There's no persistent world model; there's nothing sustained, consistent or high fidelity about this. Calling an element a world model doesn't make it so.
It's a blurry copy of existing games. That doesn't mean the tech isn't cool or fun, but it does mean that the claims are wildly overblown currently.
No idea what UGC stands for and this feels like the examples showing an AI-imagined Quake demo. There isn't actually a world behind this, just rendered ghosts of games played before.
For the skeptics, think about how bad the video generation models were last year versus today. Besides optimizations, the biggest difference between those models and todays are more training and clever tweaks. A lot of core concepts are the same or very similar.
So I think that the fact that this is so close to a real game engine (if you squint) and combined with one or two other demos that were a little bit more general purpose, is pretty big news. Because it is concrete progress on the path to being able to prompt games into existence the same way we can generate videos, images or HTML/JS etc.
It's a step beyond generating games or other code with LLMs. One can even imagine (maybe) with the right training and architecture, you could prompt your productivity tools into existence -- rendered on the fly frame by frame.
If you keep going, maybe our whole universe is a simulation in some kind of incredible alien neural network or pseudo-reality decompression system.
I dont mean to insult you but your last paragraph makes me not take you seriously at all. I could feel it in the rest of the post but the idea of an alien simulation just reaks of ayuhasca fueled quack science
> A lot of core concepts are the same or very similar.
One point about core models is that LLMs mess up long contexts because they are trained on a static system rather (next token prediction) than a dynamic system (a reactive world, like a market or a road), so they dont know that their decisions affect the the world they are observing. This means they fundamentally cant model and extrapolate how actions affect current state. They can memorize and interpolate that as a static system, which is what they do for these videos and game systems, but that is incredibly data inefficient. Thats a core problem in them simulating games rather than understanding what a text says and means. The text wont change as they interpret it. These models will need to be re hypithesized to understand that their actions change their environment and that is not a trivial change thats a few years ago, thats a change at the level of the transformers where it might happen next year or 50 years from now. theres no way to tell
What a trip. Play the car one and drive into a building. Its such cool experience having it go from wall to a completely different environment, just because it's fun and quirky.
If you look at it from the perspective of "this is supposed to work like a normal game and it doesn't" it's terrible, if you look at it from the perspective of "I have never seen a game do that and it would be insane/impossible for someone to build that experience normally" then it was a very cool 2 mins of my time.
First is the lack of internal consistency. Games are an artform, even though most studios put out slop right now, and that means they need to say something. This will allow us to interact with environments at a shallow level but will it maintain those environments in a logically consistant level? Will it be able to recreate something like Skyrim or New Vegas, where the whole point is the expansive and context heavy world and how you interact with it? It might even do that because of technological advancements (that i dont think are trivial like the agi-bulls seem to think) but then will it mamange to have the self referencial artistic vision behind it that something like DDLC or Undertale have? That self reference requires actual mechanics that mess with your filesystem. This thing wont do things like that because thats just not part of its world system. And making a whole new world system to facilitate that doesnt seem feasible for some indie games
Second is the classic radioactive steel problem of data. If this becomes used more and more, then itll canibalize its own training data. Its not producing a utility, its producing art/entertainment. People might e joy the things it produces initially, but eventually they'll get bored and demand will plummet, at which point novel nad original games wil be needed. Which this wont be able to produce. This phenomenon happened recently with the rise of battle royales. Brendan Greene was inspired to make a game out of the 90s japanese film, but will this AI? and if it cant, then will it flodding the gaming market with slop make it so real game dev knowledge isnt even prevalent enough that someone could have an idea and just implement it?
Ah yes, the "GTA-style" demo that actually just looks exactly like GTA IV because it's just an unholy amalgamation of a bunch of GTA IV gameplay videos. Truly the next generation of gaming.
God forbid we have a little incremental progress, huh?
