isn't it obvious ? synthetic data was nice to generate an underlying distribution and avoid jumps in 'data holes', but with methods like diffusion now it's not

isn't it obvious ? synthetic data was nice to generate an underlying distribution and avoid jumps in 'data holes', but with methods like diffusion now it's not as necessary