I actually like DLSS <= 4.5, when it’s well used. It’s just a scaling technique, like bilinear or 2xSai, but instead of using a regular mathematical formula to calculate the interpolated pixels, it uses a neural network. Of course the final results vary, depending on how much of the image you interpolate, the training data and if you use previous frame data and stuff like that (motion vectors, etc.).

OTOH, DLSS 5, sloptracing or whatever you want to call it, doesn’t seem to be a scaling technique (even if it most likely can do that too). It seems to be a video enhancing technique, with stability features included (anchored to 3d objects) to avoid the common morphing artifacts in early video GenAI (pre-Sora 2).

This really needs an Elder Scrolls Oblivion-style remake. Use the original engine for everything except graphics, and remake only the graphics part (and the contact surface between the visual and original engines).