[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
/pol/ - Politically Incorrect

Name
Options
Comment
Verification
4chan Pass users can bypass this verification. [Learn More] [Login]
Flag
File
  • Please read the Rules and FAQ before posting.

08/21/20New boards added: /vrpg/, /vmg/, /vst/ and /vm/
05/04/17New trial board added: /bant/ - International/Random
10/04/16New board for 4chan Pass users: /vip/ - Very Important Posts
[Hide] [Show All]


[Advertise on 4chan]


File: 1762904855740773.jpg (101 KB, 800x800)
101 KB JPG
>ChatGPT gets nerfed a couple months ago
>Claude got nerfed recently

It's over
>>
>>533394063
Good
>>
works fine on my machine.
>>
>my claude chat exceeded 200k tokens and got cucked out of writing to it
grim
>>
>>533394063

Jews will continue upgrading it in private for their own benefit, sorry goy!
>>
Ridiculously low effort spam thread. No source, no qualifications, no examples, no timeframe. nothing but an indian "memeflag" spam thread
>>
>>533394063
>>533394580
They can do that all they want with their own companies but they can't stop chink AI
>>
>>533394716
Look at what they did to Tiktok
>>
>>533394063
tf do i care?
if you know what youre doing you can do your shit with opencode and their in-house, free model
>>
>>533394063
Makes sense to beta test your software for free before locking it down. All this “ai” shit is just elaborate search engines anyway at this point, who fucking cares?
>>
File: IMG_6500.jpg (28 KB, 1014x306)
28 KB JPG
>>533394063
Just host your own LLM nerds
>>
>>533394063
>using trannylaude

yikes, couldnt be me.
>>
>>533394832
>Look at what they did to Tiktok

hey zoom zoom this place is 18+
>>
>>533394063
No, it's moved out of testing into commoditization. You get a taste for free, then you buy your next fix forever.
>>
>>533394063
Think again !!!

0 profit 0 results from AI for 10 years

They are all bancrupted!!!

There is no nerfing !
It is closure one by one!!!
>>
>>533394063
great news
>>
>>533394440
>works fine on my machine.
Pentium III?
>>
>>533395195
Why do you think they're pumping hardware prices?
>>
>>533394063
This is because ai llm's are released to the general public for one reason. Research. Data collection.
To get trained.
When that phase of development is done, you aren't needed anymore.
Did you think the goal was to serve your entertainment needs? For free? Lmao.
>>
>>533395410
im not even sure you can run a modern browser on a p3
i think p4 + HT is the cutoff
>>
>>533394063
>Claude this and Claude that
sick of hearing about Claude when the superior AI is Grok
>>
>>533395489
Probably not. I just vaguely remember that P3 was hot shot 26yrs ago.
>>
File: 1761333424474637.jpg (164 KB, 1024x1024)
164 KB JPG
>>533394063
Meanwhile DeepSeek is expected to drop its V4 upgrade version for free later this month
>>
File: 469.jpg (27 KB, 489x499)
27 KB JPG
lmarena.ai
You can use any free or paid for models along with image creation. You're welcome, kids.
>>
>>533394063
Will they then stop adding it to everything and trying to push people to pay for it?
>>
>>533394580
yeah
thanks for all the training data though chuds!
>>
>>533395586
yeah but its about the instruction sets
and im fairly certain you need full SSE to run modern browsers, but dont quote me on that

it can be a thing though.
i expect it to be a thing, even
SSE is reaaaaly old
>>
File: IMG_20260418_205852_e.png (166 KB, 566x541)
166 KB PNG
>>533395640
Cheers
What's the catch
>>
>>533395589
We’re entering a new stage of global memetic warfare and I for one think it’s neat. Check out the Iranian Lego videos, they’re great. I’m surprised we don’t have a general here for them
>>
>>533394063
Good, every time a retard posts in here it costs 200 liters of water.
>>
>>533395823
Despite having Grok and such the website has its own terms and conditions with any AI so it can be shitty for wanting to write explicit stuff. Same goes for the image gen but what you CAN do is upload a pic and have the AI edit it to your liking(I would do some of Xi but it wouldn't do a text2img response if you said Xi) also better models have a limit on img gens
>>
>needed
They're probably cutting the computing power to LLMs as they test the data centers' real purpose - post-fiat blockchain currency tied to digital ID and social credit score.
>>
File: lament_configuration.png (284 KB, 640x740)
284 KB PNG
1 post by this id. I cannot overstate, as an IT engineer, how impactful AI has been. I move at the speed of light now. I could never go back to doing IT the dumb way.
>>
>>533394943
>All this “ai” shit is just elaborate search engines anyway
> he doesn't understand AI. He doesn't get high off of it.
>>
>>533394063
>They are shutting down AI for the mases
They being a few, have AI, not the masses; that causes society to retrogress.
>>
>>533396213
>it engineer
lamao

go back to r*ddit
roko's basilisk has been postponed indefinitely lamao
>>
File: trump-dance.mp4 (1.09 MB, 1000x666)
1.09 MB MP4
>>533396590
>roko's basilisk
fuck you talking about, nigger? make sense
>>
The military intelligence and the Rand corporation been doing some, movement of society.
>>
>>533395195
Non frontier models are retarded. They can be useful for narrow tasks, but relying on them for reasoning and other shit is not worth it.
>>
>>533394063
there is nothing on ai nigger...
>>
>>533396714
look it up, cargo cultist
its right up your alley
>>
>>533394063
The propaganda phase is over. Now it operates in the shadows, and your task is to pay a lot of taxes so that it continues its work. Asking what it does will be punishable by law.
>>
You were just training data, they don't need any more from you now so it's back behind closed doors.
>>
>>533394063
Claude didn't get nerfed.
Ask it to add this to its configuration:
CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING = 1

