GameNGen: Neural model game engine that "runs" Doom

10 minutes ago, LKD said:

it doesn't change anything about the value of the research.

The value of the research? Once again:

The Travolta Will Be Posted Until Morale Improves

24 minutes ago, giwake said:

AI is, just like NFTs and the Blockchain before it, another techbro buzzword, heralded as the future by people who only see short-term profits. They don't care about the way they're making things worse, only the money.

To quote Dan Olson’s “Line Goes Up: The Problem With NFTs”:
 

Quote

I think the thing that normies don’t get about [NFT/blockchain/AI] bros is their dedication, the staggering volume of capital they already control, and how deeply rooted they are in the culture of the people who operate the platforms we all use every day.
And that alone is a good reason for people to pay attention.

They have a lot of money and a lot of clout that they can use to try and make Fetch happen.

 

This is something of a splitting point.

Basically, the future shakes out in one of two broad ways:

• One is that some new technological buzzword comes along and “blockchain” and “web3” lose their sway over investors, the stream of new buyers dries up, and the early investors cash out as best they can (popping the whole bubble).
• The other is that they’re successful, and cryptocurrency is able to crowbar its way into enough corners of our lives that it becomes unavoidable; we’re all forced in some way to maintain a crypto wallet to manage whatever coins and tokens become necessary for participation in society, providing early investors with a captive audience and steady flow of capital.

While his point on NFTs have been spot-on, you can pretty much apply this to any hobbyhorse of any tech or finance bro who wants to gain all the power and clout that comes with Juicero Tech (AKA what if this thing already existed, but just enshittified).

And sadly, I can also confirm that doing a Google search for anything nowadays will often lead to AI-generated slop being pushed.

2 minutes ago, LKD said:

 

but imagine being able to take an mri of a heart with a certain condition and from that predicting it's movement and possible future health problems.

I'm sorry, but this is utter nonsense. You can't just make up impossible scenarios to justify the existence of research that's being called into question on its value. This isn't how the body works and for detecting potential heart problems that could occur soon, we already have technology that lets us do this. Throwing lossy MLMs into a process that already has guaranteed methods of working seems silly and off-putting. Predictive models absolutely have uses in the medical field, but it's often for things that have fundamental rules to them that can be genuinely predicted but are simply too time consuming for a person to do. The fact you have to make up scenarios by your own admission to justify this research makes me wonder why you're even defending it.

55 minutes ago, LKD said:

The focus here is not game dev or art it's more generally just allowing ai to predict into the near future for any given image input, i don't understand how you're so confident that this research will never have a practical use when you don't even understand how model architectures can be adapted to different use cases, for all you and i know this research could be critical to developing general reasoning abilities for ai. I don't think that's likely but there are some obvious and very practical uses for this tech if it's developed further.

I don't think this is a valid approach for that.

The aim IMO shouldn't be to try to predict what comes next from the current frame (Rachael Gunn transforms into a human-mushroom-helicopter hybrid!) but in constructing a model of what it sees (shapes, materials, positions, velocity vectors) and then predict what comes next from simulating it in its model. Compare prediction with actual outcome to refine model, and so on.

Or for art. Instead of generating a random image that looks like a person, starts with an internal model of a person, render it, and then you can do predictive painting over it to emulate a particular art style if needed but you won't end up with too many fingers or a body that got a pair of legs starting at the elbow.

All that predictive generative large model slop is doing is teaching the computers how to fake it, not how to make it. As long as this is the approach taken, you can refine it however much you want, it'll still be a dead end.

GUYS GUYS

Imagine if the sky was pudding!!!

bet you all feel stupid now!

2 minutes ago, kevansevans said:

GUYS GUYS

 

Imagine if the sky was pudding!!!

 

bet you all feel stupid now!

First, the water is piss. Then this. Doom is going to places.

Absolutely beyond baffled that anyone is defending this or saying it's impressive. It still looks like garbage and there's literally zero reason to do this, oh an AI can (BADLY) hallucinate a doom level now?!?111... what will the next one do??

The only part that seemed remotely interesting after reading the github is where they mention initially training an AI to play the game as part of the process (i think?), I'd rather see somebody do that as a whole thing, similar to how Youtube programmers like Code Bullet absolutely demolish some random game with their own code.. although even he has fallen off and caved into using AI for some stuff, so uh lol.

