Insane Metal
Member
Nuts. This will redefine gaming.
Or not.
Or not.
-
Hey Guest. Check out your NeoGAF Wrapped 2025 results here!
Google: Project Genie | Experimenting with infinite interactive worlds
- Thread starter IbizaPocholo
- Start date
- Trailer Hype Platform
bighugeguns
Gold Member
I think if you start AAA game today, by the time you finish it stuff like Genie will be mature, maybe it's the right call to get out early.
The negativity from publishers is warranted. Nasty as always, but warranted.
Last edited:
SlimySnake
Flashless at the Golden Globes
it is trained on games that were created by other developers. without all that art feeding the training model, it wouldnt exist.
we pay $70 for art because its generated by artists with a vision. we reject samey derivative shovelware from other developers because we dont want copy pasta stuff. thats all you will get with AI.
Kenneth Haight
Gold Member
Those frame rates
MidGenRefresh
*Refreshes biennially
Go back over the previous pages and ask yourself if asking what's the "context window" length here is really a smart move, lol.
bighugeguns
Gold Member
I'm not convinced by that, I think individual or small teams of professional creatives will be greatly empowered by this. The idea of having big lumbering groups doesn't make sense anymore.
IntentionalPun
Ask me about my wife's perfect butthole
We'll see I guess. I'd bet against something like Genie replacing any sort of meaningful amount of the games industry within the next 5-6 years for sure.
edit: Stuff like this makes way more sense to me: https://www.worldlabs.ai/blog/marble-world-model
Last edited:
bighugeguns
Gold Member
5 years seems like an incredibly long time in the AI space.
That looks freaking awesome.
Last edited:
nemiroff
Gold Member
60 seconds of world state and 5 seconds memory for physics.
For comparison, Genie 1 (2024) crashed its garbled temporal stability in under 2 seconds. Genie 2 (late 2024), 10 seconds. Genie 3 from just a few months ago also had 60 seconds memory, but it wasn't nearly as dtailed and consistent.
Last edited:
videogamedude
Member
so why can't a cpu simulation including all your npc's, subroutines, story etc be 'the prompt'
why is it only one or the other
...honest question
why is it only one or the other
...honest question
Last edited:
bighugeguns
Gold Member
Would be sick if you could run parallel processes for each feature simultaneously in the same scene. Like the multimodal concept except it's interactive. Crazy.
The more I think about the potential the more excited I get. Pros are going to have a lot of fun.
Layers!
Last edited:
viveks86
Member
Even if someone hypothetically figures out a way to represent all game states in CPU and RAM as prompts (essentially a 16 GB memory dump as a single prompt), then the prompt changes every frame, which is a completely different paradigm as prompts are not meant to be processed in real time. And what is maintaining all these values in CPU and RAM? A game engine right? So then this is not getting rid of a game engine. It's just rendering on the fly. And rendering differently every single time, as they are hallucinations by nature.
Neural rendering will certainly play a major role in the future. But that's predictable and deterministic. Generative AI, by it's very nature, is not. No 2 people will even be playing the same game.
Last edited:
GermanZepp
Member
Slop this slop that. Like if 80% of all current media and art expressions weren't absolutely trash.
There are plenty of talentless hacks running the show right now.
There are plenty of talentless hacks running the show right now.
IbizaPocholo
NeoGAFs Kent Brockman
MidGenRefresh
*Refreshes biennially
People who think this will get so good that you will be able to generate a fully functional game using a prompt are the same people who think you can get to the moon by climbing progressively taller trees.
Bark
Member
Pfft, as if there will be water for your game in 2030 at this rate.
bighugeguns
Gold Member
You're right, they should put the Genie back into the bottle so we can keep selling Fifa ultimate team packs and create more gambling addicts.
MidGenRefresh
*Refreshes biennially
What are you even on about?
StueyDuck
Member
having explored this more, this isn't really what people are pretending it is; it seems like it works in 3 parts: an AI movement animation, an environment creator and a 360 mapper.