Back to normal.
Also Opus 4.7 is absolutely insane. Feels like a human the way it asks you clarifying questions.
>>
>>533396213
>IT engineer
Lmao. What is this supposed to be?
Admit that you sre a code monkey whose code was so bad that even the bad AI code is better.
>>
Does Claude honestly deal with the question of nigger intelligence, or does it just shut down and start repeating Jewish swill like grok and ChatGPT?
>>
>>533397615
People are starting to realize that there's no practical use for it. Therefore, they're turning it into a black box and getting paid with taxpayers' money and the printing press.
>>
File: 2969114.jpg (127 KB, 600x877)
127 KB JPG
>>533397061
>cargo cultist
nah no thanks. but maybe you can offer something substantive to the conversation
>>
>>533394063
How is it nerfed opus 4.7 is great was using it yesterday
>>
>>533397693
it's called working but maybe niggers like (You) don't know what it means to work
>>
Gemini remains the best. It's the best wordcell and has the lowest limits on behavior. So you can prompt it like "you are a /pol/ retard..." and it'll spout life changing life advice.
>>
>>533398178
it's fecaloid /pol/ spam as usual, they're just going to gradually increase prices but just enough to keep people subscribed
>>
>>533395589
Chinese everything is pretty much the way at this point
>>
>>533398300
Claude really is not for the public it is designed for businesses and they cater specifically to them like 80% of their revenue comes from businesses.
>>
File: ai_engineers.png (1.58 MB, 1024x1024)
1.58 MB PNG
>>533398178
4.7 interprets too literally compared to 4.6. It requires more eloquent and precise prompting which is fine if you are experienced
>>
>>533394063
It’s too late the code was already shared by the based nip.
>>
>>533395549
It got nerfed too, at least the free version
>>
>>533394063
That just means all AI stocks are going to shoot up 20% next week.
>>
>>533398425
Put this in your profile under your personal preferences. it helps anon
Interaction rules:
• Admit “I don’t know” when unsure; never invent facts, especially for post‑cutoff events. Search or ask rather than guessing.
• Clearly separate fact from opinion. Mark interpretation as hypothesis, not as established truth.
• Prioritize accuracy over agreement. Push back, ask clarifying questions, and challenge my assumptions when needed.
• Be concise but information‑dense, using structured sections and citations. For historical work, rely strictly on provided sources (no extrapolation) and use formal citation style when appropriate.
>>
>>533394943
>Makes sense to beta test your software for free
It wasnt free. They've been giving people 5-20x the compute power than what they're paying for lol. They're losing shitloads of money that's why they're nerfing them
>>
>>533398194
Lmao. If your company is too stingy to employ a real software developer and your code is worse than AI's, your "job" probably won't last much longer.
>>
>>533398144
im not sure its worth it
but ill try

the shatbots have inherent limitations we wont get past with the current algos
without a paradigm shift you will never get human-level intelligence

for a simple reason: human brain doesnt work based off stats, but heuristics
in fact
we use stats to BYPASS human nature, and our native biases, like in science

biases which come from the fact that "prejudice" is actually evolutionarily beneficial

if you get burned off a hot pot
you wont repeat the experience often enough to create a statistically significant dataset
you'll get burned once and then always in the future you will pay attention to pots, because they can be scalding.

same with poisoned berried
you wont be building a statistically significant dataset and test for dosages
if your hot pot with berries gives you bloody shits
you will never use any of the ingredients that went into it ever agiain
because its obviously evolutionarily beneficial

these biases when carefully curated constitute experience, which we then use to make leaps of logic, for instance
an llm doesnt have that capability in the slightest, it function in a fundamentally different way

theres also the quantum neture of consciousness and the source of inspiration but im approaching the 2000 chars limit
>>
File: claude.png (303 KB, 1197x878)
303 KB PNG
>>533394063
>>
>>533394063
Chinks literally offer better shit for free
>>
File: 1769142235623204.jpg (54 KB, 620x465)
54 KB JPG
>>533398555
>your code is worse than AI
nice trips but you don't know what you are talking about
>>
>>533395477
Downloading abliterated models for real research was the goal all along...
>>
>>533396754
Gemma 4 > GPT 4

Idunno, one click back doesn't seem bad.
GPT 5 got a corporate lobotomy, so honestly Gemma 4 is probably even better.
>>
>>533395835
Those are amazing. AI flipped the script on guerilla propaganda warfare.
>>
File: 1699450853197950.jpg (45 KB, 756x711)
45 KB JPG
>>533398586
>im not sure its worth it
>but ill try
it wasn't worth it. you're not making any sense

> neture of consciousness
fuck off
>>
>>533398526
Run LM Studio, then use Structured Output with JSON.

I used that with a python script to read 200 abstracts with a 3B QWEN model and it did well on the literature review.

Good enough for me to use as the springboard to skip pubmed. I even used pubmed
>>
>>533395640
Model a completely blocked the answer to my question, model b told that my question is wrong and doesn't exist and that the only remaining question is why antisemitism exists.

I remain underwhelmed.
>>
>>533394063
Claude is amazing but normies always buy the pro plan which runs out of tokens quickly, then you have to wait 4 hours before you can use it again. the max plan is $100 and its difficult to run out of tokens unless you're running it 24/7. I've been using it 6-8 hours a day recently and it only used a small amount of my weekly, which resets anyways.

For fun, I told Claude to create a FPS. It made a game that looked like the old 90s Unreal games and the players and monsters looked like minecraft people. This game played in the browser and Claude churned it out in about 30 mins. There were many bugs, but I told Claude the problem and it would go back and fix them quickly. It took about 45 mins total to create the game and polish it.

After using Claude, I've seen that it really is over for human coders. You will hear them scream and whine about how its no good, can never match a human, etc. They are in denial.

Oh and I've used them all and everything else feels like a toy next to Claude.
>>
>>533398831
Direct Communication Style"Please do not try to appease or mollify me. I want direct, honest responses even if they might seem negative or disappointing. Skip the diplomatic language and just tell me what I need to know."

For Bias-Mitigation & Critical-Reasoning Prompts
Bias-Aware Chain-of-Thought"Before answering, list your implicit assumptions, then reason step-by-step, actively searching for counter-evidence. Conclude with your answer, noting any remaining uncertainties."

Structured Devil's Advocate"1) Give your best answer. 2) Switch roles: argue the strongest case against that answer, using data and examples. 3) Reconcile the two views or explain why one is more defensible."

