This is really interesting - Thanks for sharing! It looks like these services are taking an "interview" approach, so they are building their own dataset. That makes a ton of sense, and really avoids a lot of the concern of cloning people without their consent (it would be hard to talk them into hours of interviews without their knowledge).