Because it isn't useful outside of shitting out templates, project skeletons and babby tier solutions.

Until A"""I""" can decompile a binary into easily readable code. It will be absolutely useless

File: bubble.png (139 KB, 1076x644)

139 KB PNG

>>108779577
>vibe-coding is so great and AI is stealing all CS jobs,
It's only letting """people""" shit out copius amounts of useless and dogshit web apps.

File: cocks.png (22 KB, 1013x402)

22 KB PNG

>>108779701
Can confirm, am in the middle of writing a dogshit web app.

So you saying Cuda can't be vibed?

>>108779577
6 months too early, nigger
>I told AI to refactor my code. The code stop compiling, but it is... beautiful...
and 6 months later you redditors never brought up this point again

File: dick_factory.png (31 KB, 1079x956)

31 KB PNG

>>108779772
just 6 more months AI will replace everything pls give us money just 6 more months brooo DDx

>>108779787
for sure it will happen before the bubble burst

File: dickin.png (40 KB, 1079x1233)

40 KB PNG

>>108779793

File: 78A7BF30-FCB7-446B-8840-5(...).jpg (1.29 MB, 3464x3464)

1.29 MB JPG

>>108779577
Most codetrans are Luddites and hate ai. If they weren’t ludders we would have much more advanced software

File: snailCat.png (2.23 MB, 1286x1223)

2.23 MB PNG

>>108780009
You are wrong again, being a luddite implies not using it at all, turns out everyone uses it, yet that "advanced software" hasn't come, and besides nazi-snail cat will holocaust your anus

>>108779577
because cuda cores are a hardware feature of Nvidia GPUs, they arent software. you fucking retard.

File: f75ffdef-c119-4cc3-8b18-e(...).png (2.76 MB, 1254x1254)

2.76 MB PNG

>>108780078
it takes time to build factories and revolutionize hardwares

>>108780078
Snailcats aren’t nazis, they are Luddites and they are slow. Face it, you will be replaced and left behind
>>108780525
I look like this

>>108779742
>>108779769
>>108779787
>>108779804
you have a very disturbing obsession with dicks, anon

>>108780078
based and pic is literally me.

>>108780525
I kek how AI slaves are so obsessed with just shitting out code like that's all that matters.

File: 43109AFD-4B95-4F37-A2B2-C(...).jpg (594 KB, 3383x1854)

594 KB JPG

>>108780618
You look like this (slow, weak, luddite and trans)

>>108780636
snail gods will win, cuck. slow and steady wins the race.

>>108780646
Buffcats aiGODS will win, codetrans will lose

>>108779577
The real answer is that CUDA is a leaky abstraction and CUDA code is specifically tuned for Nvidia hardware. If you could compile them to AMD they would probably be slower because they weren't tuned for AMD's hardware.
Also CUDA is kinda shit so it's not worth doing anyway.
And even if you had a CUDA compiler for AMD that's only half the battle. A lot of software comes pre-compiled and the whole build system is built around nvidia tools. You would have to replicate all of nvidia tools to even have a shred of compatibility, it'd be extremely messy.
The real answer is probably to just create kernels for AMD that do whatever you were doing on the Nvidia card.
Maybe I'm wrong and it's just a cope but I don't think I'm trying to make a CUDA compiler.
But maybe I should try to make one, I'm not sure.

>amazon winding down their internal AI bot
>microsoft quietly shelving copilot branding, apologizing to users
>openai and anthropic admitting that everyone has to move to api pricing soon
>agents still refuse to build anything more complicated than a web dashboard

is /g/ always this far behind the curve? the AI thing is over, it’s just a tool for SWE’s, a quality of life thing so we can tab-autocomplete for loops and maps. anon you… you didn’t actually think it was over for SWEgods did you?

>>108779577
the big companies already have done this and just haven't announced it to the public, why do you think both AMD and INTEL stocks have risen 120% in the past month? Insider knowledge.

>>108780765
>api pricing soon
which is already subsidized by their cash, on top of the already subsidized data center operating costs.

