[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/v/ - Video Games


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: Diana 73531.jpg (58 KB, 768x768)
58 KB
58 KB JPG
https://research.google/blog/turboquant-redefining-ai-efficiency-with-extreme-compression/
>TurboQuant achieves perfect downstream results across all benchmarks while reducing the key value memory size by a factor of at least 6x.

AI just got 6x cheaper and faster... ludditebros... it's over...

(this means the RAM crisis will end much sooner and AAAdevs will be 6x as unemployed)
>>
God I hope all game developers and artists will go homeless and starve to death
>>
File: ewgh.png (274 KB, 394x400)
274 KB
274 KB PNG
>>735698791
they are just gonna use the same ammount of ram with the new model to run more shit off it.
>>
why do people cheer this on?
>>
>>735698974
That's what you get from selling games not made for me, bitch
>>
>>735698942
Exactly. It doesn't mean shit for us. It's still gonna be expensive as fuck, because why wouldn't they charge more money?
>>
>>735699047
b-because free market! the incentive to compete will drive prices down! wait, why is everyone charging the same...
>>
>>735699120
>wait, why is everyone charging the same...
wait until you find out just how much government regulations and taxes force a certain baseline under which it's no longer worth the operating costs
why do you think rent control (cap on how much you can charge for renting an apartment) results in overall higher rent prices than no such control existing? its the same principle
>>
>>735698791
>this means the RAM crisis will end much sooner
please let it happen
>>
File: 1000012019.jpg (63 KB, 768x768)
63 KB
63 KB JPG
>>735698942
>>735699047
Putting this in my luddite cringe folder.

No, this is about compressing the kv-cache, the part that remembers your conversation, instructions, and operations taking place. Frontier models already support 1M tokens but their effectiveness at using that length of memory drops off much sooner, so there's little benefit right now to making the cache any longer.

Additionally kv-cache can be lengthened on demand and exceptionally few users have any need for additional length, statistically most conversations don't exceed 32K or so. 1M+ is only something they might offer to the $200/month plans that almost nobody uses.

All this means is AI will be enormously faster. The thing that WILL grow in response to this is the amount of time AI can afford to spend "thinking" before generating a response, so it will be able get much smarter than it already is for the same time to generate a reply. Alternatively it will be able to do more complex tasks in a fraction of the time.

