> And for theoretical fields labs don’t need human data - synthetic one works well if not better.

How could that be true? I am a mathematical rube, but surely data that works "better" than real data should be extremely suspect?