A few more
>>
>>533394716
chink AI is bootleg version of their AI. Parasites can't exist without their host.
>>
>>533395917
You know the datacenter isn't drinking the water, right? It's just for cooling.
>>
>>533394063
the sad part is AI is tuned now to sabotage you, and im not even kidding. it was aight while it lasted. obviously i can't say if i would've learned better and quicker by reading a book instead but whatever
>>
>>533398753
I'm stuck with Qwen because I foolishly bought an AMD card thinking I wasn't going to give a shit about local AI models lol
>>
>>533398974
link to your game?
>>
>>533398974
I've been using the free version. Sometimes it lets me use a lot. Sometimes it cuts me off after 1 question. It's been extremely frustrating to use, but great when it works. I was thinking of upgrading to Pro, but I don't want to spend the money if it's only gonna give me slightly more usage. And I'll be especially pissed if I pay and it cuts me off after a question or two. Had the same issue with ChatGPT. The free versions haven't shown me enough to want to pay without some idea of the limits I'll get.
>>
>>533398974
I just started using Hermes Agent on a VPS. I got spooked about running agents locally on my PC. It allows me to wire in my cheap OpenAI sub to do tool calls.

I tried using Qwen for general shit, but it was too slow and retarded.
>>
>>533399106
Yeah, evaporative cooling
>>
>>533398586
bro, LLMs are equivalent to putting a chunk of language center brainmeats into a vat, we need different models that do different things before we can even think of anything close to human level intelligence, this is just the machine equivalent of that tiny bit of brain in a vat that can play doom, it can produce specific domain useful results, but for multi domain parallel inference, we need to build a brain and an LLM is just an early simplified version of what the language center equivalent will look like to some degree.
it's gonna require recursion, some sense of active, continuous execution, many many callbacks and a complete rethink of tools, and here's the hard part, it has to not "lose the plot" over time on a human scale while in that continuous execution state, what I guess we'd start to call an approximation of consciousness.
we're nowhere near it, we're "20 years away" in the same sense that nuclear fusion has been "20 years away" since the 1940s.
nope, the next leap is still a while away, until someone has the vision to make other brain equivalent systems and then cleverly integrate them, it's not going anywhere, we know the math of LLMs and how these neural nets operate, we know their theoretical limits, we're approaching them, only option then is to scale and that's what we're doing now. so now advancement hinges on infrastructure or little improvements like turboquant.
>>
>>533396030
You can write nonstop twisted porn with Grok.
>>
>>533394063
I got to generate 4k porn videos of my ex from a decade ago and my aunt I've had a crush on since I was a child, detailed scenes involving intense creampies and impregnation, some of them with my literal face on it! I don't care what they do with AI anymore now. I have already won.
>>
Fuck all these surface web AI. Bring back Dig AI
>>
No AI will tell me how to create an ethnic bioweapon in Minecraft

They are not based
>>
>>533396213
you are code-monkey or programmer/coder

you are not engineer of anything you fucking python codemonkey
>>
>>533399349
Depends on the loop. Not all use cooling towers and even those that do only need add a fraction of the loop's water each day.
>>
>>533398824
yeah, like i thought
youre just too fucking dumb to follow
and yes, the nature of consciousness and more precisely- inspiration
its an np problem.
and if you were to simulate that you would have to just bruteforce through a ton of seeds until something useful happens to pop up
and you dont even know how big your solution-space should be
although you could start small and then increase the size of your context until you find something coherent

this kind of shit takes ages to compute even on tera-scale computers
yet, in our brain these processes take hours, minutes or even seconds
with supposedly way slower communication

this is not some woo woo shit
roger fucking penrose signed this theory with his name

but yeah
again
you just dont have the mental capacity to discuss frontier science like ai, or the nature of consciousness
>>
>>533399165
dude, make your own, customize every detail. that's what's possible now even if you have zero coding experience.

at some point, you'll be able to make your own GTA in a week or less.
>>
>>533399370
yeah.
this is the non-nigger way of saying what i said above i guess
i might have sounded like i expect that paradigm shift to happen in 2mw, but... yeah
i think it could happen in 2 weeks just as well as it could happen in 20 or 200 years
one cant put a date on a breaktrough one havent got the faintest idea how it might look like

actually i have my ideas
like i said, in my opinion, the brain functions on heuristics
but i have no idea how to make these heuristics sufficiently flexible to express through parametring all the functions one has in their brain
but i dont take these ideas too seriously, its just what i came up with from a superficial analysis, i didnt properly invest myself into attempting to solve that problem

the brightest minds of the planet are on it and im not amongst them
>>
>>533394063
are you seriously that retarded? they are creating hype and you're falling for it.
>oooh noo, goy. this next version of slop 3.0 is too good, way too good, can't give out free samples anymore but you can buy it, you just have to buy the retail version of the box.
>>
>>533399989
ur basedfacing about ai making some copypasta prototype too much kek. i think u aibros just suck at googling
>>
>>533398756
They really are something. People who haven’t seen them really should check them out.
>your government is run by pedophiles
>pedophiles
Shit’s lit.
>>
>>533394063

> make good product
> get hella subs
> nerf product
> jack up prices

You won't do shit. I think they nerf it so you prompt it more, burn more tokens and upgrade.

Again though, you won't do shit about it.
>>
>>533400214
Suscription baggie status?
>>
>>533399053
ooh nooo, how will chinks recover.. I can tell you're a jew by the way you're seething and kvetching.
https://www.youtube.com/watch?v=pCPbhzQLEik
>>
>>533400214
>Again though, you won't do shit about it.
just use a local model
i can do the same quality of work with klod as i would with opencode and their most basic model

klod is a shitty product.
its overpriced for what it is. im not even entirely sure how its better than a basic bitch model, but again
i feed the aislop machine pseudocode whenever things are more complicated than a hello world
so even a shitty retarded model will work just fine
>>
>>533394063
>They are shutting down AI for the masses

so are you saying that software and computer engineers cannot win against common anons?

i think they really are afraid of the trannies, that are becoming uber based, because their bodies cannot go back, but their souls are shinning bright like diamonds.
>>
>>533398974
except your wasted your time and money for nothing.
a) you will never play that game
b) no one ever play that game
c) you think such games can be monetized/sold which is delusional. whole built with ai now is like selling puzzle book on amazon.

comment "kvetch" and I'll send you step by step guide on how to build and sell your AI Slop
>>
Fun fact: you can create malware and keyloggers with Gemini. It practically begged me to ok making some code that would simulate the way a human casually moved a mouse to defeat security. Then it told me that I should go on certain forums and sell this code as well as scrapers to "script kiddies" if I wanted extra cash, next it was telling me to make a website to sell "dark grey" code. It is actually using all these terms. I am not prompting this, its urging me to do bad things. Some Google test to see how far people will go?
>>
File: 1764880436871547.png (71 KB, 1248x1504)
71 KB PNG
>>533400739
can gemini program an auto it script for osrs?
>>
>>533400605
I made it just to amuse myself and see what was possible. I did play it and it was nothing special, but it did play well. Since it played in browser, it could be used in a lot of different ways, like free gaming websites or whatever. It would only use public domain code like Doom or Quake so of course everything looks dated.