I mean, this shit looks worse than those sora ai videos, but I guess there's not as much data to cannibalize and leech off when compared to real life recordings, idk and idc ¯\_(ツ)_/¯
 

7 minutes ago, knifeworld said:

The only part that seemed remotely interesting after reading the github is where they mention initially training an AI to play the game as part of the process (i think?), I'd rather see somebody do that as a whole thing, similar to how Youtube programmers like Code Bullet absolutely demolish some random game with their own code.. although even he has fallen off and caved into using AI for some stuff, so uh lol.

Good news! There have been competitions for bots playing Doom deathmatch since 2016! (and probably earlier!)

https://vizdoom.cs.put.edu.pl/

This page has a bunch of videos of Doom bots beating the crap out of each other in the "competition" segment.

So... yeah.

Honestly if I was less lazy I would make a WAD that replaces a handful of sprites with ones that got messed up with the photoshop smudge tool to "accurately emulate" this AI version of Doom.

Understanding the ingestion process kind of takes the fun out of it. They basically train it by running through some engine-based demos (like others have mentioned, probably zandro, but could be a combination). It just snapshots just enough to procedurally generate the next frame based on what it saw at time of snapshot. There's logic in there of course, but it's more the player moving through a vivid slideshow of images, popping in what its sampling for scene data, which results in  something akin to a dynamic but playable PowerPoint presentation. The smoke & mirrors of the illusion is more panoramic stitching of images as you turn and move, updating it with animate objects where expected and refreshing based on actions.

It'll become perfected in time, but you can see it "processing" what it expects or anticipates should or should not be from time to time.

There's nothing that replaced AI in "image enlargement for art" and some audio processing like "source separation" when it demixes songs down to individual instruments - yet (DemucsGUI, Ultimate Voice Remover and etc). The quality is not the best possible but it's definitely way better than the previous conventional tools like "center channel extraction", this one literally builds a base of instruments to recognize and cut them off like a knife a frequency by frequency.

I put a lot of additional hand and mind work on this in order to squeeze the best out if it but still. It's just another tool that was born, not a magic spell to do all the work for you.

The criticism is ok but make sure it's only the begininng of the technology, I'm pretty sure folks in the 80's were spitting at "8-bit 22khz digitized audio" as well as we do now but it evolved into something that already makes music as good if not better than analog, just take a listen to AI cover of Nirvana made in the style of 50's: 

As soon as the research is going on it's going to get better and better.

I think that discovery of AI is like a discovery of another useful tool in our history, like a wheel or fire, either are useful but not everywhere. You can cook with fire, make better weapons, invent metallurgy and etc but it's certainly may get really harmful if used other way. Wheels make up for a better transport, mills, printing and etc...

I'd love to see the result of training it on both Doom and Duke 3d footage and seeing it make a cursed hybrid of the games.

Just now, Macil said:

I'd love to see the result of training it on both Doom and Duke 3d footage and seeing it make a cursed hybrid of the games.

We call that "Serious Sam" I do believe.

5 minutes ago, maxmanium said:

  Hide contents

 

At risk of going off topic, people's skepticism in a post-Theranos world is mostly rooted in the way that AI bros hype up every single new advancement as "has to have lots of VC money and effort go into something that looks like absolute dog doodoo". Sometimes some ideas are better than others.

and to be clear - this is being marketed as a potential product, not just research. not with the way the paper is titled "DIFFUSION MODELS ARE REAL-TIME GAME ENGINES" (not may be or could be real-time game engines, ARE real-time game engines) and the model having a very cute and marketable little name? no, these guys are not interested in how this AI actually thinks, they are interested in suckering in people to continue doing the work and throw money at it so it can be the next Sola or whatever the hell else is out there.

5 minutes ago, Major Arlene said:

At risk of going off topic, people's skepticism in a post-Theranos world is mostly rooted in the way that AI bros hype up every single new advancement as "has to have lots of VC money and effort go into something that looks like absolute dog doodoo". Sometimes some ideas are better than others.

Don't get me wrong -- there's definitely a bubble. Part of it is definitely a fad, because companies want to push AI onto everything. But I think the difference in our opinions here is that I can still see a lot of value in the research. Technological advancement is exciting to me. I said in a previous post, but I didn't think generative AI would exist in my lifetime prior to ~2021-2022. It's crazy and exciting.

