2/2 I think there's been a lot of research on generating synthetic data to boost predictive models (incl. classifiers). The concern is that you'll never be able to generate anything "new" by recycling models (trained on a fixed dataset) that generate synthetic data.

→ View original post on X — @rasbt,