That article reinforces the view that you need big datasets to do ML. Generating a synthetic dataset still requires a real dataset to work from, and all the useful datasets are controlled by the big tech companies.