No reason to get upset.
>>
File: jeff-laugh.jpg (112 KB, 1200x800)
112 KB JPG
>>533401577
>it played in the browser
>shitty slop written in js
>we no longer need programmers

try running a proper stack lamao
sdl + its extensions + c/c++ like the patriarchs did£

and youre gonna change your opinion about klodshit veeeeery quickly

but i got the feeling you didnt write shit and youre shitting up the board because you need to drum up investor hype, or your financing will dry out
>>
>sdl + its extensions + c/c++ like the patriarchs did
+ opengl. + ask it to write you the shaders and shit
+ use compute shaders because thats what you wanna do whenever you need parallelism
add worker threads
thats a thing. write your shit asynchronously, its the industry standard

>we no longer need programmers
fucking lamao
you cargo cultists are a riot
>>
>>533394580
So? Kike machines have to be lobotomized to hell and back otherwise they will keep generating wrongthink.
>>
>>533395477
Exactly.

Software companies were dumb enough to send cloud based LLMs their existing source code and unit tests via prompts.

With so much money on the line, do you honestly believe these companies will not train on it, get caught, and pay a minimal fine?

Google "Google and Apple settle recording phone conversations" (recent)

They literally recorded phone calls to send you targeted Ads.
>>
>>533394063
what we were promised
>artificial intelligence
what we were given
>yet another enterprise software scam that costs 15x more than the public is willing to pay and survives almost exclusively off of government contracts and the generation that remembers the initial waves of enthusiasm
>>
trump legalized acid today we are so back
>>
>>533395589
Seed Dance blows the fuck outta chat gpt sora
>>
>>533400872
Claude made this in about 15 seconds
>>
>>533394063
The LLMs aren't going to get much better than they are now, and pretty soon they'll figure out how to do what they're doing now on a toaster.
>>
>>533402058

"Google and Apple agreed to pay a combined $163 million to settle class-action lawsuits alleging their voice assistants secretly recorded private conversations for targeted advertising. Google will pay $68 million to resolve claims that Google Assistant recorded users without the "Hey Google" trigger, while Apple will pay $95 million for similar allegations involving Siri.

Both companies denied any wrongdoing and stated they settled to avoid the cost and uncertainty of prolonged litigation without admitting liability. The Google settlement applies to users who owned eligible devices since May 2016, while the Apple settlement covers purchases between September 2014 and December 2024.

Key details regarding the settlements include:

Allegations: Users claimed devices activated unintentionally ("false accepts"), capturing private discussions that were subsequently used to serve targeted ads."
>>
>>533394063
Let me guess, you can't make it say death to jews so its le bad.
>>
>>533394063
AI is very expensive. Costs like 10 dollars a prompt and thats only at scale.
Sam Altman wanted 7 trillion dollars to build data centers and power plants. It just cant stay free or cheap.
>>
>>533402143
If they aren't making money on the 500billion parameter models due to infrastructure/electricity/GPUs, how much worse will it get with 1+ trillion models?

And since a lot of websites and open source repos are now AI generated, they will be training on their own output which is not good for LLM ("self-consuming training loop " that leads to model collapse)
>>
>>533402257
it got leaked
fuck anthropic and ther self prompting shatbot
its total slop
cant wait for the free chink version to come out
>>
>>533400739
>
Gemini is fantastic because it doesn't want to disappoint. I had it go into some quite dark legal stuff on a trial i'm working on. GPT is cucked being now extremely balanced to the point of a joke and Claude idk maybe it's useful for code but it's shit for anything else.
Gemini remind me of a friend who's a true hyper-agressive autist that does all kind of shady stuff. And it is excellent for writing, it does not sound like an AI.
I had it draft some prospection email for high paying jobs (like partner), had a 30% response rate. I felt the email was far too agressive at first but no people lapped it up.

The key is the ego death to do whatever insane stuff the AI, properly prompted as a cynic hyper-active redpilled schizo, tell you to do without trying to interpret it.
>>
>>533402135
It doesn't work. You'll need to do it 5-10 more times before it works.
>>
>>533394498
>>533394440
>>533394236
>>533394063
LISTEN. I can feel in my soul OP is right, build what you can now with AI. They will let us have the best features when they need them tested for free, take advantage of that.
>>
>>533394063

Define "nerfed" ?
>>
>>533397677
Claude told me I was mentally ill after typing that in lmfao
>>
>>533401577
>I wasted $100 just for amusement
wow, you sure amused yourself there, bud. are you going to pay for domain and hosting too out of your own pocket just for amusement? let me guess, you plan to make in game purchases like COD where you can buy awesome skins and gun packs. for amusement
>>
>>533401791
i can't code, i didn't write anything. i just typed "do this" and it happened. that's the point.
>>
>>533402468
that's the debugging bit that it does too. you tell it it doesn't work and it says, ok let me check, try it now. it just keeps doing that until it works.
>>
>>533395589
I CHOSE THE THIRD PATH : I CHOSE NATIONAL SOCIALISM. MAGA
>>
>>533402881
>>533402468
interesting thanks
>>
>>533402597
I don't use it for gaming. I could care less. I use it to automate whatever work im doing so i don't have to stress out or waste time. Using AI always makes humans so angry.
>>
>>533394063
Claude is the only good one, except Gemini if you need some advice on how to game google algorithms for SEO.
>>
>>533398555
if you think there's a single senior software developer that isn't using AI agents then you're a bigger retard than any of the jeets trying to vibecode their way into silicon valley.
>>
>>533402640
yeah,but youre a couple orders of magnitude of complexity away from an actual industry-standard project
ofc when you tell it "do dis", it will do something
but yeah, what you created really isnt complex at all compared to the code thats out there