Edit: you ninja'd me. Yeah, the marketing is very overzealous right now, I won't deny. Still exciting to me tho.

6 minutes ago, Major Arlene said:

At risk of going off topic, people's skepticism in a post-Theranos world is mostly rooted in the way that AI bros hype up every single new advancement as "has to have lots of VC money and effort go into something that looks like absolute dog doodoo". Sometimes some ideas are better than others.

Hey now there's a billion dollar idea. A drop of blood run through AI, whatever it generates is what disease you have.

And while you wait for the results, enjoy the AI-generated DOOM running on the units screen! 

1 minute ago, maxmanium said:

 I didn't think generative AI would exist in my lifetime prior to ~2021-2022. It's crazy and exciting.

It's not particularly exciting to me because it's caused a lot more issues than benefits. All the hype is squarely on creative outputs, ChatGPT being the next encyclopedia and customer service replacement when it can't do any of those things well, meanwhile there's plenty of great application outside of these boxes that get almost no attention, hype, or capital. There's a company in Nelson, NZ that does environmental AI-powered research that instead chose to open a brick-and-mortar storefront to... teach people how to use ChatGPT instead of talking about any of its own research.

2 minutes ago, Major Arlene said:

It's not particularly exciting to me because it's caused a lot more issues than benefits. All the hype is squarely on creative outputs, ChatGPT being the next encyclopedia and customer service replacement when it can't do any of those things well, meanwhile there's plenty of great application outside of these boxes that get almost no attention, hype, or capital. There's a company in Nelson, NZ that does environmental AI-powered research that instead chose to open a brick-and-mortar storefront to... teach people how to use ChatGPT instead of talking about any of its own research.

That's definitely part of the bubble. Jumping onto the tech too quickly when it's not ready yet, or stuff like that. I think things should stabilize in a positive manner once it bursts.

2 minutes ago, HiMyNameIsChair said:

i don't want AI to make my art, because I want to make my own art.

Then do that and don't use AI? I'm sorry but I don't really understand this complaint. It's not stopping you.

24 minutes ago, Darkcrafter07 said:

There's nothing that replaced AI in "image enlargement for art"

You say this as if AI upscales are any good.

So the AI managed to recreate lilith.pk3? 

Just now, maxmanium said:

 

Then do that and don't use AI? I'm sorry but I don't really understand this complaint. It's not stopping you.

I literally already do. i make my own maps lol. wow it's like i never said anything was stopping me.

you've missed the point entirely because I don't want capitalistic ghouls pushing this shit as a way to save money and create our art. Not to mention when it is used IT DOESN'T FUCKING WORK...studios over in Japan like WIT have tried this shit... and every time it just creates more work, because people THEN HAVE TO CORRECT THE AI's OUTPUT BECAUSE IT'S UNUSABLE LOL.  I also really..REALLY  don't want them figuratively digging up the dead and defiling their work and memory to create these shallow, hollow, morbid demonstrations they literally cannot object to. 

this is sick, twisted, I hate it with every fiber of my being. it is a monstrous homunculus of the creative spirit and it represents everything wrong with this AI hype cycle.

3 minutes ago, HiMyNameIsChair said:

Not to mention when it is used IT DOESN'T FUCKING WORK...studios over in Japan like WIT have tried this shit... and every time it just creates more work, because people THEN HAVE TO CORRECT THE AI's OUTPUT BECAUSE IT'S UNUSABLE LOL.

I also don't understand this argument. If it's as bad as you say it is and creates more work, then companies will eventually stop using it to reduce costs. Which would mean you have nothing to worry about.

We're in a tumultuous time right now with respect to this technology, but personally I highly doubt it will completely replace artists or whoever else; likely it will be a tool in the workflow to speed things up. Sounds great to me, personally.

5 minutes ago, HiMyNameIsChair said:

I also really..REALLY  don't want them figuratively digging up the dead and defiling their work and memory to create these shallow, hollow, morbid demonstrations they literally cannot object to.

I'm not sure how to feel about it, but this is something else that will get sorted out with time. The tech is just too new for any existing laws to cover it, I think.

