Google has good engineers and a long history of high throughput computing. This, combined with a lack of understanding what ML research is like (versus deployment), led to the original TF1 API. Also, the fact that google has good engineers working in a big bureaucracy probably hid a lot of the design problems as well.
TF2 was a total failure, in that TF1 can do a few things really well when you get the hang of it, but TF2 was just a strictly inferior version of pytorch, further plagued by confusion due to TF1. In alternate history, if Google pivoted in to JAX much earlier and more aggressively, they could still be in the game. I speak as someone who has at some point knew all the intricacies and differences between TF1 and TF2.
The results appear promising, but to validate its value they will want to get it to work on other related mammals. From a commercial and ethical perspective, I suspect that there will be customers who would want to their aging pet's life with 50% probability, even if it carries a 50% chance of immediate death.
He is talking about work that involves a deep understanding of the domain and data, on smaller datasets that require a lot more framing to make sense of.
I understand the author to be saying that "small data AI" activities as a fraction of all AI work in the US is lower, because in the US it is so much easier to make more money doing big-data work instead.
TF2 was a total failure, in that TF1 can do a few things really well when you get the hang of it, but TF2 was just a strictly inferior version of pytorch, further plagued by confusion due to TF1. In alternate history, if Google pivoted in to JAX much earlier and more aggressively, they could still be in the game. I speak as someone who has at some point knew all the intricacies and differences between TF1 and TF2.