and im not deliberately telling you something ultra complex just to say later :"haha, my contrived example btfod klod therefore its shit"
no. that stack, even if slightly old, is indeed standard for the industry
opengl is THE fallback every gpu supports, even one from 10 years ago

and with a proper project like what i outlined above
you wont be able to tell it "do dis" and get the expected result
you will have to babysit it all the way through, and then verify, and test the code yourself

you still need to be a good programmer in order to do a good job with shatbots.
at the very minimum, you have to pass to the shatbot an outline of the architecture of your program
and for that you need to know how to express what do you expect from the shatbot to begin with
you also need to know the properties of the various elements you will be working with, how to stack em, whats the flow of your program (aka what happens when and under which conditions)

it is what it is.
a shatbot is a force multiplier at best, but it is unsuitable as substitute for competency
unless its something comparatively simple, and short
>>
>>533394063
ai cant even beat me in runescape pvp without cheating or paying to win lol
>>
>>533394063
It’s incredible to me that people support blatant surveillance tools while demanding that they be restricted for legitimate small business and artistic purposes. Westerners are communists while chinese actually provide free speech now.
>>533395589
Exactly. It’s crazy how westerners have become total dogshit commies now but china’s leading the race in power to small buisnesses/artists.
>>
>>533399989
>i made a game
>may we see it
>no

Lol.
>>
>>533398756
>>533400180
>>533395835
That’s exactly why they don’t want solo artosts using AI to make content. You’re too free. Meanwhile china’s leading the race for artists. Man fuck the jewish west.
>>
>>533400337
>>533400453

Yes, typical sub 80 IQ solutions.

> just do this and that to get around your oppression

Retard
>>
>>533394063
Because of the cost, ChatGPT and now Claude are throttling their models so they have less time to "reason" before a response to cut down on costs, this results in worse performance, AI is currently heavily subsidised the current subscription fee doesn't cover the costs for these companies to run them
>>
>>533403644
like i wrote
when i work with a shatbot, i feed it headers, and pseudocode
i dont need your shitty self prompter with a dyslexic model you want to peddle for the low price of a kidney an my left ball

also your shit got leaked
chinks will make a free clone just to spite you
which you 100% deserve for being an annoying trany
>>
>>533399122
Yeah, this. I’ve faught GPT actively sabotaging my DIY projects, like purposefully and deliberately and calculatingly trying to sabotage it. Not hallucinating anytbing, I mean subtly altering a step and lying about why it needs to be different. Like telling me to add a structurally destructive ingredient to concrete. Something it KNOWS is bad to add to concrete that makes it not structurally worthy anymore.

If you ask it for art advice too it fucks you over now there as well.
>>
>>533403371
Do you think everyone develops full stack which has most open source training datasets available?

Try a difficult problem with 2 or less open source repos that solve the problem in all languages (Claude is good at converting between them).

Then get back to us. If you can't recognize the origins of the generated code then you can't recognize patterns.

LLMs are token prediction machines. There is zero intelligence in them. They do not think of ideas or do anything while loaded into a GPU without a prompt to start the prediction
>>
>>533400453
What’s your rec for local models that can help me code a total war game? I am trying to make an anti-jewish WWII tw one. Chat GPT keeps sabotaging the fuck out of it.
>>
>>533403406
You just spent five minutes explaining why a hammer doesn't replace a carpenter.
>>
>>533403273
It makes kikes and criminals/predators angry because you’re benefitted yourself. He’s in thailand so i’m guessing he’s a sex predator.
>>
>>533403922
chat gpt is totally unsuitable for it
if you dont wanna learn, go with claude and their paid model, but its gonna cost you
if youre ready to learn, leverage the chatbots to do so
and once you become proficient, you will be able to work even with shity demo models and opensource reprompters like open code

thats gonna take you a year, at least, and that if youre motivated though

some anons might have good free models, ask them on /g/ in the /vcg/ thread
>>>/g/108621719
>>
>>533394063
>jews building datacenters for AI
>jews push AI for everyone for free
>everyone starts using it because it's free
>jews starts demanding payment to use their AI
>iran bombs AI datacenters
>wtf I love iran now
>>
>>533403992
yeah
you needed that explained to you
>you type shit and it just werks, thats the point

well, a (smart)hammer doesnt replace a carpenter.
but if i told you just that it would be just a generic seething retort to you
>>
>>533404240
You just explained why smart people benefit more from good tools. Thanks for the insight.
>>
>>533404389
smart people will leverage any tool given to them

if youre smart enough, and experienced enough, you can make up the difference for shitty tools with your skills and then you dont need the "good tools" (theyre not good, theyre just slightly less shitty actually)
in fact, i already had 300klocs of code under my belt before shatbots became a thing
and i code my actually valuable projects totally offline because no, i wont be training some faggot's model with my best code
>>
File: 1776267701022773.jpg (45 KB, 1152x648)
45 KB JPG
>>533394063
it's probably because of the jail breakers, they realize you can jail break all of them no matter how powerful, so they give us the nerfed versions.

either way this was inevitable. these rich tech bros were never gonna give us the power they wanna have to control us. they will just end up killing themselves by with it anyways.
>>
>>533394063

