Post

gen AI is an attempt to industrialize knowledge work. Luddites keep coming up in this discussion, because it's incredibly relevant. A short thread, I guess

John Breen@breen.tech

I’m not convinced “(Generative) AI is here to stay”. It’s massively expensive to train models, and companies are burning billions and billions of VC dollars on it with no clear path to profitability. The idea that we should all get familiar with LLMs because they’re “here to stay” rings hollow

Luddites were skilled professional independent textile workers. Making clothing was very labor intensive, and the labor involved required a lot of training. Obviously, capital hated this. Or proto-capital. This was early in the rise of capitalism, so there's a lot of mercantile power structures here

Anyway, a variety of technological improvements made it possible to produce shitty textiles with big industrial machines and easily replaceable labor. This gives all the power to capital, and obviously capital loves that. So capital went all in on shitty textiles. Note the state of textiles today.

Did people want shitty textiles? No, not really. People did want cheap textiles, but that's not exactly the same thing. Which gets us closer to the point I'm working toward. It didn't matter that it's not really what people wanted. Capitalists forced it on them, anyway.

Capitalists monopolized markets, undermined workers, and sicced state military force on the workers who opposed it. This was directly harmful to everyone: workers, customers, society at large, and considering the waste produced by the modern fashion industry—the environment, too. And yet.

So, industrial textiles sucked. But, they were only producible on terms set by capitalists. And capitalists forced out all other options. Gen AI is industrialized knowledge work. It doesn't matter to capital that the result kind of sucks. What matters is that they can dictate the terms.

And so, to the original point in the post I quoted: bsky.app/profile/acab... Gen AI products are wasteful, unprofitable, and terrible quality. That's not what will decide if it's here to stay. The deciding factor is whether they can force all other knowledge work products out of existence.

John Breen@breen.tech

I’m not convinced “(Generative) AI is here to stay”. It’s massively expensive to train models, and companies are burning billions and billions of VC dollars on it with no clear path to profitability. The idea that we should all get familiar with LLMs because they’re “here to stay” rings hollow

That's so unlikely I'm comfortable simply calling it impossible and moving on. It's going to get worse, not better. But even if a miracle does occur, that will not change the power dynamic that is driving this whole frantic mess.

0 reposts2 likes

What do you think change means? And anyway, you'd have to ask Sam Altman. I'm deferring to his analysis. He thinks AI will need to consume several times more electricity than humanities entire current energy budget.

It’s fascinating that you’re deferring to the analysis of the guy y’all call a huckster and a scammer other times. Sam Altman is nothing but a salesman. I don’t take anything he says at anywhere near face value. I’m perfectly capable of learning information directly.

What I’ve learned is that there are significant scientific efforts underway to dramatically reduce the amount of energy LLMs need, both for training & inference, like cut by 90% type of reduction.