What a time to be alive. Our most advanced technology is used to cheat on homework and play video games.
It used to be that more computational power was desirable because it would allow for developers to more fully realize creative visions that weren't previously possible.
Now, it seems that the goal is simply visual fidelity and asset complexity... and the rest of the experience is not only secondary, but compromised in pursuit of the former.
Thinking back on recent games that felt like something new and painstakingly crafted... they're almost all 2D (or look like it), lean on excellent art/music (and even haptics!) direction, have a well-crafted core gameplay loop or set of systems, and have relatively low actual system requirements (which in turn means they are exceptionally smooth without any AI tricks).
Off the top of my head few years: Hades, Balatro, Animal Well, Cruelty Squad[0], Spelunky, Pizza Tower, Papers Please, etc. Most of these could just as easily have been made a decade ago.
That's not to say we haven't had many games that are gorgeous and fun. But while the latter is necessary and sufficient, the former is neither.
It's just icing: it doesn't matter if the cake tastes like crap.
[0] a mission statement if there ever was one for how much fun something can be while not just being ugly but being actively antagonistic to the senses and any notion of good taste.
That's not _quite_ how temporal upscaling work in practice. It's more of a blend between existing pixels, not generating entire pixels from scratch.
The technique has existed since before ML upscalers became common. It's just turned out that ML is really good at determining how much to blend by each frame, compared to hand written and tweaked per-game heuristics.
---
For some history, DLSS 1 _did_ try and generate pixels entirely from scratch each frame. Needless to say, the quality was crap, and that was after a very expensive and time consuming process to train the model for each individual game (and forget about using it as you develop the game; imagine having to retrain the AI model as you implement the graphics).
DLSS 2 moved to having the model predict blend weights fed into an existing TAAU pipeline, which is much more generalizable and has way better quality.