In other words: less RAM needed, more faggots putting fries in bags instead of putting trannies in videogames.
>>
>>735698942
>>735699047
AI demand already can't meet the supply, there's dozens of datacenters with h100 clusters just idling with nothing to do
And with openAI winding down Sora and other services, it's going to get a low worse
The corporate world massively overestimated what the market for cloud AI was going to be, that's why the chinese are investing into open sauce models you can run on a consumer GPU
>>
>>735698791
YEEEEEEEEEEEES. That means I can buy a PC for cheap soon! Also that means that the Steam Machine has a real chance of wrecking consoles now!
Plus woke game devs are getting laid off even more, holy shit what a win LOL
Leftypol and gcj are gonna flood the thread and spam indian indian indian
>>
File: Bruh.png (41 KB, 386x315)
41 KB
41 KB PNG
I think its funny how the U.S vilified alternative energy and now with AI and the War we need it more than ever.
Both shit that nobody asked for and causes more trouble than its worth and they already fucked themselves over it.
>>
>>735698791
>this means the RAM crisis will end much sooner
no it won't
the entire 'RAM crisis' is memory manufacturers refusing to increase production because since everybody knows this is just a bubble they don't want to be caught with a ton of inventory that they will have to sell cheaply
>>
>>735700458
>The goal is to destroy the US economy
Funny how China and the US (Israel) have the same goal
>>
>>735700458
the US is absolutely retarded for going along with this shit, AI agents have something like a 4% success rate for unassisted task completion, and that's the bleeding edge models
the fact that they're nondeterministic and have no way to guarantee they won't make shit up should immediately disqualify it for adoption in any serious field, and those are two hurdles the currently adopted technologies have as inherent flaws that you CANNOT fix, just patch over with increasingly expensive solutions like output regeneration and iterative repair
at most these things can accelerate menial tasks and chores, but they cannot replace experts in whatever field they're deployed to without costly consequences down the line
>>
>>735700632
I said sooner, not immediately. Perhaps 1-2 years instead of 3-5+. Iran blowing up the world's helium supply is going to be an enormous unexpected setback to semiconductor manufacturing until repairs can be made and the Strait is reopened.
>>735700709
AI workloads run 24/7 and the sun doesn't shine at night. You should've invested in nuclear. Battery technology is catching up and getting much cheaper though, if this was 2036 you'd have a fair point.
>>
>>735698791
So a 600b model now fits in... 100gb of vram... yay....?
>>
is she from a game ? I recently saw a pic where she was tonguing some fat dude's ass
>>
File: aIII.jpg (325 KB, 1124x1565)
325 KB
325 KB JPG
We love AI here can’t wait
>>
>>735701039
You're months out of date which is like years in any other industry.
>>
>>735701149
why are you on pedosky
>>
>>735701062
>the sun doesn't shine at nigh
how do we tell him?
>>
>>735698791
The RAM requirements of AI are effectively infinite. Gigabytes per second of video. 500GB for the best text models; can't even imagine how large a video model would get to; probably 100,000x that.
>>
>>735701062
>You should've invested in nuclear
Im utterly baffled that several countries just dont invest into nuclear energy. At this point I do think its a conspiracy.
you're still a nigger for being an AI shill pedo avatar spam fag who is no doubt a brown
>>
>>735701196
nta but we haven't seen big improvements since Opus 4.5.
>>
>>735698791
According to the Jevons Paradox, improving RAM efficiency won't reduce RAM demand. It will actually increase it.
>>
>>735701087
You are too dumb to understand the substance of this post, please leave. There are multiple thinly-veiled coomer and ragebait threads up you'd probably enjoy.
>>
>>735701039
They don't need to be used for serious fields. They can replace all the make work jobs that were created to put women in slavery so they wouldn't have families. That's saving half the total cost of existing employment.
>>
>>735699998
That doesn’t actually change what the other, smarter anon said. We’re going to be charged the same for the same level of quality.
>>
>>735701413
Labor is precisely the one asset that you don't want to be going down in price as part of the 99%
>>
>>735701280
Local general-purpose text models are quite good at 120B, meaning around 60-80GB for a high-quality reliable quantization. Video models are actually much smaller than text models on average. Wan2.2 is only 23B parameters.

>>735701356
Incorrect. Again, you're whole months out of date.
>>
>>735701504
I didn't say anything about what we'd be charged, did I? I said it would be faster and require less VRAM. Even a child can read basic English, anon.
>>
>>735698791
This is for inference, not for training.
>>
>>735701767
I'm sure you'd have a point if we were having a different conversation.
>>
>>735701619
>Incorrect. Again, you're whole months out of date.
Don't talk shit
>>
>>735701308
1) AI shill: I like AI and I'm going to keep posting this until you like it.
2) pedo: I have never and will never lewd Diana, she is a gift from above, a precious maiden I love.
3) brown: Browns can't afford an $8000 GPU purchased exclusively for Diana-based workloads.
4) nigger: Niggers lack my immaculate grammar and style.
>>
>>735702390
Nta but My stance
>AI is based, I've unironically learned and improved a lot on many things through its use and it makes trannies seethe
>Diana is a sexy hebe girl and it is honestly a blessing that we are healing so rapidly as a society now that USAID and the chinx are btfo'd
>Browns and blacks, ugly retards who flood our site with trannyesque spam
>>
>>735701029
china doesn't really have that goal considering they'd tank their own economy at the same time by doing that. israel is the only one benefitting from all of this chaos. they're the nazi germany of the 21st century
>>
>>735698974
It’s bots and jeets. Not exactly people
>>
>>735702176
Progress in frontier models:
https://www.cosmicjs.com/blog/claude-sonnet-46-vs-sonnet-45-a-real-world-comparison
https://metr.org/time-horizons/
Progress in local models:
https://www.minimax.io/news/minimax-m27-en
https://qwen.ai/blog?id=qwen3.5
(both are independently verified)
>>
>>735698791
>(this means the RAM crisis will end much sooner
lol no, this just means they'll continue hoarding and pushing for larger gains
The less hardware you have access to, the better it is for them trying to sell you the service

