AI/LLMs DLSS 5 will be the worst thing to EVER happen to video games

To me the main point, that you seem to have left unaddressed, DLSS5 was turned off and similar oddities still existed. For example, The odd oblivion eyes, they were there even without DLSS5!
Please refer to the third paragraph of my post. If tech makes something worse it's not absolved by virtue of something being bad in the first place.

I didn't watch the video and honestly haven't really paid much attention, other than to chuckle at the hysteria.

But I'm curious about this from a technical perspective. I think what you are describing is this:
1. First, the original mesh of triangles as written by the programmers is still used to model the scene in 3D, in memory, along with all the texture mapping and ray-tracing and whatever other techniques are used today to determine the color of each pixel.
2. The scene is rotated so that the viewers P.O.V. is oriented along the z-axis (maybe this is done earlier; I don't really know) and everything is "collapsed" into 2D
3. DLSS 5 is now used on the resulting 2D image...I'm guessing using the some of the same "adjacent pixel" logic as generative AI.

Thus no "changes to the geometry" in the sense of altering the triangle mesh, but possibly some re-touching that looks as if it could have resulted from geometry changes.

Is that correct?
It generates new images and presents that as the frame for the user. The input it uses to generate those images is the colour data (i.e the 2D rasterised frame you'd see with DLSS5 off) and motion vector information (the directions things in that frame are moving). That is the sum total of the data it uses to infer anything it generates. It doesn't know anything not in the FoV like where a light may be, nor does it know anything the engine itself knows such as what material an object is made from and how that should interact. When it alters lighting and whatever else that's all inferences on DLSS5's part. It's also not going into the game's files and altering model meshes or textures but it is image generation and the images it generates are different than the original 2D input frame.

For a general user the technical steps that take place to result in the final presented frame are irrelevant. The frames they are shown are what matter because that is the end product and that is what people will be experiencing and interacting with. So when a general user says "this changed XYZ" it truly doesn't matter if X, Y, or even Z, were changed because what they're saying is "XYZ now look different". Which is obviously true. The new frames do look different. Zeroing in on how their initial phrasing of the complaint is actually incorrect from a technical perspective doesn't actually address the complain itself. They're not wrong for not liking how a thing looks nor for perceiving changes between the input and output frames. It's great to educate people on how this stuff works but they're not actually incorrect about it from an experiential standpoint and that's the entire basis of the complaint. Ignoring that at best misses the point, and at worst is deliberately misleading. Given the rest of that guy's video the latter seems more likely.
 

log in or register to remove this ad

Got it. So basically how I summarized it, but with the addition of movement vectors (which makes sense).

Thanks. Interesting stuff.
 

Ignoring that at best misses the point, and at worst is deliberately misleading. Given the rest of that guy's video the latter seems more likely.

...or I could also imagine it's just typical Internet pedantry. "Actually...."

Sort of a corollary to the "never ascribe to malice that which can be explained by mere incompetence" theorem.
 

More or less, yes. But I do feel the additional information is important because of the context of this conversation is around a video dealing with the same topic. As mentioned in my initial post about the video I'd typically be more generous but there were a lot of demonstrable lies that he knew he was making based on the information he was showing and he explained how the tech worked. To me that clears most doubts that he simply is unaware of how faulty his logic is.

I'll also mention that how DLSS5 works isn't a guess on my part or anything of that nature. Nvidia has said it's purely a screen space effect and the only data DLSS5 has is the frame and the motion vector data. If the user can't see it on their screen then the AI can't either.
 

In a moment of Irony:

TL;DR: NVIDIA's DLSS 5 trailer, praised for enhancing game visuals but criticized for character faces, was removed from YouTube due to a copyright claim by Italian channel La7. The automated takedown affected all related videos, including NVIDIA's official one, highlighting issues with YouTube's copyright enforcement system.
 

Recent & Upcoming Releases

Remove ads

Top