> When you hear a company talk about the promises of synthetic data, you should run far far away. The fundamental problem is that in order for synthetic data to

> When you hear a company talk about the promises of synthetic data, you should run far far away. The fundamental problem is that in order for synthetic data to be useful for model training the generative synthetic model must have already solved the problem at hand. I've heard this group is pretty good, and they seem to be proud enough of their synthetic data to publish papers on it: https://ai.googleblog.com/2020/02/generating-diverse-synthetic-medical.html https://ai.googleblog.com/2020/02/generating-diverse-synthet... Illuminate me, why are they wrong?