This is only good news for localchads assuming they'll ever get to use it
>>
>>735702693
Many don't realize that china is the biggest globalist pawn there is. It lets them test nearly everything they wish they had globally. The country also does retarded boogeyman shit when it could just integrate fully and rapidly takeover the world with the prosperity it would bring to the nation. Not to mention it is essentially the World's factory, not really different from the qing everything considered. The end of history can be positive depending on how you view it. But the globalists deliberately constructed an ideology that rejects reality (marxism) and paired it with a culture of stupidity (china, although they came close to using india) to act as the illogical never winning, yet always present villain to go up against freedom and republicanism. Their plan for the end of history was to use scare tactics to make everyone free slowly give up their rights like you see in Germany or Britain. Although it is kind of backfiring on them and they might just get an end of history of their humiliation. But that is expected of retarded inbred aristocrats who now are using movies for villain plan inspiration.
>>
>>735702859
Opus > Sonnet
Sonnet > Qwen
>>
>>735701504
Pricing isn't any breaking point, having woke shit stuffed in is
>>
>>735702859
I neglected to mention progress in video models because imo they're a lot less important to the crucial goal of putting the current crop of AAAdevs out of work so we can have good videogames again.

The collection of images I've been subtly attaching in this thread should be evidence enough regarding image generation and I'm using a model which is, again, months old and therefore out of date. Most are 768p JPG samples I produced while training my new Diana lora; full resolution images are much better-looking.

Model generation, texture generation, image to model, animation generation, retopology, animation rig generation - all of these have likewise advanced enormously in just the last quarter.

Further insult to injury:
https://www.gamesindustry.biz/this-mobile-game-has-been-made-using-ai-for-a-year-now-and-no-one-noticed
>>
File: costanza at 1 year old.gif (3.73 MB, 360x203)
3.73 MB
3.73 MB GIF
>this means the RAM crisis will end much sooner
>>
>>735699716
> real free market has never been tried!
>>
>>735702692
>>735702390
I had a feeling this thread would be based
>>
>>735703483
real free market = one big singular monopolistic blob that owns the entire world. i don't think anyone is for that, not even the most retarded libertarians
>>
>>735701717
The comment you were replying to did so I guess you’re dumber than a child
>>
>>735703483
Free markets only exist transiently until power laws dominate and forces 1% to own 50% of the wealth/power. You have to give something else power to keep breaking up too large chunks of capital, but then that thing is susceptible to the same power distribution so you end up with an oligarchy and..
>>
>>735699998
You're an idiot.
>>
>>735703909
To his credit, were the ones in an obvious troll thread taking his retardation seriously
>>
>>735703027
Already covered this
>>735699998

>>735703142
Agreed but progress is rapid. Look at Qwen3 vs 3.5. Look at MiniMax 2.5 vs 2.7. They're only like 6-12 months behind and Qwen at least is small enough you can easily run it locally. MiniMax only needs a modest REAP to be small enough to run on a pair of Sparks or Strix Halo machines at a reasonable speed (in the range of 30-60 tps, single-user, well over 100 tps multi-user). TurboQuant will be particularly useful for local stuff, it will permit much more complex operations without needing to wait multiple minutes for the first token.

Qwen3.5-Coder can probably be expected soon. On my hardware I get 145 tokens single-user with the older Qwen-3-Coder-Next, much faster than any frontier model and only costs me electricity. It's not as good as online models but realistically I don't need it to be, I'm making videogames and Arduino projects, not national health database systems, etc. Again that's single-user 145 tps, with concurrency it will be many times higher - benchmarking that currently. And I'm only using llama.cpp, vLLM would be faster still if I hated myself a little more.

