Tom Lieberum comments on Race Along Rashomon Ridge

[Tom Lieberum]

If the subset $W_{\text{interpretable}}$ of interpretable models is also “nice” in the differential-geometric sense (say, also a smooth submanifold of $W$), then the intersection $W_{\text{optimal}}\cap W_{\text{interpretable}}$ is also similarly “nice.”

Do you have any intuition for why we should expect $W_{\text{interpretable}}$ to be “nice”? I’m not super familiar with differential geometry but I don’t really see why this should be the case..