Watching these comparisons I get much less interested in the technology to be honest.
Its like the more these technologies get refined (ahem)... the more they start to deviate from the real color information. Look at the differences in the first shot: Nvidia DLSS has a dark foreground and bright cloud/background of the scene, AMD FSR is the polar opposite. DLSS has 'thinner lines' where heavy aliasing would occur (power line is a great example) while FSR has much better definition of shadows in the background and a fatter line on heavy aliasing, something that can also be spotted on the more distant building facades.
The true picture is probably somewhere in between both implementations, color wise likely leaning more towards the DLSS version, albeit more stable and less blurry in places.
I'm struggling to see improvement over native, rather, its just 'different', but absolutely warped from the intention the developer had, making it... IMHO... a worthless technology. Its cheating, and apparently not true reproduction. How far will the trickery go to achieve a few more frames? The implementation here is absolutely not good for green or red.