FSR3.x and earlier, sure. But AMD catched up a lot with FSR4. While DLSS is still marginally better, for most people the difference isn't really noticeable anymore.

DLSS is definitely the gold standard still.

FSR4 is a LOT better than FSR3, probably even better than all of the old CNN DLSS models used by Switch 2 and older games. But they still haven't caught up fully.

Want to point out that dll injectors make using modern versions of DLSS/FSR possible in many old games that don't have official support. The various injector apps people have written are pretty easy to use too. 

DSLL all day, works great going from 1440p rendering to 4k output. On a side note, Frame Gen can go **** itself. I have tried it a few times on a 4090 and can't stand it. Ghosting and lag issues.

As someone who doesn't yet have a RTX card, I use XeSS on Marvel Rivals because it's the only way I can run the game well and get a decent framerate (on super performance mode). I can't say it looks pretty, in fact it looks absolutely hideous, but hey, I do get frames out of it

DLSS seems to be better, especially for me. With an RTX 5080 I can take advantage of multi frame gen, which is a big plus.
There are some games where I prefer the look of FSR - Final Fantasy XVI for instance, but even then, I still choose DLSS because it's so easy get 240fps at 4k.

Unfortunately for consoles, that point is moot.

On Switch 2 for example there are games using FSR over DLSS-Lite, because the game was originally designed for Xbox/Playstation... Or the developer has noticed that during motion FSR provides a better output as DLSS-Lite falls apart when trying to provide anti-aliasing during motion. (But tends to have reduced fizzle on particles.)
Case in point: Donkey Kong. - Where they use FSR1.

It can't be adjusted by the end user sadly.

Conversely... Many multiplayer games will notice that injecting DLSS into a FSR game or vice-versa will flag the anti-cheat software resulting in a ban.

DLSS for sheer fidelity, but FSR easily wins on flexibility. FSR has improved dramatically between iterations as well, and the gap isn't as big even in the fidelity department today. 

That's true. Although in DK's case it might just be that the motion vectors weren't exposed during its development on SW1 and they didn't feel like doing it as the game moved to SW2. FSR1 is just a spatial upscaler. Oddly, Nintendo's internal teams seem to be very resistant to TAA (or AA at all) actually. Shows in how few of their initial SW2 games use DLSS.

Developers of course, if they wanted to support games with updates, could basically swap out DLSS models (unless there is something unusual about Switch 2's implementation) as new models become available to them over the course of the SW2's lifetime and SDK updates. I'd expect that to happen if SSM's(like Mamba) replace transformers, since their main advantage is efficient and fast inference with minimal quality loss, which SW2 could benefit from.

My first contact with AI Upscaling was Resident Evil 2 RE.

Previously, I had a pretty weak GPU (GT 730 2GB), so even in low settings, the game used to lag a lot. When I turned on FSR, it got much more playable. Today I have a better GPU, and I don't need the upscale, but since I like to use Ray Tracing, I use DLSS in Quality Mode to get more performance.

Both DLSS and FSR helped me a lot through different situations.