Because they use inordinately more resources than traditional methods. Running the actual complete doom game uses about as much electricity as having an AI generate just a handful of frames.
Yeah, thats fair, as a means of playing doom this is maybe as bad as it gets. But it was a research paper. It's not really being proposed as a game engine, it's just a dumb demo of what you can do by prompting a finetuned diffusion model with user inputs. As it stands the paper is reasonably interesting. Especially the need for noise injections. I feel like people are giving the researchers a lot of hate just because Elon retweeted it, and he's lame.
I think the issue is that really, this is research for research's sake, and has no truly useful destination. In of itself it's incredibly wasteful, able to only predict 60 frames of content from of 40 year old game, but the potential use of predicting frames for a game without using an engine is.... What, exactly? Making games that are far more process intensive for the same results? Simulating real environments based off already existing images, which we can already do with those same images?
We have limited resources and manpower to pour into research, but we're using it to have computers do what people already do for fun, instead of concentrating it on applications that could free up humans to do those same things with much less effort. Each of those frames used more electricity than most people use in their entire lives, but there's no potential in it to actually improve those lives. In a world with limitless clean energy and no major geopolitical issues it could be nice to have these kinds of treats, but we don't live in that world, so it's just a massive waste of resources that could be used to meaningfully improve the lives of billions of people around the world.
Because they use inordinately more resources than traditional methods. Running the actual complete doom game uses about as much electricity as having an AI generate just a handful of frames.
Yeah, thats fair, as a means of playing doom this is maybe as bad as it gets. But it was a research paper. It's not really being proposed as a game engine, it's just a dumb demo of what you can do by prompting a finetuned diffusion model with user inputs. As it stands the paper is reasonably interesting. Especially the need for noise injections. I feel like people are giving the researchers a lot of hate just because Elon retweeted it, and he's lame.
I think the issue is that really, this is research for research's sake, and has no truly useful destination. In of itself it's incredibly wasteful, able to only predict 60 frames of content from of 40 year old game, but the potential use of predicting frames for a game without using an engine is.... What, exactly? Making games that are far more process intensive for the same results? Simulating real environments based off already existing images, which we can already do with those same images?
We have limited resources and manpower to pour into research, but we're using it to have computers do what people already do for fun, instead of concentrating it on applications that could free up humans to do those same things with much less effort. Each of those frames used more electricity than most people use in their entire lives, but there's no potential in it to actually improve those lives. In a world with limitless clean energy and no major geopolitical issues it could be nice to have these kinds of treats, but we don't live in that world, so it's just a massive waste of resources that could be used to meaningfully improve the lives of billions of people around the world.