I still call BS on a lot of those being shared on social media, though. Most of what i've seen being generated in realtime is basically google earth but it puts a character on it.
I think there's a little bit of lying going on in some of those tweets or pedo tweets, (what do the peds call their tweets on that app?)
I still call BS on a lot of those being shared on social media, though. Most of what i've seen being generated in realtime is basically google earth but it puts a character on it.
I think there's a little bit of lying going on in some of those tweets or pedo tweets, (what do the peds call their tweets on that app?)
Last edited:
Hudo
Member
Well. It's something, I guess. This is probably a solid basis for future work.
A.Romero
Member
I understand how this model work. I also think that resisting stuff like this is like resisting MP3 back in the late 90's. I don't think traditional gaming will stop existing so people still wanting to pay for a unique artistic vision will be able to continue doing so but I don't think the mass market cares.
To me wearing a VR device and describing the scene I want to play sounds pretty amazing but I would also be willing to pay for gaming as it is today.
"Attention is all you need" (the white paper that kicked off LLMs) was published in 2017, GPT was released to the public in november 2022. That's less than 4 years ago and it has come a long way from generating crappy 2d images to being able to create stuff like this. It doesn't make sense to think that this technology will stay like this forever, specially considering how much money is going into it. I bet that in 10 years it will truly be like magic.
nemiroff
Gold Member
Ditto.
ChorizoPicozo
Member
Hemingwayoffbase
Member
I think AI is helping people understand art on a new and deeper level.
Why is Super Mario 64 charming and incredible when the AI copy of it feels cold, lifeless and creepy?
It's because there is no one behind it. It's hollow in the truest sense. A world created by a person (or even a team of hundreds of people) all had actual minds behind them saying "this is an idea I've had as it will be cool to share it with another person."
AI isn't that and by definition it can never be that. There is no one behind the wheel, so the absolute best you will ever get out of it is something that looks like a human made it, but you know deep down that it's just random nonsense.
Even if something cool accidentally happens in an AI generated world, it means nothing.
Why is Super Mario 64 charming and incredible when the AI copy of it feels cold, lifeless and creepy?
It's because there is no one behind it. It's hollow in the truest sense. A world created by a person (or even a team of hundreds of people) all had actual minds behind them saying "this is an idea I've had as it will be cool to share it with another person."
AI isn't that and by definition it can never be that. There is no one behind the wheel, so the absolute best you will ever get out of it is something that looks like a human made it, but you know deep down that it's just random nonsense.
Even if something cool accidentally happens in an AI generated world, it means nothing.
bighugeguns
Gold Member
I think the gigantic mega corporation stacked with some of the smartest people alive has a very clear vision and will deliver on it faster than we can reasonably react. I think it's very foolish to be overconfident that this is a nothingburger and Genie 3 smashed some maintained illusions of safety.
I also think thanks to years of rabid predatory monetization there will be very little love lost between a lot of consumers and various publishers.
Last edited:
viveks86
Member
Sure. This will end up being its own genre. This could be a great use case for a VR theme park, where every person's experience is unique. Over time, it will last longer and reach good enough quality that people will pay money for it as an experience, or as a place to brainstorm ideas... or meditate.... or go on a trip... or just burn time. They would inevitably become social spaces too where everyone can share the same experience. And then there is always the use case for porn. Or just like regular VR, it may all be a cool fad that can't withstand the test of time and gradually falls off people's radar. We don't know. Could go either way and that prospect can be worth some gambling. But the pushback is when proponents stretch the idea to claim the whole gaming industry will be disrupted and that game engines will all become obsolete. Like... CDPR's stock price is dropping right now because Genie can dream up 60 seconds of interactive video at 720p/24 fps. That's plain ridiculous.
Last edited:
Ass of Can Whooping
Gold Member
You know this is google, right?
bighugeguns
Gold Member
Breh this is deepmind, they're the skunkworks of tech, publishers ain't winning a fight with these dawgs, ever.
