By using this site, you agree to our Privacy Policy and our Terms of Use. Close

Forums - Gaming - Yes, I support (some) AI use in gaming

sc94597 10 hours ago

By the way on the topic of LEWM, casual models, and this thread. 

Note this is a model that can be trained on a single GPU and has only 15M params. 

https://arxiv.org/abs/2603.19312

Joint Embedding Predictive Architectures (JEPAs) offer a compelling framework for learning world models in compact latent spaces, yet existing methods remain fragile, relying on complex multi-term losses, exponential moving averages, pre-trained encoders, or auxiliary supervision to avoid representation collapse. In this work, we introduce LeWorldModel (LeWM), the first JEPA that trains stably end-to-end from raw pixels using only two loss terms: a next-embedding prediction loss and a regularizer enforcing Gaussian-distributed latent embeddings. This reduces tunable loss hyperparameters from six to one compared to the only existing end-to-end alternative. With ~15M parameters trainable on a single GPU in a few hours, LeWM plans up to 48x faster than foundation-model-based world models while remaining competitive across diverse 2D and 3D control tasks. Beyond control, we show that LeWM's latent space encodes meaningful physical structure through probing of physical quantities. Surprise evaluation confirms that the model reliably detects physically implausible events.

Of course this isn't a generative model nor a traditional VLM. Just wanted to point out that casual DL models are a thing that actual people are researching and implementing. 



Around the Network
Norion 8 hours ago
CaptainExplosion said:
Norion said:

That doesn't change the fact that data center water usage is tiny. Do you any source for noise being a relatively significant issue for data centers? Cause the US has 4-5k of them currently but that's nothing compared to the hundreds of thousands of factories.

As usual you point the finger at industry that was there before AI to distract from AI's glaring flaws.

Here's your proof. Now stop being a shill for technology that's destroying the world one way or another.
Leynos said:

There is an entire city in Nevada that will be without power in a year because the energy provider chose the Data Center over a city of people. You need to remember that power is also for hospitals and people at home with medical devices. Not to mention AC for Nevada heat. It only takes one for it to be ok to happen to other towns and cities. (mutliple local news videos on this) It's already taken water away from just houses a few years ago. They go where the promise of profit goes, not your needs. So go ahead be a fucking dipshit and make your Nintendo AI images or AI memes. It's only fucking the planet and people, including yourself. Go ahead be a frog in boiling water.

^THIS!! Everybody who advocates for AI and data centers has such a thick skull it's a wonder they have any cognitive functionality. The only good data center is one that's been demolished and never to be rebuilt.

I'm not distracting from that cause I've pointed out major concerns various times, I just care about being factual. Can you point out how that article shows that noise pollution is a relatively big issue for data centers compared to other major industrial facilities? The amount of electricity they need is a concern that will need to be rectified by ramping up electricity production but it's funny that the article tries to make the water usage sound scary by saying a large one can use up to 5 million gallons a day when the US as a whole uses hundreds of billions of gallons of water everyday. The water usage is insignificant and tiny compared to how much something like golf courses use.

Also the bolded part really shows just how impossible it is to take you seriously on this topic. The bloody internet wouldn't function without data centers. They're not a new thing that only started being built the past few years, they're extremely important facilities crucial for modern society to function so for the love of god please educate yourself on this stuff cause the ignorance is off the charts.

Last edited by Norion - 8 hours ago

OlfinBedwere 7 hours ago

Even if we accept for the sake of argument that the whole data center argument is a "silver bullet" argument against LLMs in the here and now - leaving aside the fact that cloud computing has been a widespread thing since the early 2010s, and even if Google and the like mothballed all their AI models tomorrow they'd likely just convert the data centers into ones handling e-mail and cloud storage - that's probably not going to be the case forever.

15 years ago, IBM wrote an AI model that was able to compete in a special episode of Jeopardy and whooped the asses of two of the show's greatest champions (and thought Toronto was in the U.S.; yeah, even back then hallucinations were a thing). At the time, running that model required a giant server farm of the likes used to run modern-day LLMs. Nowadays, that model would almost certainly run on a PC with 2 or 3 high-end GPUs and 128GB+ of memory - something prohibitively expensive for most people, but not as utterly ludicrous as the hardware required back then.

So yeah, I've said before, I think we're in for an "AI bust" in the next year or so... but if it's anything like the "Dotcom Bust" of the early 2000s, it'll just delay things until the technology catches up to the point where any smartphone can do what you need huge data centers to do today (and again, I think some people under-estimate what's possible even right now with on-device processing).



SvennoJ 7 hours ago
sc94597 said:
SvennoJ said:


The 'AI' I worked with was mostly deterministic, fuzzy logic and neural networks. It all ran on simple PCs, no need for data centers. LLMs are not trustworthy, first of all because the absorb everything without any understanding of what is real, fake, fact, truth, lies, propaganda, fantasy etc. It's SQL on steroids yet without accountability, introducing 'hallucinations', false answers. 

Just want to point out that "AI" today can be arbitrarily deterministic in the sense of getting the exact same output for a given input by setting the temperature hyper-parameter = 0 or even a low decimal value. Many AI workloads can run on local hardware; I train smallish CNNs and VIT-CNN hybrids on my local PC with training sessions being like 10 hours, as an example; online learners are also broad research field; and Local LLMs are pretty popular and capable relative to proprietary models, even if they don't top the benchmarks. I'd actually argue that open-source is in a much better relative position in the 2020's than in the 1990's, when even things like compilers were proprietary and compute was even more centralized. The big difference is the share of the overall economy that these companies compose is much larger than then. 

I also think "brute-force search" isn't quite what is happening. Reinforcement-learning isn't brute-force and that is how "LLM's" have been post-trained since late-2024. In RL, paths are pruned based on the reward function and not traversed exhaustively. I also don't think the heuristics learned in the pre-trained base models are really "brute force" either. Having said that I don't think LLMs are human-like intelligence though. "AGI" probably will be achieved through an array of casual models like LEWM that interact with generative models in yet-to-be-known ways, and there probably will have to be some online learner involved as well. 

I don't disagree with you, just wanted to say by brute force I mean these huge data centers parsing all data curated or not. It's not elegant at all, just throwing more processing power at it and hoping for something to arise.

Yeah LLMs are not human-like intelligence. Then again human-like intelligence is what got us into today's mess. :/ Not something to strive for if you ask me lol.