You know the popular meme about DOOM being able to run on new, unusual devices (like refrigerators). Well, DOOM managed to run even on... AI! Modern diffusion models allow you to literally “generate” the game in the process of its passing. Through predicting the next frame of the classic shooter DOOM, you get a playable game at 20 fps without any kind of underlying real game engine. This video is from the diffusion model.
Tobias Briones
2mo
Welcome to over-engineering, which teaches you that just because something (kind of) works doesn't mean you have to do it. Further, it will not satisfy the original requirements because of the tradeoffs you didn't ask for. Over-engineering is a severe issue in this industry, more than you think. Of course, the example of the video is a good 𝗮𝗰𝗮𝗱𝗲𝗺𝗶𝗰 exercise. Doom 95 is one of the classics I grew up with, I'm glad to see it here 😃.
Patrick Weber
2mo
You can see the game struggling with object permanence 🤣
Ernst Reidinga
2mo
This is verry cool, and I think we will use this in the future. Imagine a game that changes, and is different to play for every player. New levels, new enemies new adventures could be generated on the fly. I definitely think we will see more of this in the future!
Jan Maarten Buis
2mo
Very cool, although not a very efficient use of computing power
All those inconsistencies remind a dream-like experience. A dream-themed game would be a great application for an even more, "deliberately broken" AI engine.
Is it possible to place here the reference to this model?
Cool but a bit glitchy and seems really weird to generate this in image space instead of generating it in 3d and then rendering it.
Chris H.
2mo
Branch prediction/probability goes brrt
The VCD guy
2moSo: terrible power consumption, no object permanence, no consistent rule. But it also has some defects, I guess…