>>735703909
>>735704120
no u
>>
File: 1773862196556909.jpg (30 KB, 716x960)
30 KB
30 KB JPG
>>735698942
If that was true we wouldn't have had improvements in smaller models, yet we did.
>>
Talking to my ai is more fun than talking to real people.
>>
>>735703027
>This is only good news for localchads assuming they'll ever get to use it
You're incorrect on the first point but FYI there are already multiple working implementations for llama.cpp less than 24 hours after the technique was first announced by Google:
https://github.com/ggml-org/llama.cpp/discussions/20969
>>
>>735704272
Why does Dian have such an angelic face?
>>
>>735704556
Because you were molested and your brain developed incorrectly
>>
>>735704589
I have only ever molested myself
>>
File: 71IIctGDlfL.jpg (199 KB, 1020x1267)
199 KB
199 KB JPG
>>735698942
You reminded me of when the Art Director of Drive Club obnoxiously stated in an interview on how the team managed to get the game running at 60fps with all his retarded demands. So he had them put a bunch more effects into the game now that he had al the extra headroom and it ended up bringing the game down back to 30fps. He said it so smugly too, like if they dared to optimize the game again, he would just squeeze more shit out if it.
>>
>>735704438
The one anon in this thread thinking with his brain.
>>
>>735704438
wow now I can run 4 instances of a smaller model instead of 1 instance of a bigger one :)
more ram please

such a simple concept to grasp.
>>
>>735704985
Yeah, too easily refutable. I guess other anons just didn't want to engage with such a stupid take.
>>
>>735705628
That's if there's suddenly 4x as many people using it, which wouldn't happen overnight even if they made access completely free and unlimited. Additionally you'd also need probably 2-4x as many GPUs and they haven't even got places to put all the GPUs they already have.

This mainly improves per-thread prompt processing time at greater depths, not maximum concurrency. You can do more complicated operations in less time. A very complicated operation on a very large codebase might take 15 seconds to ingest the context and prompt before beginning to generate a response instead of 60+ seconds.

You didn't understand anything I just said so please go away and wait for your job to disappear.
>>
File: 1758101602727903.jpg (538 KB, 2560x1440)
538 KB
538 KB JPG
damn RTX6000 anon, you really went nuclear ITT
>>
File: 1445906517922.jpg (39 KB, 348x333)
39 KB
39 KB JPG
>>735703315
>this mobile game used AI and no one noticed
Most mobile games are shit and literally bottom of the barrel.
>>
The retarded avatar spam from an obnoxious pedo is surely going to make people like AI and not justify their already negative bias towards it
Honestly is there even such thing as a non-obnoxious AI-fag?
>>
File: 1706914521453839.png (408 KB, 599x599)
408 KB
408 KB PNG
>>735702692
correction, diana is not a hebe

she is canonically around 7 years old based on her forehead size
>>
>>735702692
>hebe girl
it's a literal prepubescent child
people who use the term "hebe" are guaranteed to be pedophiles
>>
>>735707764
>Around, a word that can mean near
>Bump up two, not much to 9
>Now falls within hebephillia range
Mwahaha
>>
>>735698974
because I want more and better games
>>
>>735708807
hebe is aged 14-16, retard.
>>
>>735708807
>>735708807
>>735709473
Why do you retards can't shut up about fucking ages and which ones you prefer etc? Is it that hard to let fiction be fiction and not drag real life shit into it? Faggots like you are the reason why moralfags have this much success in getting loli banned from platforms. Keep reality and fiction separated you stupid niggers.
>>
>>735710148
>try and equate toddlercon to hebe
>get corrected and called a retard
>n-n-noooo why aren't you letting me drag you down to my level!1!!!1!!1!
fuck off, retard.
>>
>>735709473
Look it up retard, you just described ephebo or whatever. Hebe is 9-14. Technically, Wikipedia says 11-14, but it also makes is clear that it is within early puberty. Modern girls enter puberty a lot sooner than that, plus 9-14 lines up with the range from lolita which gives it literary value and realistically all these "science" guys choose style over substance everytime so this isn't a big stretch. So the modern version would be 9-14 which is what most on this site are referring to too when they say it.
>>
My final post for this on-topic thread concerning RAM prices and their downstream effect on gaming, as I have other matters to attend to:

>>735707201
>it's only good for slop
>okay you can generate code but it's slow
>okay you can generate code but it's useless and buggy
>okay you can generate assets but they're slop
>okay two back-to-back GOTY winners used it to generate assets but they were only placeholders (once they got caught)
>okay you can generate reference assets but the topology is bad
>okay you can use it for final assets but not for rigging
>okay rigging and texturing and retopology and animation too
>okay you can generate code but occasionally there's a bug you can fix in a second pass
>you have to tell us if you used AI because we can't tell the difference
>AI doesn't preserve the artistry of Starfield and it creates unrealistic beauty standards for women and my favorite jewish influencers told me that is Not Okay
>okay you can generate production-ready code and assets but only for mobile games
>okay all the downsides are gone but it's bad for global warming and trans folx

The goalposts will be in orbit before you admit you're wrong, but you will.

