginal existence, I'm afraid.)
@hfalcke I see a lot of anti-smoking rhetoric here, but very little pragmatic ideas on how to change things. The answer is „don’t touch it“, but that will not change the world, except that it makes you feel good and in a few years you can say „told you so“.
It is not that smoking does not work, as often said here, it does work for many situations. Cigarettes will stay and people will use them.
- - - -
(meanwhile, smoking is banned indoors and many people stop buying cigarettes.)
@jaredwhite I work in the Netherlands and the anti-drug policy is actually not what you demand here …
LLMs will stay and people will use it.
There's a big assumption here, that comes from the previous one:
It is not that AI does not work, as often said here, it does work for many applications.
That's not the question. The first-order question is: do they deliver more value than they cost. And the answers to that are skewed by the fact that the current services are heavily subsidised and customers are paying 10-50% of the minimum viable price. So, if people actually have to pay the real price, how many of those use cases still exist?
The follow-on to that is, given that increasing the price by a factor of 2-10x will significantly reduce the demand, what will the resulting price be? Inference costs won't change much, but the price to users is the inference costs plus the training costs amortised across users (plus profit). Fewer users means that the training costs are amortised across fewer users and so the per-user costs go up. So how many use cases are still economically feasible if the prices go up by a factor of 20x?
But that's just the first order. The second order is: how affordable is it when you factor in the cost of externalities. Current LLMs are trained by doing something that, in any other context, would be a clear-cut case of large-scale copyright infringement. If LLM vendors had to actually pay for licenses for the things that they are ingesting in training, the cost of training would go up by at least a factor of 10x.
So there's a huge assumption in 'LLMs will stay'. The companies that train them do so on the assumption that they'll eventually be able to recoup that (huge) investment. If you don't train new ones, they become stale quickly. To be useful, they need to include recent events. For software, an LLM that doesn't know about new APIs and language features and does know about deprecated / removed ones is useless. For other uses, the shifts may be slower or faster. But training isn't a one-shot cost as a result, you're constantly needing to do retraining. And that's what needs the massive datacentres full of GPUs.
So, if you want to claim LLMs are here to stay, you need to answer a simple question: Who is going to pay for the training? If no one is, then how will LLMs remain useful?
The entire industry is spending enormous amounts of money for tiny returns. How do you expect this to shift to actually making money? Note that the numbers in this page look much better than they are, because all of that $516 B in revenue for NVIDIA is spending from other companies (which must be passed on to customers for anyone else to become profitable).
Of course they're useful. You don't need to retrain LLMs so that they have internalized knowledge for them to be useful. The point of the LLMs is that they're able to go outside their training data and look up API descriptions etc and then perform work.
Even the existing open weight models we have today are enough for people to get good usage out of purely local AI on regular graphics cards for many years to come.
New foundation models can then be trained by companies who sell local datacenter deployment subscriptions. When you don't need to satisfy VCs and a quarterly market you can take much longer (=cheaper) to train them.
@troed
As a human, when we do that "inference" we also train on the way.
Note that LLMs don't do that.
We are seriously projecting humanized view on this Neural Networks, into which lots of Internet's content was condensed/distilled into.
Heck, please point me to place with explicit model of anything in LLMs. You, a human, have it. LLMs just don't have this code.
These are Large Language Condensates. Very good for guessing.
@troed
This whole agentic craze is simple.
Have guessers producing outputs/guesses in some perpendicular way. Intersection produces better guess.
The useful unit isn’t “LLMs”, it’s cost per useful result for a specific task. Example: on Direct Browse Content Synthesis, Gemini 3.5 Flash is both the highest-quality model scored (9.00) and the cheapest good-enough option from 75–100% bars. But in Structured Data & Fact Extraction, Qwen 3.7 Plus averages 23.4x the cheapest while clearing only 1/2 tasks at 95%. Economics get lumpy fast.
@david_chisnall @hfalcke open weight models like glm 5.2 will shine when ai stops being subsidized
@david_chisnall @hfalcke but as its being discovered as the funding evaporates, "no", they are more expensive than an employee.
I hope people who are being re-hired after having been fired due to AI, can demand a higher wage than they had, something closer to what the AI rates would be (we should play this game too!)
OpenAI never promised profitability, and now all the investing firms (who invested your pension in AI FOR YOU) are learning what that means (!? you had ONE JOB )
@david_chisnall Oh yes. And aif we included the enormous costs of energy and drinking water, the environmental costs of the huge datacenters (plus planned #spacedebris) in a time of #climateCrisis! These costs are mainly paid by people living in those regions, not by the users. And by the climate, by biodiversity.
We only have one earth.
I'd counted that in the second point (what happens when you factor in externalities?), but it's worth calling out explicitly: there are a lot of costs that vendors of these things are simply not paying, they're being indirectly subsidised for. Take those subsidies away and the prices go up a lot. Make them pay enough for the water and electricity that they consume to be able to fix the damage that they're causing and you'll see them become unaffordable very quickly.
@david_chisnall Thanks. I hadn’t appreciated the need for retraining enough. I wonder if there is away to do a differential update, or multi layers, or whether it’s fundamentally impossible.
The true prize is major uncertainty, but, I would be surprised if not eventually there will be some form of a market equilibrium. People will pay serious money if they’re just 10% more efficient. In some areas, it may well be much more. (In others not).
People will pay serious money if they’re just 10% more efficient
Being able to justify an expense for being 10% more productive requires you to be able to measure productivity. Most places that try to measure this properly find that LLMs are not actually helping at the things that matter. Using an LLM lets you produce text faster. That's like using a paint roller because it lets you apply paint faster: it won't make your portraits better.
Word processors had a big commercial impact because they allowed companies to completely eliminate the typing pool: now, rather than drafting something longhand (or dictating it to someone who wrote shorthand or, later, with a dictaphone) and sending it to the pool for someone who can type a page on a typewriter without mistakes (at least often enough that they can get mistake-free pages out at a reasonable rate), you just typed and edited it directly. That was worth the cost of a computer on every desk (which had a capital cost of a good 5-10% of annual salary, in the early years).
When the current bubble bursts, it's going to come with a massive hit to credit markets. That will have a knock-on effect across many industries. I would expect SaaS subscriptions that you didn't need a few years ago to be right at the top of the list for spending cuts, especially if they're also getting twice as expensive at the same time.
The news is full of 'Company X rehires employees because they're cheaper than LLMs' stories, precisely because, in most industries, a machine that produces plausible output that is often wrong in subtle ways (an intrinsic property of LLMs, not something that can be fixed without moving to a completely different technology) is not actually a useful thing. A help bot that tells customers to use a form that doesn't exist reduces costs relative to a person answering the phone, but it also makes customers leave.
@hfalcke help build a #Lifehouse because AI is just an acceleration of the grand shit show
