I think negentropy shines when there is only a small difference in two datasets and you can't simply tell (and request) the difference