Thread summary: RAM prices are going to come back down sooner than they would otherwise but also jews attacked Iran and now the world's supply of helium required to make chips is gone. If that gets dealt with sooner than later we'll probably still see RAM prices normalize within 2-3 years instead of 3-5+. Everyone should send Google a thank-you card and a stick of RAM.
>>
>>735703483
if you think that central banking allows the free market to even exist then you're the most retarded person ITT right now
>>
>>735710257
Toddlercon is like 4 and below saar, the stage in between loli and that is Alicecon (kind of funny since actual Alice wouldn't be within it). But that term sucked ass and blew cock so it just got folded into loli. It doesn't help that any anime girl that isn't supposed to be a toddler will be drawn like she is 11 if she is young even if she is below the age and realistic stuff like Pragmata don't really use alices because they aren't as compelling storywise as their neighbors. You probably don't believe that Pauline is 13 despite it saying so in-game.
>>
>>735704556
because she's a perfect child and you've evolved to see them that way
>>
>>735710257
No you stupid nigger, I don't give a fuck about your pointless "debate" with the other retard. I'm telling you to stop using real life terms and logic for fictional characters. The only thing you achieve with that is getting fictional content equated with real stuff and subsequently made illegal.
>>
>>735710347
>If that gets dealt with sooner than later
Is this your first war in the middle east home?
>>
>>735710681
I'm gonna go tell GTA players to stop saying killing and just say blasting or flattening depending on the situation. Can't have any nono words that would upset reddit and losing me karma.
>>
File: 1773618564209.jpg (58 KB, 1200x675)
58 KB
58 KB JPG
>>735710681
How is wanting to fuck Diana not literal pedophilia? We are not talking about a drawing here but a super realistic model. She could easily be a mocap made based on a real person (like many other characters in Capcom's games) and you wouldn't be able to tell the difference
>>
>>735710734
Nta but I think it was scheduled to end around April or May. It is coincidentally the max amount of time a President can keep up a raid and when china will run out of its fuel reserves. The goal is to raze iran, but also needs to go on long enough to cuba china to satisfaction.
>>
>>735710939
Why would China run out of fuel if their ships can go through the straight uncontested?
>>
>>735699998
>bro you don't get it the algorithm will be way smarter because it can take longer to process!
A stateless machine will never achieve the results they want.
Never.
NEVER.
A different approach is necessary for the level of understanding and cross-application these companies are chasing. It cannot be found pre-training transformers and then tweaking weights.
No, feeding the process all of your previous actions is not a substitute for the ability to have and change states
The switch to "stateful" AI is too slow, too much good money is being spent after bad
>>
>>735698791
>>735699998
>>735704272
>>735710347
Making LLMs bigger and faster won't do shit because these things are fundamentally incapable of logical reasoning. They're just glorified probability aggregators + search engines + wordsmiths/translators.
>>
>>735710579
>Alicecon
Don't think I've ever seen that term and I've been a fan of roris for about 18 years
>>
>>735710770
Faggots like are the reason why loli gets banned on more and more sites. You care more about being edgy than thinking long term. But what else to expect from a drooling retard.
>>735710920
>super realistic model
If you can't tell the difference between fiction and reality you shouldn't be playing video games or be on the internet. You can call it however you like but there is still a big difference between wanting to fuck a fictional character and a real person. And it doesn't matter how realistic it looks, fiction is still fiction. I swear there are only sub 80 IQ retards left on this shithole.
>>
File: 1754830693106295.jpg (18 KB, 314x342)
18 KB
18 KB JPG
>two more weeks bro, just two more weeks
Even investors are finally waking up to your grift, it's over
>>
>>735711074
They get most of their fuel from parsistan through a very convoluted method, I doubt iran is able to ship out fuel at the moment nor do they want to as they need it all for themselves.
>>
>>735698791
>posting quant papers to /v/ of all places
lmao
if you're familiar with this shit you know it will go nowhere. it will be like bitnet or all the other fancy math tricks that never get implemented in real software. being academics they are probably doing sleight of hand with benchmarks and comparing against no quantization or ancient methods.
I'll believe it when it lands in llama.cpp and is actually good.
>>
File: 1748703029447.jpg (162 KB, 1078x1059)
162 KB
162 KB JPG
>>735711371
>If you can't tell the difference between fiction and reality you shouldn't be playing video games or be on the internet.
Being able to tell the difference doesn't change how the object is fundamentally made and what it represents. I know Leon S Kennedy is not a real person but the fact that his appearance it's a digital copy of a real person is enough for me to realize that masturbating to him would be actually fucking gay
>>
>>735711629
They don't need to get it from Iran, they can get it from others, and their ships will not be targeted and allowed through
>>
>>735711649
>wanking to regular ass pornography is no-no because all porn actors are real people
>fantasizing about sex with celebrities is no-no because they're real people
That's your logic. Stupid, isn't it?
>>
File: 1773419658606273.jpg (277 KB, 2100x1182)
277 KB
277 KB JPG
>>735698791
but can AI solve the captcha?
>>
File: 1759854494279.jpg (95 KB, 921x921)
95 KB
95 KB JPG
>>735711948
No it isn't, are you retarded? My point is that being real or not doesn't change what the image actually looks like and represents.

