Renders at a lower resolution (to make game run better), then uses phycology tricks to postprocess and overlay specific little things to make it seem not as low of resolution.

Nslopia decided that since they were using “Deep Learning” to “Super Sample” and do what I said above, they figured, why not use “Deep Learning” to make the image look “better”, too?

Putting it overly simply (ELi5)…

DLSS uses machine learning to do two things:

1. figure out, in real time, what’s going on on-screen
2. change it in some way

All that DLSS 5 changes is to use real life professional photography in stage 2, instead of trying to be subtle and use existing higher resolution videogame images/look. It’s an interesting experiment just for fun, but some disrespectful, greedy, simpleton MBA with bad taste obviously saw it and thought “oh yes”.