Really surprised by some of the comments here: a mix of hatred, ignorance and lack of foresight
This is a great way to:
- Have a fail-safe / fallback in case the software does not work as intended
- Allow them to continue training the already immense dataset using real data (as opposed to synthetic data)