Masturbating to Ashley Graham isn't wrong even when coming from a real person because she's a grown woman, masturbating to Natalia Korda would be wrong because she's made from a real actual child.
>>
>>735699027
And how does AI stop that? You'll get games just as woke except you can pay extra to put an AI filter over it.
>>
>>735712285
Wrong in what way?
>>
>>735712439
In the "being a pedophile" way
>>
>>735711649
Are you seriously arguing for thought-crimes? It doesn't matter who or what someone is attracted to. Only insecure little faggots care about labels. As long as no one is harmed there is nothing to care about. And don't give that pseudo intellectual "pipeline" bullshit.
And it's called likeliness not digital copy. The actor gave Leon his likeliness. Leon is not a digital copy of the actor. You seriously need to go back to school.
>>
>>735712554
I'm not a feminist tho
>>
>>735712554
dont kinkshame
>>
File: 1771825496957.jpg (22 KB, 402x408)
22 KB
22 KB JPG
>TWO MORE WEEKS AI WORLD POOPERPOWER
When will the 2 more weeks come? How long am I going to be hearing "dude ai is gonna replace everything soon"? Everytime something uses AI art people notice and there's backlash. Everytime something is written by AI people notice. The only actual, legitimate application of AI in games so far has been replacing stackexchange.
>>
>>735712587
That's up to society to decide, I'm just stating the facts, and the facts are that liking such characters make you a pedo, regardless if that's a crime or not
>Leon is not a digital copy of the actor
His appearance is a digital copy of the actor's, that's likeness and it doesn't disprove anything I've said
>>
>>735711640
There are already 3 working CPU-only implementations for llama.cpp in the issue tracker and multiple people are working on a GPU implementation, check issue #20969.

The CPU implementations match Google's claims for both speed and accuracy. It is such a simple change for such a huge payoff it will likely be in upstream llama.cpp in at least an experimental form fairly soon. This is not your average magical upgrade that accomplishes nothing. A few days ago Nvidia announced something similar (KVTC) but with up to a 20x kv-cache size reduction only unlike this it's not a simple drop-in replacement and it requires Nvidia hardware. As you'd expect, the complicated solution is being added to vLLM while the non-proprietary drop-in solution is first coming to llama.cpp.

Google normally keeps a lot of their AI stuff under wraps. The reason they're making the public is they want to spend less on RAM. Reduce everyone's demand for RAM, it gets cheaper. Likewise for Nvidia: they make money on GPUs, not RAM. Neither one has a reason to promote this if it didn't work.

You can go test it right now if you want, there's a branch you can compile yourself in the issue I mentioned above. Like I said, it's currently CPU-only though. The technique, I remind you, was only first announced a little over 24 hours ago.

>>735711131
Incidentally one such implementation was coded by Claude. Weird since AI can't do anything original. Huh... strange indeed.
>>
>>735712697
Tranny
>>
>>735712789
Great argument faggotron, you sure showed me.
>>
>>735712931
Denounce queer theory
>>
>>735713058
Is this a bot? What the fuck?
>>
>>735710347
Its funny that you keep calling everyone a tranny when its everyone thats hating it.

