> Basically 2010-2018 was an open field for ML/DL research with old(ish) methods being rapidly applied to low hanging fruit and large datasets with newly cheap

> Basically 2010-2018 was an open field for ML/DL research with old(ish) methods being rapidly applied to low hanging fruit and large datasets with newly cheap compute. Absolutely! And what many folks need to continue to remember is that many scientific disciplines and domains are really just starting to wrestle with the utility and implications of this first generation of deep learning tools and applications. I graduated with my PhD in atmospheric science from an R1 just over 4 years ago; at that time, very few people were looking at how DL provided useful tools for their work. These days, the field is inundated with folks playing with these tools and knocking tons of low-hanging fruit off the tree - it might not be "deep", revolutionary research, but it's fomenting a mini-revolution with respect to R2O and real applications of what had previously been somewhat niche science. There's no reason to think this trend won't continue. New tools let new generations of scientists take new stabs at their discipline, and of course the low-hanging fruit drops first as folks get their bearing, build skills/experience, and - most importantly - prove efficacy so that they can get funding for more ambitious work.