I didn't write about this publicly somehow, but it's quite a theme in my private conversations in the past few weeks.

TL;DR: Big AI labs are openly asking to be regulated. Which I personally am reading in one way and one way only. They need a safe way to deflate the bubble.

In other words, they need plausible deniability that "it's not us, it's those darn external circumstances", and that "it's all for the greater good", and "you'll be better off", and "we're saving humanity in the long run".

While in reality the only set of entities they are saving are the AI labs themselves.

The "beauty" of macro-economics, the more I study it, is that the line between "it's the reality" and "it's what we declare to be the reality" is getting more and more blurred as the stakes get higher.

And if we'd be living in the real world we'd be inevitably headed towards an AI bust. "The big AI short" as some already call it.

Will it be good for the planet? I have no idea. But at least it would be somewhat fair. Companies that can show profit will get more investment. Companies that are only chasing the hype would lose out. We'll correct big time and move on.

But the reality does not work like this. As in, the fake, financial reality does not work like this. If anything, history teaches us that financial and fiscal instruments get more and more sophisticated from one crisis to another, such that what's "materially true" becomes less and less, well, materially important.

So it's a genuine inflection point. AI token prices may go down 10x or they may go up 10x. Companies can go bankrupt or create new tillionaires. We seriously can not predict. It's the "generative adversarial reality" at its finest, and, as a mathematical model it does not fail to deliver.

There's one thing I know for sure: the current model is unsustainable. Slowing down AI development and claiming it's Uncle Sam and Co. that were behing this push to slow it down is the safest way out for most powerful parties.

Hence, I'd say it's far more likely than not to happen.

How exactly it will happen — I honestly can't predict. We may well find ourselves in a dystopian world where running a cheap model hosted overseas would be subject to 10,000% tarrifs. Or we may well find ourselves in a world simliar to the post-Covid world, where we just chose to collectively forget the madness we all we pushed through.

But to quote Dr. Strange, we're in the endgame. That is, we're in the endgame of transformer-based LLM architecture that required these many gigawatt-years of training and inference. "I saw 14,000,605 futures", and everyone is now trying hard for us all to end up in those two of them where the collapse is not that bad.

And, chances are, they will succeed. At least, personally, I'd bet quite a bit that they will succeed.