I think a good rule of thumb when deciding to criticize someone's project is to pretend it was created by your own children or your best friend. Would you be as harsh and close-minded if it were created by someone you love?
Incremental progress towards what? This is literally going backwards. I can play GTA IV on my Playstation 3 right now. I could almost 2 decades ago.
But now I can instead play a version of GTA that resembles what dreaming about playing GTA would be like, in which I can press a button and, after 10 seconds of latency, watch my "character" awkwardly walk into a building as the world melts around him, all while consuming literally 100 times the computing resources that the original game required to run. And this is apparently revolutionary.
If this was created by someone I knew, I'd tell them to learn Unity or something and make an actual game.
The direction this tech is heading seems pretty exciting: just uploading an image and instantly having a playable generative world to explore, sounds like the OASIS game from Ready Player One? :) even if it’s still rough around the edges, but imagine how much better it'll get over the next few weeks/months?
...but why? What's the point in playing a game that isn't an artistic expression, to communicate with another human? If there's no vision behind it, no expression to enjoy, then does it mean anything at all?
I suspect the people who play games to understand the artistic point of view of the game maker are a minority. What is the artistic point of view of Mario Kart?
People play games because they’re fun. They’re challenging and entertaining and interesting. Highly subjective. Whether these neural-procedural games will be popular hinges more on whether they’re engaging or repetitive imo.
The thought of being to upload a single static image of some scene with a humanoid-like entity in it and then being able to infinitely walk around a procedurally generated world with the same theme is insanely cool.
Still glad I stumbled upon seeing this, regardless of whether it's a marketing tactic. The future is amazing.
There's no persistent world model; there's nothing sustained, consistent or high fidelity about this. Calling an element a world model doesn't make it so.
It's a blurry copy of existing games. That doesn't mean the tech isn't cool or fun, but it does mean that the claims are wildly overblown currently.
No idea what UGC stands for and this feels like the examples showing an AI-imagined Quake demo. There isn't actually a world behind this, just rendered ghosts of games played before.
For the skeptics, think about how bad the video generation models were last year versus today. Besides optimizations, the biggest difference between those models and todays are more training and clever tweaks. A lot of core concepts are the same or very similar.
So I think that the fact that this is so close to a real game engine (if you squint) and combined with one or two other demos that were a little bit more general purpose, is pretty big news. Because it is concrete progress on the path to being able to prompt games into existence the same way we can generate videos, images or HTML/JS etc.
It's a step beyond generating games or other code with LLMs. One can even imagine (maybe) with the right training and architecture, you could prompt your productivity tools into existence -- rendered on the fly frame by frame.
If you keep going, maybe our whole universe is a simulation in some kind of incredible alien neural network or pseudo-reality decompression system.
I dont mean to insult you but your last paragraph makes me not take you seriously at all. I could feel it in the rest of the post but the idea of an alien simulation just reaks of ayuhasca fueled quack science
> A lot of core concepts are the same or very similar.
One point about core models is that LLMs mess up long contexts because they are trained on a static system rather (next token prediction) than a dynamic system (a reactive world, like a market or a road), so they dont know that their decisions affect the the world they are observing. This means they fundamentally cant model and extrapolate how actions affect current state. They can memorize and interpolate that as a static system, which is what they do for these videos and game systems, but that is incredibly data inefficient. Thats a core problem in them simulating games rather than understanding what a text says and means. The text wont change as they interpret it. These models will need to be re hypithesized to understand that their actions change their environment and that is not a trivial change thats a few years ago, thats a change at the level of the transformers where it might happen next year or 50 years from now. theres no way to tell
What a trip. Play the car one and drive into a building. Its such cool experience having it go from wall to a completely different environment, just because it's fun and quirky.
If you look at it from the perspective of "this is supposed to work like a normal game and it doesn't" it's terrible, if you look at it from the perspective of "I have never seen a game do that and it would be insane/impossible for someone to build that experience normally" then it was a very cool 2 mins of my time.
