Currently Online 
| curl-6 said: Upscaling is one of very few legitimately positive uses for AI technology, and pretty much nobody takes issue with such use; what we're seeing here though is just a hideous slop filter that makes everything look ugly and uncanny, like those horrendous deepfake ads on social media. |
Outside of your personal opinion - which I disagree with (at the very least the intensity of it), what is the social harm of the existence of an optional toggle feature like this? You keep suggesting it is being forced on you, but DLSS even today is optional and that likely won't change with DLSS 5.
I also don't really see the difference between inferring pixels (DLSS 1+), frames (DLSS 3+), denoising rays (DLSS 3+) and inferring lighting, contact shadows, and (possibly) PBR materials in DLSS 5. DLSS 5 is a lot closer to upscaling in this sense than it is to an auto-regressive VLM, in that it is works best when kept deterministic, small and targeted to specific inference-cases. A VLM can't achieve this in real-time (well it can if you set the temperature hyper-parameter to 0 and have the requisite hardware, but then you would get an even more boring output.)
Last edited by sc94597 - 14 hours ago
