sc94597 said:
DLSS, including DLSS 5, doesn't use VLMs/LLMs. The equivocation of LLMs with AI/DL generally is where the irrationality arises. As far as AI technologies go, DLSS is pretty ethical: 1. It is trained on data provided by developers consensually. Because these models need to inference at real-time they can't be as large as a VLM (Video-Language Model.) So instead the requisite training data is the buffer data available in a game engine, most importantly motion vector data. 2. They are relatively small, small enough that if you had the data you could train them on a few tens of thousand dollar workstation. Meaning the environmental impact is also quite small. Labor costs are probably the bulk of the costs for training DLSS models. 3. Their usage is purely option. DLSS has always been a toggle-feature. The only main ethical issue that arises with them is that they can displace some digital artists by reducing available jobs. But that is also true of other rendering technologies, like ray tracing or even middle-ware like game engines. |
I use DLSS myself from time to time, especially in more demanding titles. My main gripe with the technology is that it removes focus on optimization in favor of software bypasses, this reduces the required competence required to utilize tools and engines for development. UE5, with all its baked-in functionality, led to similar issues; developers have enough "auto-fill" options to not need proper skill and understanding of various, singular tweaks and adjustments. All in all, this leads to poorly optimized games that function poorly on hardware that can't use the latest generation Super-Sampling and/or fram-gen, or even games that run poorly with DLSS enabled, taking ages to fix (if it happens at all) by developers who skill and insights.
