OK that’s a much more defensible claim. It will be interesting to see whether synthetic data works in practice, and if so, where it is most and least effective. But we can’t rule out its potential a priori.

→ View original post on X — @erikbryn,