File: 1768481836621178.jpg (36 KB, 550x535)

36 KB JPG

>>108780078
>>108780525
can some AI do that from a single prompt or did the guy who make this prompt then tweak the photo with photoshop and more prompting to fix the text and make it meme-tier?

t. asking from my condominium

>>108780915
you can do that with a single prompt if you use models like gpt-image-2 or nano banana 2.

for example this prompt

A realistic tabby cat with a large snail shell on its back sits centered on a cluttered futuristic workstation. Arrange the cat facing the camera at eye level, occupying the middle foreground. Surround it with holographic blue UI panels at varying heights and angles: a brain-shaped diagram labeled 'SNAIL CAT OS v4.2', a protocol flowchart with steps 'OBSERVE — THINK — DECIDE — EXECUTE', a small tablet display showing 'SNAIL CAT: CHIEF OF THINKING & DOING', and a floating suggestion box reading 'A SUGGESTION: Optimal solution ready? ACCEPT / REJECT'. Place an open notebook with a pencil and a detailed ink sketch of the cat at lower left, a black coffee mug on the left with white text 'AI IS A TOOL NOT A BOSS', and a laptop with code visible at lower right. Use cool cinematic rim and key lighting, subtle warm practical lights in the background, high detail, photorealistic texture on fur and shell, shallow depth of field, 35mm perspective, dark cyberpunk tech workshop atmosphere.

makes the attached image via nano banana 2 or if you use gpt-image-2 it's https://litter.catbox.moe/pri6f8zr7v2fx854.png alternatively I could have uploaded a picture of the actual snailcat instead of just describing it with the prompt and it could have used that design properly.

File: 1778265765755-019e08e5-e5(...).png (1.78 MB, 1408x768)

1.78 MB PNG

>>108780946
if I had used AI to submit my post it wouldn't have forgot to attach the image, more proof that humans are inferior. :^)

File: dont read the book.jpg (574 KB, 2048x2048)

574 KB JPG

>>108780078
>>108780525
nyoooo stop fightan

File: 1778266356749-019e08ee-30(...).png (2.09 MB, 1536x1024)

2.09 MB PNG

>>108780955

File: 1765351230059381.jpg (25 KB, 602x402)

25 KB JPG

>>108780946
>>108780952
intredasting, thanks for the qrd
the text still looks a bit messy and parts of the background don't make complete sense but it's another giant step from that time when image generators couldn't understand text at all

another thing I noticed is that it now understands how things ought to be used
like, cables go into the laptop, are uninterrupted and have connectors instead of looking like a pile of spaghetti
that crayon looks like a normal crayon, it has an eraser that's a bit used
the techy stuff behind the cat seems believable, at least from a sci fi perspective, I mean, it has clearly defined buttons and stuff
and now it seems to understand spelling and grammar

the progress is happening very fast, it's flawed but normalfags seem to continue to have a fundamental issue with grasping the long term evolution of things
I still remember this cow and the
>ai can't do hands
meme
remember that most normalfags still believe ai can't think because it's a "database" kek

>>108781007
it's just a next token predictor bro it's just maths which as we know has nothing to do with human brains which is based on flesh, not maths

>>108781041
what if we connect the AI to a juicy chicken breast? it will give it the meat it requires for sentience

File: 1751596147987679.gif (2.77 MB, 268x498)

2.77 MB GIF

>>108780995
cool, it even came up with some legible python code related to the pic, but the number lines are fucked up. It seems that criticizing AI work gets increasingly nitpicky as time goes on but to be fair real artists shit make similar mistakes all the time

yeah, i think op is onto something.
has anybody employed ai to reverse engineer compiled executables back into comprehensible source code?

File: 1776264381122.png (505 KB, 952x952)

505 KB PNG

>>108779577
I've just got to a vibeslopped project and the only way to fastly understand what's going on is to slop myself a documentation (zero useful wikis, nonexistent onboarding, on any question you have an answer "just ask ai"). Tasks? Slopped through github api. Reviews? You guessed it. Code? Or fucking course.

I feel like I'm going insane from this soulless abomination of the job. And I will surely do if I will be forced to use Claudie after this month expire. I just don't understand why people who hate programming SO FUCKING MUCH do not just go fuck themselves at some useless middle management position.

>>108781041
You questioning the AI shizo

>>108780856
>why do you think both AMD and INTEL stocks have risen 120% in the past month
rampant speculation, but that's unthinkable because we're in the "new paradigm!!!" part of the bubble

>>108781049
shieeeeeeeet dis nigga finna gonna build frankenstein

File: 1749795084858809.png (247 KB, 620x387)

247 KB PNG

>>108780856
shitcoin tier economy but instead of jeets in discords they have jews in suits which makes it legit

>>108779804
Kill John Lennon, kill John Lennon

>>108779577
>if vibe-coding can do 99% of webslop why can't we use to make one of the most complex tasks in CS
Nigga, you have no idea of what you're talking about lol. It's almost the same as saying:
>if the microwave can do instant noodles then why it can't do Mole Poblano?

