Agree. I came here to say this. 2012, with Andrew Ng's effort to use a neural network on a large farm using CPUs for the compute to find cats in youtube video was the first "modern" AI attempt. As I recall all that compute, and their accuracy was like 60%. Vs YOLO which can run in real time on one GPU and find cats by breed.

That was the start. But 2022 does feel like something special. It might be the "knee" of the S curve - when things start to go exponential. Hard to say, due to the recession throttling the money flow into AI it may slow things down a little.