I just wanted to share this because I’m finding it fascinating. Google Deepmind just announced Genie 3, a system that will generate a world that you can walk around in, based on a single image. At least that seems to be the promise. Doesn’t have to be our world, either. Imagine taking a screenshot from a game and feeding it into a system that you can then walk around in and change. I mean I know we can already walk around in game worlds.. but maybe from a point and click adventure or something!
This is all second hand info as Genie 3 isn’t available to the public yet. I think the real purpose of it is to build ‘digital twins’ to train AI models for robots and such, but you know we nerds will bend it to our will once it becomes available.
The embargo for this lifted earlier today so you’ll probably see a lot of discussion of it from people much smarter than me; I’ll just share the video:
Groovy, right? And here’s a the official link to learn more: https://deepmind.google/discover/blog/genie-3-a-new-frontier-for-world-models/
[Image generated by Google Imagen 3 using the obvious prompt: Make me an image for the top of a post about Deepmind’s new Genie 3 system]
Space. SPAAAAACE. /Spaaaa-aaaaace/
I have a use-case in mind. You might’ve guessed it through my subtle hints.
Imagining a No Mans’ Sky 2 using this sort of tech though, and just… Yuss.
Now if someone could just hurry up and crack the whole BCI (Brain-Computer Interface) stuff, we’d be set.
I figured by the time I got to the age I am now, I’d have a data port in the back of my head. I’m more bummed about that than I am of the lack of flying cars!
There’s an AI out there that can PLAY No Man’s Sky and I really, really would love a system where you can create an AI ‘friend’ to play MP games with. There’s an AI that can play the game. There’s an AI that can chat with the human. I feel like we’re so nearly there!
It’s interesting that they’ve cracked the persistency problem, which has been the huge issue for all the other AI World Simulators I’ve looked at but they’re still only claiming persistency measured in minutes. Until persistency… er… persists, the kind of uses I imagine we’re thinking of can’t really happen. Also, the graphical quality of the examples they’ve picked out is… not great. It’s going to be a good while before we see something in this field that looks like it has more than novelty value, I think.
Yeah but “a good while” in the AI space might mean a year, they way these things are progressing. Though I suppose a lot depends on how well they can monetize it.
But I have to admit this is the first AI World Simulator I’ve stumbled upon so my excitement level is probably spiking due to the novelty (for me) of it.