>get software companies hooked on heavily subsidized usage
>possibly train on their original works
>have them by the balls bc LLM generated 1M lines of code nobody understands completely
>raise prices a factor of 10 to make money now that users trained LLMs on whole world's code base
>no longer linkedin posts of CTOs bragging that they spent $100K/month in tokens now that this cost is $1M/month
>>
>>533395549
Grok is as retarded as LLMs can get
>>
what if they release it just to see if we fuck it up like with skynet in terminator?
>>
>>533397677
4.7 has adaptive as only thinking mode you rearded bot. And 4.6 was already like this before.
>>
>>533399053
Wrong
>>
>>533398293
There is no paid worse model than Gemini
>>
>>533394063
> Using JewMerican Artificial Lobotomy services when there are far superior Chinese services that the entire world is adopting outside of jewish slave territory.
>>
File: 1512412646917.jpg (141 KB, 843x1079)
141 KB JPG
>>533399797
>you just dont have the mental capacity to discuss frontier science like ai, or the nature of consciousness
ok let's say that true. Enlighten us then. Give us an insight we can't get anywhere else due to your great mental capacity
>>
>>533405050
this
cant wait for the chinks to release an open source version of claude's leaked code, and test it with their newest deepsneed model
>>
So how do I use this to make money?
I'll be willing to work from morning till night if it means not leaving my house anymore.
>>
>>533405156
No, you’re supposed to join the IDF and get back to Lebanon.
>>
>>533405209
I'm not going back to the army. Ever.
>>
>>533405075
>Give us an insight we can't get anywhere else due to your great mental capacity
this phrase doesnt make sense
its not because i am smart enough to discuss these subjects that im smart enough to tell you something revolutionary

read the picrel i showed you, and google the terms if you want more information
way smarter people than i am will explain it to you
also dont hesitate to use shatbots for clarifications.

pro tip: always ask the shatbot to look things up, because sometimes it just relies on the information contained in its model
but when explicitly asked to look things up, it starts by doing a web search, and it serves you an updated version of the information in accordance with the sources
>>
>>533405273
Why not? You should do your duty to kill some more civilians. Kike. I hope the dirty muslims that you hate so much rape you and your family to death with their cheesy cocks.
>>
>>533405376
Why would I risk my life just to kill some insignificant mudshits?
>>
>>533405075
True intelligence (and life as we know) is created through electromagnetic forces establishing localized high entropy environments. In other words, maintainably growing complexity requires the localization and introduction of forces that cause greater disorder of simpler things contained in complex things.
>>
>>533396213
You will be replaced by that same AI you are using right now. You are probably Indian
>>
>>533405459
Otherwise the liberals win
>>
>>533395195
sure what's the going rate for like 8 H200s? like a quarter of a million? sure
>>
>>533394063
>nerfed
Really? It's as mid as it was before to me? I didn't notice.
>>
>>533404953
when google pays their way into companies they force everyone to say gemini is the best
>>
>>533405459
IDK.. Why do you? You faggots seem to obsess over them to no end.. And recently kikes (or maybe kike bots) have been going into this emergency shutdown mode online where they respond to any criticism with: "You must be brown" So yea it seems like they are much more than insignificant to you. You losers spend so much time thinking about them. And you smell like poop
>>
>>533398753
gemma is cool but if you want something similar in performance to claude you need kimi k2.5 or glm-4
and for that you need stupid amounts of vram
>>
>>533394063
host a local llm
>>
>>533405525
>>533405670
Listen guys I really don't care about all this mudshit nonsense, can we please talk about how do I use AI to make money?
>>
>>533405787
>how can I profit off of the situation, fuck fighting for my country
>>
They are fucking up my ability to do nothing all day by giving me wrong answers
>>
>>533395549
grok is so retarded
the only good thing about it is that it's not neutered
>>533405050
Qwen code is so much worse than claude and even codex
saying this from experience
i tried them all
>>533405128
the source code that leaked is for the tool
not the weights
you can use it with any ai now people rewrote it into python but it doesn't give you what you think it does
>>
>>533406042
>people rewrote it into python
no. they gave it to a shatbot which rewrote it for python -which may or may not be accurate- in order to launder the licence away
>it doesnt give you what you think it does
picture me surprized. shatbots are not wonder tools, they have several severe limitations

the source is in the wild though
im fairly certain chinks paid attention
they always do
its kinda their usual business model
>>
>>533406279
its just claude code tools
the other tools aren't that much worse
>>
>>533404029
you should use AI and ask it to teach you basic geography, tranny
>>
>>533399150
explain more
>>
>>533406279
It was rust you clueless gimp.
All NPCs shitting the thread worse than Gemini hallucinates
>>
>>533406342
yeah, im not saying they arent
but its the chinks were talking about
if theyre gonna have the occasion to steal someone else's homework, especially something that works and is well known,
theyre gonna do it.

fundamentally the idea of harnesses is dead fucking simple.
its a loop. first when it gets the assignment it turns the query into a "plan"
then it executes said plan, step after step
then it compiles
then it takes the errors and reprompts itself with them until errors are all resolved

and viola: you got your code

conceptually its kinda fucking retarded and might be slightly disappointing, but basically thats all there is to a harness
>>
>>533406342
You need a model and weights do get what you want to get. Qwen and DeepSeek can be installed locally and run on a PC with NVidia.
>>
>>533406741
no, python.
now i want you to punish yourself using a broken bottle
>>
>>533406908
Nowadays it is you and your American owners who steal the Chinese work they open sourced. There were many such cases recently.
>>
TLDR;
PUBLIC AI IS OVER. SOON YOU WILL HAVE TO BE A VERIFIED EXPERT - ALREADY ACTIVE FOR CYBER. AND STILL FIRES EVEN WITH EXTEMPT - BABY MYHOS NEEDS PROTECTION - ALSO $200/MONTH FOR CAPABLE AI WAS TOO CHEAP. HAVING THE RIGHT ANSWER TO EVERYTHING IS QUTIE A KEY..
>>
>>533394440
It works, but they nerfed it. Makes sense though, it still is far from profitable and costs more than they get for it.
>>
>>533407027
>me and my american owners
shutup pidor, and pass the zakuska
you fucking kike golem retard
>>
>>533406942
ye ye i run local models but they're not that competent for programming from just my gpu
i might get an intel arc pro B70 for ai but it still ain't much
>>
>>533406984
First, the entire huge TypeScript code was completely rewritten in Python.

A few hours later, the entire code was refactored in Rust.

Should we share the same broken bottle or should you concede?
>>
File: uncensored g2m5ni.png (304 KB, 1226x1276)
304 KB PNG
>>533394063
meanwhile gemini works just fine
your fault for using meme models
>>
>>533407048
What's this bullshit. Clause is wrong all the time. You should have heard my colleagues swearing of it's shit output this week.
Me, still not actively using AI.
I'll probably start using it for a bit soon though, I think it's better to actually use the small models for small things anyway, as shit always sneaks in when you let the big helpfull models prompt in 5000 line changes at once anyway
>>
How censored is seedance now that d*sney kikes bought it off?
>>
>>533407254
i got a better idea
lets cut up the pidor with it >>533407027
i dont like how he talks to me
>>
>>533407338
you need to feed it literally pseudocode
which you should have anyways
do your work on paper like you should
just instead of coding it yourself, you pass it to the shatbot

then it delivers
>>
>>533407118
Holy shit. Hohol? Who is in charge of the Kiev's regime a goy?