>>108779577
CUDA is an entire API stack. And it’s copyrighted to fuck.
That’s like saying “why don’t we vibe code our own AWS?”
Do you mean all of AWS? And how do you plan to deal with the lawyers?

I’d just use ROCm as the base since it’s already open source and vibecode whatever shortcomings it has with CUDA.

>>108780955
wtf is this model, good edit but text too blurry for gpt/gemini standard

>>108780430
Is that like how many barracudas are in the GPU? Haha. Sorry I'm new here

>>108780009
>i am an ai chad but i can't vibe code software because the trannies are luddites and i need them to have more advanced software
what the ever loving fuck are you even talking about anymore

>>108780952
>Strap on shell
THAT SNAIL CAT IS AN IMPOSTER

>>108779577
Cuda being _the_moat_ is a lie that keeps being copy pasted by gullible tech journalists. Having cuda is not nearly enough. nvidia could probably give it away for free(as in freedom) but better keep it closed because it serves as retard magnet
You need the whole software stack to be good fit for purpose(not to mention hardware duh), firmware, driver, libraries, yes applications too needs to target the hardware.

By the way the brand new intel cards do not support native 4bit math. Blackwell had this in 2024. AMD rdna4, the consumer cards do not have this either. The race for even lower precision isn't over either. This isn't something you can patch in with software.

>>108784243
>You need the whole software stack
So vibe code it, isn't everyone left behind by vibe coding, didn't it replace coders and can one shot every program

>>108784248
If we had access to the same information as nvidia does we could do it without vibecoding.

>>108784275
but isn't the point that you don't need access to the information since you can just vibe code it

>>108784299
With vibecoding you can fill the knowledge gaps the programmer has with the system prompting itself and searching the web. But the knowledge isn't on the web or in the weights either. Maybe if it hacks itself into nvidia, but even there I suspect some of the information only exist in peoples biological heads, not even on computers.

>>108784311
>the knowledge isn't on the web or in the weights
I thought ai replaced everyone and it can train itself, it knows assembly so it left everyone behind

>>108779577
because ultimately it's just a tool and a compliment to developers

>>108779577
There is open source cuda that huawei's gpus use.

>>108780995

really nice i like this one

>>108780714
>Also CUDA is kinda shit so it's not worth doing anyway.
Semi-Related question, but if this is the case, why why AI researchers have such a heavy preference for using Nvidia hardware? The main justification I always hear is "CUDA CUDA CUDA"

File: 1772478798189822.png (718 KB, 1366x768)

718 KB PNG

>>108784299
>>108784374
That's not entirely the case. AI by itself can get something you want done done as a starting point but only a full beliefs. They are all knowing oracles that can just pull shit out of their asses. That works the first time.

I vibe coded this space https://huggingface.co/spaces/AiAF/Civitai-to-HF (I consider this thing still in the alpha phase. Technically does what it's supposed to do currently, but I need to add more features and you fix all the shortcomings later on today)


However, what I ended up doing was I cloned and then pointed it at existing working spaces. So I got a good idea as to how a space is supposed to function and be programmed along with having it look at the official documentation:

https://huggingface.co/docs/hub/spaces

I then went a step further and get cloned the official gradio code base onto my machine and then pointed the model at Dad so that it not only knew how space is we're supposed to work but how gradient works well (radio is a core part of how most, if not all spaces operate).

If you give it the right information and ample relevant information. You can do some good shit with it. But you can't just snap your fingers and have it one shot something like >>108784248 is implying should be the case. Llms are not all knowing. I thought this was made clear all the way back in 2024. They're OK general purpose assistance and pretty decent at programming related tasks. You should also keep pic rel in mind when trying to do these kinds of things. Being deliberate about your planning, and going one step at a time will save you a lot of headache, even if it means you can't necessarily shit something out super duper fast (Which was always an unrealistic expectation even from SOTA models. Even if "AGI" becomes a thing anytime soon it will be in unrealistic expectation even for that)

>>108779742
>>108779787
>>108779804
what prompt did you use to generate the drawings, anon?