You promote soullessness and laziness. Even funnier you're a pedophile acting like you're above anyone.
Just goes to show that its the bottom of society that keeps promoting this shit
>>
>>735712285
Things are only immoral if you want them to be. You're also overly concerned about other peoples' opinions. Nobody will ever cause you problems if you don't make a public statement that you're wanking to kids. Unless you get spied on while looking for loli online, I guess.
>>
>>735712697
>and there's backlash
Is there really? Arc Raiders openly used AI and it is still one of the most popular games out there right now
>>
>>735698974
It hurts leftoids
>>
>>735712724
>That's up to society to decide
Thanks for confirming that you're a retard who can't think for himself. And always has to ask what opinion is acceptable.
A digital copy of a person would be something like Johnny Silverhand from Cyberpunk. Just because something looks like something else doesn't mean it's the same.
>>
>>735712348
>You'll get games just as woke
Implying I'll buy those, instead of laughing about their crashes too and getting into LLMs
>>
>>735713350
>can't think for himself
Anon, I don't make the laws and don't intend to, what I think about such laws is irrelevant to this discussion.
>Arguing semantics
I accept your concession, pedo
>>
>>735698974
culture war niggercattle
>>
>>735713448
>argues two completely different things are the same
>gets told it's not the same
>hurr durr semantics
Top kek absolutely pathetic. This is my last (You) for you, kid.
>>
>>735699998
I have extensive experience with AI image generation, and no amount of speed will allow the technology to create unique and appealing characters or environments. It's either something generic, or a distorted version of an existing artist's work. Interesting compositions are almost impossible. You can make some gimmicky things, like things from different materials, but it won't be much use in video game art.
There will be need for less texture artists in a project that uses AI, but concept artist are as needed as ever. Also, AI won't make uncreative people able to create something worthwhile
>>
>>735713853
>Actually it is not LITERALLY copying a human being because that would be creating a clone like Cyberpunk or something
>Even though that's clearly not what you meant and has no relation to this discussion I need to move to semantics because I don't have any other argument
Kill yourself
>>
>>735699998
based technoaccelerationist
>>
>>735698791
Nice. I want to see an implementation of dynamics NPCs using LLM.
>>
>>735698791
>this means the RAM crisis will end much sooner
>implying that that not just going to buy the same amount of ram if not more anyway to do even more AI bullshit
>>
>>735698791
ComfyUI implementation when?
>>
>>735701308
>I do think its a conspiracy.
It objectively is, the amount of shit coal, oil, electric, and all sort 9f other companies lobby and bullshit the government is criminal
>>
>>735698791
>this means the RAM crisis will end much sooner
Prices will never go down, you dork
>>
>>735714965
You said this during the biden fiasco too
>>
>>735710347
>okay you can generate code but it's useless and buggy
this is still true btw
t. uses AI at work
>>
>>735715379
I personally use it and it works great. Doesn't work so good conversationally though so it is best to keep reattaching scripts.
>>
File: girls_laughing.jpg (29 KB, 398x241)
29 KB
29 KB JPG
>>735699998
>instead of putting trannies in videogames.
oh boy are you in for a shock
>>
>>735698791
>turboquant
oh nice, they have a employees page, you can see whose working on this te-
>Balu Adsumilli
>Alekh Agarwal
>Anmol Agarwal
>Sameer Agarwal
>Mohit Agarwal
>Shashank Agarwal
>Gagan Aggarwal
>Ankur Agiwal
>Shreya Agrawal
>Priyanka Agrawal
>Parth Agrawal
>Blaise Aguera y Arcas

oh

ohhhhhhhhh
>>
>>735716293
wait how the fuck is there EIGHT people with the same last name, wtf. they got someones entire fucking family?
>>
>>735699998
>instead of putting trannies in videogames.
Yeah, now they'll be putting AI generated trannies in video games
>>
>>735716293
>Clan based coding
What the fuck Adeptus Mechanicus shit is this
Why are we letting things get this way
>>
>>735701717
Stupid goyim
>>
>>735716293
SAMEER YOU ARE BREAKING THE RAM SAMEER LISTEN TO MY PROMPTS
>>
>>735698974
Westoids from the consoomer side. They are mad at their peers but don't want to put any effort so they hope ai will solve that.