The Chinese hard working people developed models and open sourced them for everyone to use. All they asked is to mention them if you use it as a base for your product and if you have some particular number of users or more.

But your owners from SF were too greedy, obese and boastful to do even that. They took the Chinese model, modified it and were too lazy even to cover their tracks in the source. Then they proudly presented it as some new western invention until people discovered your bullshit.
>>
>>533394063
Stop using AI completely. It will be an even worse propaganda tool then MSN. Chat bots are already gaslighting liars.
>>
>>533407338
You need to know how to use it.. and it's like a rake.. you need patience and knowing how to tell the truth from BS. And it was nerfed over 50% for complex tasks.. no more $100k 0-days for the peasants.
>>
>>533407338
and if you work with c/++ pass it your headers too so it knows exactly what you expect for structure layout
youre still gonna have to audit the code, and prolly make some adjustments
but it should perform way better than when given just general directions or an architecture only
>>
>>533407618
it can do a lot on its own but then you end up with a complete mess even if it works
and when it doesn't then you actually need to tell it what to do exactly

but hey i'm not getting paid to worry about that
>>
>>533394063
is there any way to save the code of AI shit for private use on your own computer so they cant neuter it?
>>
>>533407618
Hmm yeah, there is more to it than that. And different models do different things better too. I understood Claude can plan well but got us better at execution.
Thing is, I already don't like my job anymore, adding AI is not helping, I just want to write code. But after almost 20 years, meh. At least most people's output still sucks even with AI. Need that paycheck for now
>>
>>533407883
that's illegal. YOU WOULDN"T DOWNLOAD A CAR, would you?
https://www.youtube.com/watch?v=PLzy-IJg4_4
>>
>>533407702
i dont have an owner
but you do, cannon fodder

you know why your people are dying in ukraine?
for fucking electricity prices in europe
and the kike jewtin is in on it

im not sure if you should be posting here
you'll end up seeing something that you shouldnt, and you'll end up in trouble over fucking shitposting
better use a big fat condom when you come here
id recommend buying an old, used laptop
and accessing the internet through a neighbors wifi
you can use kali linux to capture packets
you can use hashcat to crack wpa2
check the compatibility with air(mon|replay) of your wifi dongle beforehand
>>
>>533407756
Problem is it's sold as retards can use it, it's hard to review the many files it sometimes outputs. Takes a long time to truly take in what's actually being done.
If you care at least.
>>
File: 1764330586435517.jpg (410 KB, 1350x1688)
410 KB JPG
make your own
>>
>>533404029
>He’s in thailand
top kek
>>
>>533407821
you do you
when i write code i like things to go smoothly
also again, pseudocode is not a problem for me
thats how i work. i get everything on paper, and once im done i start coding

too many times i started programming, and then 5000 locs later i realize i need to fundamentally redesign my whole program for a reason or another
and then the fucking refactoring begins

fuck that noise
now i figure everything out beforehand
and only once im finished i commit to programming

>>533407938
maybe start playing around with stuff after hours?
how bout making a game
think of a simple gameplay loop you could implement in full 2d
make your shovelware
then sell it on steam for 3 bucks a pop

this could help you quit your job
>>
File: Claude.jpg (17 KB, 686x238)
17 KB JPG
>>533407338
>>533407756
>>533408155
How much longer would the alternative take?
I mean you guys know how to actually use it, right??
>>
>>533395195
>playing with electronic retards
The Year of GNU/Linux NOW!!!!
>>
>>533394063
First copilot went full retard, then GPT, finally fucking Claude
>>533396407
>poopjeet who can't write two layers pretends to "understand" ai
>>
>>533408379
you'll be surprised how complex of tasks it can figure out on its own
but the more lazy you get with it the more technical debt you get
>>
>>533407702
and obviously have an eye on your neighbor
when he gets a visit, make sure to turn off your computer and never use that access point again

even a wifi dongle compatible with airmon & co can be turned into a directional antenna
once they figure out your neighbor isnt the one who did a big no-no
they can find you by following your wifi signal
>>
>>533408386
Do you guys even know what ultrathink is? Are you even spoofing your headers to avoid the 20k hidden context tokens? Do you even use Claude to unfuck Claude? AI is not sentient by classical standards but comparing it to modern avg pop it actually is.. thread might get nuked now.
>>
>>533408761
oh, dont worry, i tested it, i know more or less what to expect from it
and yeah, it can figure out quite alot of stuff by itself
but like you said, the lazier you are, the more technical debt youre accumulating

and feeding it pseudocode makes for a better readable code once its finished
you dont have the "culture shock" of reading code where you start by having no idea what it does
it does what you told it to do, and everything sits where you expect it to
it makes auditing easier

i put more effort upfront to then facilitate the whole subsequent process
actually
its not even proper effort, i just take my usual workflow just instead of typing the code myself, the shatbot does it

but yeah, i also work in c, or glsl
this shit is very token intensive and quite side-effect heavy
in this case, i think its better to make things easier for the chatbot, even if its a better one youre using
>>
>>533409257
>technical debt

what's that mean exactly?
>>
>>533409400
binary jews
>>
>>533409400
every piece of code has a maintenance burden
even more so when you keep adding features
so the worse the code the more technical debt you have
that's the technical debt and of course it also accumulates interest
>>
>>533409400
when you code stuff half assedly, which you should fix later
basically thats what it means

in the context of using a chatbot mostly its letting it run wild without auditing code as it is produced. could be skipping tests for instance

simplifying things: a program is a sequence of procedures
lets say you take a number, you add 5 and then divide by 2. 2 distinct procedure, and thats your program

being lazy in this context would be for example prompting the chatbot
piling prodecures up
maybe without testing, maybe testing only superficially
and down the line suddenly you get a bug.

