I was looking for "NeRF" in here but reading the paper it seems like they don't use neural networks to represent anything.

They seem to just optimize the positions and shapes of gaussian primitives as well as the reflectance properties.

Certainly a lot more "explainable" than a NeRF.