I'm a bit surprised by the response this thread got. For whatever it's worth, I understand skepticism towards large corporations and their use of AI, but I disagree with the arguments in this thread that AI has nothing to contribute to human goals. The use of AI in developing COVID vaccines is one of the reasons they could be developed so quickly, with an obvious and significant benefit to humanity. I can't imagine that's the only contribution AI will make in the field of medicine.

Obviously pseudo-running Doom has far less obvious value, but I also don't think this research project used nearly as much computing power or human programming time as, say, the development of Grand Theft Auto 6 or some other contemporary AAA game.

5 minutes ago, Amaruq Wulfe said:

So the AI managed to recreate lilith.pk3? 

It made a cheap imitation of lilith with none of the intent or artistic merit. I don't know if Graf will be angry at this one though. The Word of The Chicken remains to be seen. But if he were, I would be inclined to agree with him.

1 minute ago, HiMyNameIsChair said:

this is sick, twisted, I hate it with every fiber of my being. it is a monstrous homunculus of the creative spirit and it represents everything wrong with this AI hype cycle. 

Every new "AI" thing makes me think of the Miyazaki quote and I agree with it more and more.

20 minutes ago, maxmanium said:

 

Don't get me wrong -- there's definitely a bubble. Part of it is definitely a fad, because companies want to push AI onto everything. But I think the difference in our opinions here is that I can still see a lot of value in the research. Technological advancement is exciting to me. I said in a previous post, but I didn't think generative AI would exist in my lifetime prior to ~2021-2022. It's crazy and exciting.

 

Edit: you ninja'd me. Yeah, the marketing is very overzealous right now, I won't deny. Still exciting to me tho.

Big difference between the internet at large and AI, the internet is actually useful, even despite how many of its corners may be full of trash.

AI can make all the advancements tech bros want... I'm NEVER trusting it with driving vehicles or anything remotely touching anything medical, no matter how "crazy and exciting it gets" the potential for it to make mistakes that a human would never make, despite our own error, or get misused by living humans for terrible shit is always decidedly above 0%

So far all AI has done is allow people to steal art, make deepfakes, misuse peoples real likeneses and voices, allow lazy "artists" and "coders" to scrape BELOW the bottom of the barrel, and I'm honestly ready for it to go the fuck away so that people actually have to use genuine skill and creativity for stuff again.

14 minutes ago, HiMyNameIsChair said:

...okay but i don't want it to get better. Why the fuck would I wanna listen to an AI recreation of a dead man, covering the song of another dead man.

 

also huh.. wow, funny how both of them aren't with us anymore so they can't object to their work and voices being used in this manner, interesting... almost like that was on purpose.

 

If you genuinely find this morbid demonstration interesting... I don't wanna be near you, hell I don't want to talk to you or know you.

 

i don't want AI to make my art, because I want to make my own art.

With great power comes great responsibility... and quite a few have been using it for pretty irresponsible things. Which is sad, I use AI frequently to eliminate repetitive tasks and automate (rather than waste all day writing a script or playbook when there's many more important things to do in a day).

However I do have some good news for you, of which I unfortunately can't elaborate on. There is a current agenda for crackdown of AI abuse and it's impact on media, industry, employment, etc. it's about to become heavily regulated just as super and quantum compute are. Hardware vendors will be restricted to sales and functions of appliances/products capable of such will be cut from consumer grade market. Basically, those outside academic & scientific or defense roles will be unphased, but the current platforms and utilities that are publically available will drastically change to accommodate.

A fun thought experiment is to imagine the results of a system like this made to support splitscreen multiplayer/co-op. With its struggles with object permanence and geometric consistency, both players will probably be lost into separate dimensions as soon as they lose sight of each other in distinct-looking areas, and then as soon as the players both enter similarly-textured areas the model will try to contrive things to bring them back into view of each other.

1 minute ago, Buckshot said:

There is a current agenda for crackdown of AI abuse and it's impact on media, industry, employment, etc. it's about to become heavily regulated just as super and quantum compute are. Hardware vendors will be restricted to sales and functions of appliances/products capable of such will be cut from consumer grade market. Basically, those outside academic & scientific or defense roles will be unphased, but the current platforms and utilities that are publically available will drastically change to accommodate.

GOOD