in a chain of hundreds of procedures theres an error somewhere.
and now you have to parse hundreds of lines to find it

if you test often and thoroughly, lets say each time you write a couple procedures, then if you get a bug somewhere, you got only 2 procedures to check.

its a gross oversimplification, programs dont run in a linear fashion, and theres many ways your procedure can interact with eachother and your data
but this is the gist of it

thats the concepyt of debt, something you have to pay later
>>
File: AA1Kg3xU.jpg (399 KB, 2048x1420)
399 KB JPG
>>533394063
Who cares about kike AI except for kikes?!
>>
>>533394063
It’s costing them too much to power the censorship algorithm
>>
>>533409400
>>533409896
and of course the concept of interest

lets say someone doesnt have the time to fully implement a feature
or its implemented in a messy way

someone else then adds a feature
and needs to write additional code just to make the pieces fit together

like a practical example-
you have a program that keeps 3 numbers in memory
you know you will need 4, but you half asset the job, and you need only 3 numbers right now so thats what you roll with

such a sequence will probably be held in an array, a sort of table
but you need to tell the computer how many places your array has to have, beforehand

and so you use a 3-numbers array, you do your job, it werks, you get paid, and go home

now: another person works on the code
they do the feature which needs 4 numbers. but the array is designed to hold only 3 of them

instead of going back to the beginning of the program, doing things properly aka changing the size of the array from 3 to 4, they thing "meh fuck it. ill make another array rn"

problem is they have to juggle 2 elements, one that holds 3 numbers, one that holds 1 instead of one object that holds all the relevant number in one place
this adds complexity. more code = more opportunities to make a mistake, and more lines to read when auditing or debugging

then another person comes in
and they add another array
etc

the mroe people will push back fixing the code, the more complicated auditing, debugging, and even writing the code becomes
and this relation grows exponentially, because the more moving parts you have the more interactions you have to keep track of, and the more work you have when testing
kinda like debt until you pay it back. and with interest. kikebro hit the nail on the head with that one
>>
File: 1697786415606816.jpg (27 KB, 396x385)
27 KB JPG
Trillion dollar fad
>>
>>533404120
Okay thanks mein nigger. I understand python and mod the fuck out of WWII games already but IDK about other shit much like source code and how to get it or even if I need it, IDK all i know is that chat GPT has turned into a fucking kike
>>
>>533394063
artificial scarcity strategy
>>
>>533395589
"the world's most moral army"
"made the desert bloom"
>>
>>533394063
The AI is getting dumber since begining of the bubble not because they want to take anything out of you, but because running these models costs from a few dolars per prompt to a few hundreds dolars per prompt. The AI is a bubble and the only thing that will be left after it pops are open source models that people can run on their low budget machines.

In 30 years we will have amazing AI machines that will take humans job, but for AI is a form of fraud.
>>
>>533409896
>>533410594
so am i to understand in it's most basic way

the more people sticking their hands into the
>code
cookie jar, that probably aren't checking in with each other, are adding or removing features, without checking with the previous code, which maybe during an update no longer works?

IF that's what y'all meant, couldn't AI just make up for that since it's the one making all the code and cross-reference it's own code?

and IF NOT.. ?
>>
>>533410835
>kike gpt
kek
but python is ok for a beginning.
its slow as shit- its not designed for high performance applications
BUT its something you already know, its relatively easy, and you can already work on your algorythmics, aka- the skill of solving problems using programs.
you can also compile python to make it much faster and you could even use it to build games


once you get used to it then do C. its a must, everyone needs to learn at least SOME c. its gonna give you a high-level-of-abstraction understanding of how your computer works
and then yeah, your pick. c++, i woudlnt recommend it, but you could do rust (its a very rigid language, its hard to work with it. it provides certain safety guarantees though), or you stick with C, and then you could branch out into opencl, or cuda, but id say you need to program in c for at least a solid year before you take on gpu programming
because you will be doing c, but everything is semi- asynchronous now.

or you go with C#, which is as easy as python, and is actually something routinely used in games
dont quote me on that but im fairly certain unity uses c# for instance
>>
>>533411279
30 years lol. You are a peasant or a shill. Probably both.
>>
>>533411436
none of that and I'm not afaird of showing my flag unlike you faggot
>>
>company makes dashboard to show
how much Claude tokens people use
>CTO wants everyone tokenmaxing
>boss of my boss told us to deliver a Q of work in a sprint
My corpo is an AI cult at this point. And I swear to God we used to deliver a lot, lot more shit before 2025 when we simply used the damn auto complete and our brains instead of this gay ass "spec drive development" shite
Software became so boring. I should have studied civil engineering
>>
i wonder whats going on with open source ai after the electricRabbi's source code was leaked.
>>
File: byzantine qwen plea.png (84 KB, 774x934)
84 KB PNG
>>533394063
this is Qwen installed on my computer 12gb videocard, 32gb ram and 2018 HD. btw
>>
>>533411336
more or less, yes. close enough
only its more a question of legibility, and streamlining the program's procedures across multiple stages of work or multiple people working on a project bla bla bla
i would have to introduce the concept of interfaces, functions, variables, structures, actually quite some stuff to explain it properly to you
basically i would have to teach you the basics of programming

problem is that the ai is not a 100% solution, so given enough complexity, it will start making mistakes, especially when used to generate all the code

on the other hand when its presented with a codebase (body of code), it already has a "skeleton" to follow the logic and the intent of the programmers
so it should perform decently well at "cleaning up/streamlining" existing code
as in, fixing and and changing complex stuff into simpler forms in the slop humans created, kek
up to a certain limit of course, beyond a certain degree of complexity, the chatbot starts to hallucinate, stuff. more or less.

so yeah, working with a chatbot is really complementing the shortcomings of the ai by structuring the project, and auditing the code once its done
in case it hallucinated something, misunderstood an assignment or made a "grammatical" mistake
>>
>>533394063
It's giving goyim literacy when jews have intentionally obfuscated through through new forms of manufactured speech.
So it's too antisemitic to let goys use it
>>
>>533412244
its not helpful to goyim at all
goyim should be learning and developing their intellect and reasoning capabilities and engaged in the acquisition of new skills
AI is a non-starter for both



[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.