A.Romero
Member
I don't know, that's how it works right now but I don't have trouble imagining a situation where after prompting and generating the environment and all of that it could be transfered to some kind of permanent instance where people can use.
Regarding the stock market well, that's how it is. People not in the know will react to what they think are signs of change. The bigger the bet, the bigger the pay off. If they wait to move the money until the change actually happens then they have actually been left behind and lost the chance.
I do think this has the potential to disrupt entertainment in general. I've been experimenting with music generating models and it's kind of mind blowing how well it works.
Ass of Can Whooping
Gold Member
They're AI devs throwing shit at every wall. Somehow I don't think they understand the development pipeline that devs want or need
falconhoof
Member
Lots of people told me that blockchains would take over once the blocks got bigger. Just needed a few years to improve the tech and then everyone would be storing everything on there and we'd solve the world's ills. Billions of dollars were pumped into that industry based on these early claims. The billions of dollars became a validator for others to invest, because no idiot would throw billions of dollars at an idea without expert knowledge that it was all going to work.
Genie 3 has a context window of 60 seconds, it just needs to be bigger and then there's no holding it back! We've gone from six fingered renders to photo real cats on skateboards with lasers. Victory is just around the corner.
Or it's not, it's fundamentally not, based on the current core of LLM. Most fixes to help improve things like context have been more money, more hardware, more compute. billions of dollars.
It's an arms race. Whoever gives up first loses, but who actually wins. Remember the push to get to the moon? For that window of time, money and profit were not in question, it was do anything to land man on the moon. Did it open up a new paradigm for humanity, did we establish bases and create new industries off world? No, once the battle was done, the money dried up and no one went to the moon any more because it just burned so much fucking money that it was unaffordable.
Getting to the moon did push some things forward, we gained some great tech from it. I have no doubt that we will from this AI push too, but the sheet amount of money being burned now is insane. The cost of rendering some slop GTA VI mock up is unsustainable. Little Timmy is not running this on his console unless he wants to bankrupt his parents over a weekend gaming session.
No one cares about affordability or optimization right now, because they're just building bigger rockets to be the first to get to the moon. We could have left the solar system by now using something like Project Orion, maybe it would have been great, but the return probably wouldn't be worth it apart from to say we did it.
We're burning cycles on AI generated slop right now just to say we did it.
archangelmorph
Gold Member
So much smoke and mirrors.
Anyone calling these interactive videos "full video games"...? I have a bridge to sell to you bruv...
At best you create a collidable environment mesh you can wander around in.
There's no interactivity depth here. There's no actual gameplay systems.
It's fascinating tech, but it's literally a million miles away from "one sentence prompt" -> "GTA V: THE WHOLE GAME!!!!"
Anyone calling these interactive videos "full video games"...? I have a bridge to sell to you bruv...
At best you create a collidable environment mesh you can wander around in.
There's no interactivity depth here. There's no actual gameplay systems.
It's fascinating tech, but it's literally a million miles away from "one sentence prompt" -> "GTA V: THE WHOLE GAME!!!!"
bighugeguns
Gold Member
I think what's happening is people are using basic prediction of about 3 years ahead and making money moves accordingly. Don't want to be holding the bag if Google deepmind decides to surprise everyone again, which they will.
It's basically video game corp risk vs Deepmind risk.
Last edited:
falconhoof
Member
Not even that. Its "collision" is a probabilistic event, everything it shows and does is just based on probabilistic outcomes. No mesh, no collision, no booleans, no integers to increase or decrease. Just a best guess at what happens next based on a context window of prior inputs. There's no content to probe, no characters, no buildings, no floor, no ammo, no pickups, each image is a probabilistic rendering based on a guidance prompt and prior input state.
