> I’ve been doing AI for decades

Considering that "AI" before 2016 or so was, in terms of results, a whole different category, this may not be the flex you intend it to be. ;)

I started doing deep learning in 2011 after a visit from Andrew Ng, prior to that I was doing old school neural nets (RBMs), random forests, Bayesian nets, information retrieval (search engines), symbolic AI, and expert systems. It’s really only the transformer class (2017) that kept on scaling (didn’t plateau) but I think it took to 2019 before that was really widely known. I got some really good results out of squeeze and excitation in 2017 and knew then attention models were the future. I guess how long I’ve been doing it depends on your definition of AI, I think the future of AI will probably work alongside solvers and ontological reasoners which I worked on at university in 2003 though the tech goes back much further.