If you haven't seen it already, you will very soon. Sam Altman, the face of OpenAI, talking about how RAM is not used efficiently if said RAM lives in people's computers, and not inside a datacenter.
Of course, this means that we have been given permission to be upset, and to be transparent, I'm tempted to reply with attacks as well.
The way I see it is this: We seem to have fallen into a trap of maxxing. Everything has to be optimized, utilized until the hooves of the horse reach flesh. Resources, after all, are scarce, and thus we mustn't be brutes who allow the fruit juices to drop to the ground.
Maxxing, maxing, maxxxxing. Looksmaxxing, tearsmaxxing, tokenmaxxing, auramaxxing. There's a trend here, something that unites all these different ideas and behaviors.
To be fair, I can see what Altman is trying to say as something that is not entirely incorrect. You see, I do sleep. Very little these days, I admit, but I do find my pillow eventually. So that means my 16GB of RAM, for at least six or seven hours, are just sitting there, doing nothing. THE HORROR.
Seems to me like Sam, and other CEOs in this field, are simply committed to looking at everything through one lens. After all, it wasn't that long ago that Sam basically stated, again, correctly so, that training an AI is expensive, yes, but training a human for a job takes twenty years. He says this to point out that there is no such thing as job efficiency that comes without a cost, without a climb.
The problem with that statement was that what he called a cost, in the case of the human, of course, has another name: life.
And although owning RAM is not in the same realm, so to speak, it does point to the bigger problem here.
How about freedom?
I could, not that I would, of course, buy RAM for my grilling practices. I could use it to slap caramelized garlic on top of pork chops, and there should be nothing anyone could do to stop me. The mere idea that I should not be allowed to buy RAM, or that I'm not within my rights to do so, is not a discussion I would even entertain.
But hey, at least we can call this transparency of a kind. But also, we can clearly see the necessity of not becoming too complacent.
No matter how much I'm told that local inference will never be as good as closed models, I simply don't think that means much. The liberty for us to explore, to learn, to grow, and to own our data, our infrastructure, our online existence, should be a human right, as far as I'm concerned.
Data dignity is a thing, after all.
How about privacy maxxing? Let me add that one to the list too.
MenO