His point is the highest and best use of the chips is to be full of model weights predicting the next word 24/7 instead of maybe being in a laptop that's off or underutilized sometimes.
It just so happens that the highest and best use is also what makes him the most money.
Nevermind the fact that there was no RAM shortage before he and his buddies decided to hyperscale their word guessers.
It'd be a bit like if an almond farmer came to a town with plenty of water, used up all the water growing almonds, and then tried to argue that growing food for everyone is a higher and better use than townfolk taking baths or watering their vegetable gardens.
It's not just ideologically bankrupt utilitarianism, it's ideologically bankrupt self serving utilitarianism.
Something being self serving doesn't make it false, and you don't have to rabidly deny basic economic theory just because you hate the guy.
(I say some of this tongue in cheek).
Does he actually call for expropriation without compensation? That part just feels like people massively extrapolating from what he said.
The small snippet, no, you're correct, he doesn't call for expropriation, though on its own it does seem an odd thing to say.
The full podcast, no, you're correct, the interpretation folks make isn't reflected in the podcast itself.
In the full scope of what the public believes it knows about Sam Altman, from allegations of familial abuse, to the board fiasco, to the expropriation of art and other works by humans to train the frontier models with reckless abandon, one sees that the absolute cynicism by the public writ large regarding anything Mr. Altman does or says is probably appropriate.
Perhaps unfortunately for Mr. Altman, as is the case for most public figures, his statements aren't seen as encapsulated logic statements to be taken individually, but in the scope of his broader public persona.
Let me know when you're ready to move onto advanced economic theory, including what happens to efficiency and value when a tiny number of entities concentrate an in-demand commodity in their hands and refuse to share.
But you're denying both basic economic theory and basic political theory. If RAM is to be allocated, it has to be done in a safe manner too. Something like publicly funded and controlled datacenters, protected against data leaks and spying by private interests who want to use people's research to then steal their credit, ideas and businesses.
Let's now move from basic politics to economics - upwards of a Trillion dollars of debt has been sank into AI, most of it for free, subsidized AI services provided both domestically and internationally - how's that for "basic economics"?
But there's more, for $1 trillion, you could build 100 chip factories for RAM or other chips - 100 of them - 10 times more than we now have. Not building any or not building enough, means you're severely overpaying for RAM and compute - how's that for "basic economics"?
If Sam wants to plan the economy for efficiency, let's do it the way it has to be done, simplistic self-serving BS doesn't cut it.
The man caused a worldwide RAM shortage. Who gives a flip about context?
I’ll take lower utilization for on prem thanks. Should I give my solar panels to my power company so they can overcharge me too in the name of “efficiency”?
That doesn't make any sense because solar panels hooked up to the grid already benefits from high utilization. However if there is some national energy crisis, and there's a bunch of solar panels sitting idle, it might very well make sense to at least connect them to the grid so the energy isn't wasted.
If you are the cause of artificial scarcity due to hoarding, you don’t also get to claim to be the solution to the problem. That’s just running a racket.
Or should, for example, gamers all have to stream video games from Geforce Now instead of having their own PC?
You don't have to agree with his values (e.g. you could believe that individual ownership of computation is more important than overall efficiency) but the statement here is logical and more or less obvious.
> A lot of people want local inference. And somehow, they assume that the problem is solved now. Because, we are going to have open-source models. We’re going to be able to run them on our computers. And so we’re not going to need all the stuff in the data center. But from a global allocation perspective, actually, it’s quite terrible to have everybody have a bunch of the world’s limited supply of DRAM sitting in their own computer. And, being used 5% of the time. You want maximum utilization in the cloud if you’re trying to deal with this very supply constrained environment. And somehow everyone’s... Many people’s intuition is the opposite of that.
"Scarcity is because of AI demand. Data centers are set up to meet that demand. It’s quite rich to say “evil companies are taking up all the ram” when they are serving billions of users, and you want to do the exact same thing for yourself."
So we’ve got rent seeking behavior squeezing everyday people on both sides, and if anyone says, “hey, I don’t like how this is going”, your answer is to say, “So then set up a factory?”
That’s so tone deaf it makes “let them eat cake” sound downright empathetic by comparison.
Is there any source for this beyond online clickbait? I find it hard to believe that RAM is piling up in warehouses while simultaneously every AI company and every consumer is desperate for it.
So let’s not be easy marks. Let’s be a bit skeptical and look for some corroborating evidence. One obvious place is to check whether the AI market actually behaves like what we expect to see when demand outstrips supply. First and foremost, we would expect to see rapidly rising prices and very high profits for suppliers.
But we don’t see that. Anthropic and OpenAI are very much competing on price. Which, funny, is the opposite of what you would expect of companies that are supposedly struggling to keep up with demand. And apparently razor thin if not negative profit margins, too. Why is that? If demand is so high, then why are they keeping their prices so low it hurts? They aren’t running a charity here.
All the RAM belongs to me is not too far off…
The issue is that it can be applied to everything else, yet it isn't.
We can maximize utilization of wealth if we distribute it.
We can maximize utilization of land if we build more communal pools/third places and sleep in pods instead of individuals each building their own.
We can maximize utilization of transport if we all stopped having our own cars or private planes and all used public transport.
We can maximize utilization of everything if it was available to everyone at all times.
Humanity just hasn't been interested in maximizing utilization, and I certainly doubt that Altman is. That is the crux.
TL;DR: It's only a fair take if it's a principled take.
Edit: Even constraining it to their specific RAM usage, it's also highly inefficient for each person to individually pay them for their own generations over deduplicating generations into open data stores, where I can just look at previous output instead of regenerating my own.
Even things like decommissioning an old model can be extremely disruptive if all your workloads are tweaked for it. Like when Opus 4.8 dissappear from copilot in one company I work for. Suddenly our PR reviews started missing issues they always caught before.
How is this a relevant factor as long as the demand is genuine and not some attempt the corner the market? Otherwise blame all the people/companies buying the tokens.
And they're already cornering the market! If memory prices hadn't quadrupled, I bet you'd see lots more commercial activity working on local models. OpenAI is actively preventing this entire market segment from growing, AND they're hurting tons of tech-dependent markets at the same time. It's no coincidence that this allows them to ask for higher prices.
How is that not taking things out of context (or rather, leaving them stripped of it)? Scarcity caused by what? For the benefit of who? Utilization for what purpose? "Serving people", heh.
I wouldn't want RAM or a GPU for a decent price to run local models, but to do normal computer stuff. That is what is scarce, RAM and GPU for people. Nobody would give a flying fuck if ONLY data centers had trouble acquiring RAM.
If someone fuckers buys up all the drinking water to make enemas 24/7, and people survive on the rain water they catch in bowls, and then those guys says actually, we should have those too, they're just greedy sociopaths, no need to complicate it.
He's not using even one percent of it. I'd love to help maximise the utilisation of his resources.
Why does the rich cunt think his own observations, about scarcity and allocation, don't apply to him?
This is not logical because RAM a resource we want to use. There are many uses of RAM that don’t involve LLMs, and to assume that “max utilization” and “max utility” are the same thing is naive at best. The entire reason there’s scarcity is a problem of OpenAI’s own doing (along with the other labs).
Does that mean... anything?