And the even bigger question is who cares besides Nvidia fans ?
Temporal techniques at 60 FP|s are just as good, go watch land of Nanite UE5 and tell me what improvemenst ML would bring to the upscaling ? If the answer is none, then who cares ?
There are 14 DLSS games so far, why do you think that is ? Who is going to bother with the training and unique drivers per game ?
I am more interested in GPU designs that make the temporal better or more accurately less costly in cycles, which will apply to 99 % of games.