Everything you said doesn't dispute the above comment and agrees with its core premise:
"In terms of model improvement, yes synthetic data can help. In terms of the arms race? No. True examples provide knowledge that is unique. "
"In terms of model improvement, yes synthetic data can help. In terms of the arms race? No. True examples provide knowledge that is unique. "
I do not agree on that because as I explained, with domain knowledge it is very much possible to shape the data generated for augmented learning - beyond the plain statistical variations of GAN and similar, which are obviously of very limited value in training.