It's hugely annoying when people downplay ideas that required years of human intuition to iteratively divine.
Regardless of your position on whether the current scaling trajectories are sustainable or ethical, we didn't have a good mechanism to generally apply ML to arbitrary, heterogeneously, and un-schematized data.
I don't think "AI" is going anywhere. It's here to stay, but it's also massively overhyped and still somehow misunderstood by those who never got a CS education. There are a lot of charlatans still making absurd claims long after it should have ended.
Do you have a POC of this era of "AI" working out as expected without the part I added after the '+'?
I'm only saying that this is not like the rest of computer science. It's borderline brute force. That's the bottom tier for algorithms and often considered trivial.
Transformers are not a particularly significant academic breakthrough in computer science or engineering, but they are an application of what we used to broadly call "big data". The research had a much smaller impact on the result than the sheer scale at which we started to throw hardware at it and the very permissive situation we found ourselves in regarding the data that feeds it.
As a consequence, AI is more like 50% political, 30% sales pitch, and 20% "tech".