> Why would you do this instead of just generating real data?

The idea would be that harnessing the AI's ability to create massive amounts of regurgitated old data quickly and then transmuting it into 'new data' somehow is faster than acquiring real data.

I mean I believe you, I'm not in this field nor a genius, so if the top AI people are seeing it as a problem then I have to assume it really is, I just don't understand it fully.