I see two issues with this.
First is the lack of internal consistency. Games are an artform, even though most studios put out slop right now, and that means they need to say something. This will allow us to interact with environments at a shallow level but will it maintain those environments in a logically consistant level? Will it be able to recreate something like Skyrim or New Vegas, where the whole point is the expansive and context heavy world and how you interact with it? It might even do that because of technological advancements (that i dont think are trivial like the agi-bulls seem to think) but then will it mamange to have the self referencial artistic vision behind it that something like DDLC or Undertale have? That self reference requires actual mechanics that mess with your filesystem. This thing wont do things like that because thats just not part of its world system. And making a whole new world system to facilitate that doesnt seem feasible for some indie games
Second is the classic radioactive steel problem of data. If this becomes used more and more, then itll canibalize its own training data. Its not producing a utility, its producing art/entertainment. People might e joy the things it produces initially, but eventually they'll get bored and demand will plummet, at which point novel nad original games wil be needed. Which this wont be able to produce. This phenomenon happened recently with the rise of battle royales. Brendan Greene was inspired to make a game out of the 90s japanese film, but will this AI? and if it cant, then will it flodding the gaming market with slop make it so real game dev knowledge isnt even prevalent enough that someone could have an idea and just implement it?
Buzzword soup.
Ah yes, the "GTA-style" demo that actually just looks exactly like GTA IV because it's just an unholy amalgamation of a bunch of GTA IV gameplay videos. Truly the next generation of gaming.
God forbid we have a little incremental progress, huh?
I think a good rule of thumb when deciding to criticize someone's project is to pretend it was created by your own children or your best friend. Would you be as harsh and close-minded if it were created by someone you love?
Incremental progress towards what? This is literally going backwards. I can play GTA IV on my Playstation 3 right now. I could almost 2 decades ago.
But now I can instead play a version of GTA that resembles what dreaming about playing GTA would be like, in which I can press a button and, after 10 seconds of latency, watch my "character" awkwardly walk into a building as the world melts around him, all while consuming literally 100 times the computing resources that the original game required to run. And this is apparently revolutionary.
If this was created by someone I knew, I'd tell them to learn Unity or something and make an actual game.
An AI slop generator is not even closely related to progress.
Amazing! Imaging a future where everyone can create, play, and share their own games...
Yes. Welcome to 1981.
https://en.m.wikipedia.org/wiki/Sinclair_Research
The direction this tech is heading seems pretty exciting: just uploading an image and instantly having a playable generative world to explore, sounds like the OASIS game from Ready Player One? :) even if it’s still rough around the edges, but imagine how much better it'll get over the next few weeks/months?
...but why? What's the point in playing a game that isn't an artistic expression, to communicate with another human? If there's no vision behind it, no expression to enjoy, then does it mean anything at all?
I suspect the people who play games to understand the artistic point of view of the game maker are a minority. What is the artistic point of view of Mario Kart?
People play games because they’re fun. They’re challenging and entertaining and interesting. Highly subjective. Whether these neural-procedural games will be popular hinges more on whether they’re engaging or repetitive imo.
Most people who go to watch movies dont care about the art part, still art.
I am skeptical of the idea that any automatically generated world is necessarily playable in any fun sense of the word.
Super cool feature that one can upload their own images and interact like a game! Fun examples from X: - https://x.com/chongdashu/status/1940655090127516017 - https://x.com/farfetched_ai/status/1940597724107493462 - https://x.com/the_carlosdp/status/1940657574535483742
Seems like super spammy gorilla marketing to me. Odd that the two positive commenters here have nearly zero karma or activity.
I mean, what if it is?
The thought of being to upload a single static image of some scene with a humanoid-like entity in it and then being able to infinitely walk around a procedurally generated world with the same theme is insanely cool.
Still glad I stumbled upon seeing this, regardless of whether it's a marketing tactic. The future is amazing.