It's such a fundamental shift in the authoring of content that's basically pointless to use the same terminology. Video games are collision, meshes, sprites, textures, physics, databases. This is none of that. There are no files on a disk you can open. You cannot ask an artists to tune up level 3 because there's no level, no art, nothing to tune up in that way.
Last edited:
ThatOneGrunt
Member
Don't need this for gaming but would love it for VR.
DownLikeBCPowder
Member
falconhoof
Member
Are we equating human lived experience to AI data feeds now?
I get the point from one angle but you got to be careful where you draw the line.
falconhoof
Member
frame by frame AI. It's basically an interactive video. There's nothing behind it like a game engine or polygonal models or game logic.
Guerilla Munch
Member
I remember pretty recently they unveiled clones of Doom and Minecraft being generated on the fly with all kinds of terrible bugs and dogshit performance, now you can enter any prompt you want and explore an interactive world. It's astonishing how many people downplay what a catastrophic paradigm shift this is.
kevboard
Member
Lol, this experiment already looks as good graphically as GTA 6 and it took some guy 5 minutes to generate it.
But sure, it's "AI slop"...
the only way to not see it as AI slop is if you have both, no clue how video games work, and no clue how these generative AI models work.
because if you knew both of these things, you'd also know why this is not even remotely useable for anything other than a tech demo that people play around with for an hour and then discard again.
but I guess I shouldn't expect someone on a gaming forum to know how video games function... that's my bad.
Last edited:
drotahorror
Member
I'm definitely gonna try this out if normal humans can use it. Pretty impressive.
Moridianae
Member
I think it's unique. I think it's the *speed* at which a lot of this stuff will advance that will cause the most issues. I'm not just talking about Genie--AI is becoming a pervasive element everywhere.
It has positives and negatives, whether those negatives end up being catastrophic remains to be seen.
It has positives and negatives, whether those negatives end up being catastrophic remains to be seen.
TheThreadsThatBindUs
Member
I think folks are missing the point.
The utility of this isn't for making games for the end user to play. The utility would be for devs to use it to rapidly prototype gameplay ideas.
Imagine at the pre-production stage of game development, you want to quickly prototype and idea to see if it might be fun. With this is instant and trivial. That matters a lot.
The utility of this isn't for making games for the end user to play. The utility would be for devs to use it to rapidly prototype gameplay ideas.
Imagine at the pre-production stage of game development, you want to quickly prototype and idea to see if it might be fun. With this is instant and trivial. That matters a lot.
bighugeguns
Gold Member
The end product is the most important thing, how you get to it doesn't matter to consumers or investors. That's the harsh truth.
That said, selling CDPR stock is fucking idiotic because if anyone could utilize tech like this positively it's those guys. Ignore the virtue signaling about never using AI, investors will physically force you at knife point to use the tech if it's deemed it will speed up production and save money.
I think some publishers are jumping the gun attacking Google etc and picking PR fights with groups right now way beyond the scope of their ability, I don't think that's very smart.
Last edited:
ZehDon
Gold Member
We've seen this already with Microsoft's Quake 2 demo ages ago, and I'm not seeing anything here that fundamentally addresses the core issues from that example. Frame persistence is still a non-starter here, so it's functionally useless still. Prototypes are often created exclusively to test "feel", where feel is defined as everything from control responsiveness to long-term gameplay loop behaviours - you can't test any of that with this type of system yet.
Using code generators to generate local compute applications to serve as prototypes works today, and yields better results faster. This type of fully generated approach might be great at some point in the future, but generating frames at 30FPS with full scene and frame persistence, zero hallucinations, and full player responsiveness is a zero sum game: either you're there, or you're not. And the tech is years away.
Using code generators to generate local compute applications to serve as prototypes works today, and yields better results faster. This type of fully generated approach might be great at some point in the future, but generating frames at 30FPS with full scene and frame persistence, zero hallucinations, and full player responsiveness is a zero sum game: either you're there, or you're not. And the tech is years away.
Last edited:
Similar threads
- Poll