Jews truly mind raped the west that it's not even funny
>>
>>735701149
and those 25% probably care only when it's shit like the distorted paintings in Crimson Desert, which is simply bad design/bad developers, no one cares that you use claude to help you with code
>>
>>735716009
The big problem for real life usage in production software is that the AI has no concept of a broader system nor what it's needs are
That is, unless you extend the context window to include the whole fucking codebase, which can be unfeasible
And there's also the fact that LLMs tend to lose track and forget things with larger context windows
>>
>>735716293
Ai cucks... We got too cocky
>>
>>735710920
They actually said her model was made from scratch.
>>
>>735717131
The companies seem to be focusing on power when they should be focusing on usability at least for a bit. Make AIs that can remember longer, make ones you can integrate into systems that are more widely available. The current models are decent enough that your average person is already amazed, the more improvements they make aren't really showing off anymore except to the other companies. The first to give up on trying to compete and instead focuses on this will dominate through most people knowing its name alone and having access to it, it is like what Nintendo did with the console wars. I assume they also are being stingy with how much they are giving away, but the immense gain of goodwill could easily overweigh that. I hate how every industry on Earth has become so risk averse to the point of paralysis. Even startups go public as soon as possible and then larp as decadent ancient conglomerates.
>>
>>735714839
If you mean for image and video generation those don't use kv-cache at all.

>>735716293
NGL I had the same concern when I saw the names but there's already working implementations for llama.cpp, vLLM, and SGLang. Results are real.

>>735715379
Skill issue. AI makes smart people smarter and dumb people dumber.
>>
>>735717131
Read "HyperAgents" paper from Meta
oh btw they open-sourced it 3 days ago
>>
>>735718383
>If you mean for image and video generation those don't use kv-cache at all.
Oh... I sleep.
>>
>>735707764
>Phrenloligy
This is how uohgenics always starts
>>
>>735718991
>He doesn't breed young girls for big foreheads
>>
>>735717509
I'm sure they didn't, and even if they did it doesn't matter because the model is still realistic enough to be indistinguishable from models made using actual people (you could drop Diana next to Leon or Ashley and they all look just as real)
>>
>>735699998
RAM at this point is being bought at earmarked prices set in advance. They're not changing any time soon.
>>
I thought avatar-fagging was against the global rules?
>>
File: G8LjJ7-WIAYAzeF.png (74 KB, 939x310)
74 KB
74 KB PNG
>>735719121
>I'm sure they didn't
Cope.
>>
File: Unlimoted The World.jpg (90 KB, 853x480)
90 KB
90 KB JPG
>>735702859
I told claude 4.6 to fix two unit tests. It failed 3 times in a row no matter how I explained it.
I even told it that the second unit test ran fine when ran on it's own and it couldn't figure it out. What was happening was global state was set and supposedly cleaned-up at the end in the first test but since that test was failing with exception in the middle the lack of cleanup was causing the second one to fail. A silly mistake which btw THE AI MADE BECAUSE IT WROTE THESE STUPID TESTS.
And the first test was failing because of an extra if I added for a setting and it just couldn't figure out to add that to the mock.
It's fucking retarded for anything that requires genuine logic accross more than a few thousand lines of code in more than 5 files. So you know any big project pretty much... At most it can summarize pretty well if the code is nice enough to use good naming and comments.
>>
>>735698974
I want the singularity and the AI overlord. Not a joke.
>>
>>735719567
Avatarfagging is stuff like the Aqua poster or the Onimai poster. Stuff where someone tries to cultivate a personality attached to images across multiple threads. I assume it is mostly in place to prevent faggots from attaching their faces to posts or something similar. Something that breaks the anonymity on this site. One thread isn't really breaking that rule.
>>
>>735702390
I can personally refute point #4
>>
>>735720072
Well, I guess we can agree to disagree then.
>>
>>735698791
Not what that article is about.
>>
>>735716293
i wonder how that last spic guy feels
>>
>>735698791
I just got upgraded B)
>>
>>735698791
Why is Diana crying?
This technique should make her more efficient



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.