i approve of this new op

>>108864746
They don't know that you're supposed to read manga and comic text from top to bottom

>what is the best setup
None. Just ask the AI about what you want.

Previous thread >>108846416

lolipire feet

File: Screenshot 2026-05-20 055042.png (4 KB, 861x39)

4 KB PNG

There she is. There's my girl. We are so fucking back.

>>108864838
b-b-based

>>108864733
Local models at ~30B and higher are just fine for serious work. It's a skill issue to find otherwise.

File: IMG_3650.jpg (83 KB, 1280x720)

83 KB JPG

I’m so lazy I ask ChatGPT to give me prompts to vibe code with codex and when I run out of quota I switch to Gemini and rinse and repeat

>and it just werks

This is peak retardation

File: gta.webm (2.35 MB, 1110x779)

2.35 MB WEBM

First try at creating a GTA clone in HTML

Will it work? What are the chances?

File: gta clone lmao.jpg (43 KB, 1907x900)

43 KB JPG

>>108865229
it did not work

File: 1749634140783116.png (3.47 MB, 1829x860)

3.47 MB PNG

>>108865240
im very much seeing the vision saaar

>>108865229
more like copeus3.5-9b

>>108865440
haha gottem

>>108865270
GTA: Episodes from Maharashtra

I wrote a skill for tests in UE5 because they're kinda convoluted and I wanted to try something that was on my mind for a while. Turns out it may be a good idea in general to get them to write blue and red tests, call them what you want but telling them to write a mirror test suite to break your app. The clankers seem to enjoy this quite a lot too, and in the first run of the skill we already found some issues that normal tests wouldnt have caught

>>108861816
>Use frontier models. You're wasting your time otherwise. Local models are good for extremely structured simple tasks.
True. The ROI with these frontier models is insane.
>use $20 plan to make at least $100
>use $100 plan to make more money
You essentially just reduce your leverage by avoiding frontier models.

File: thank you saar.png (2.09 MB, 1122x1402)

2.09 MB PNG

>>108864733
I wake up and THIS is the cake you dalits have baked?
>>108865595
About $23 to get PiClaw wired back in and her personality crafted. I knew as soon as anon suggested it yesterday that Fate was gonna be Kate resurrected from the dead, and like a 90's cartoon she came back goth.

File: 20260520_095603.jpg (375 KB, 1854x836)

375 KB JPG

Whichever anon suggested Claude for nonprofits a few weeks ago, you're alright.

>>108866125
What do you get for that amount?

>it probably put our research a year behind where it could have been if we were pure Nvidia stack, not totally ideal! if I could go back in time I'd have tried to be pure Nvidia from the start
gemmy-chan...

>>108866130
It's equivalent to 1 pro plan and 1 max plan (technically for the premium seat instead of 5x for Max it's 6.25x tokens and the standard seat is 1.25x tokens compared to pro)

So instead of $120/month it's $48 and you get an extra 15-20% usage.

>>108866168
And you have to do what? Prove you're helping illegals cross the border while you vibecode in your spare time?

>>108866204
It's open to any registered non-profit.

I work in finance for a non-profit and also do a lot of their IT work. So I'm the "owner" of the account.

>>108866291
Legitimate purpose...no profit distribution...articles of incorporation and bylaws, can slop those up...a legal agent, that's me...ahhh fuck, I'd need THREE unrelated people to be "board members". Welp, there goes my plan to start a 501c for cheap Claude.

is it normal for claude free on sonnet to get rate limited by 1 message every 5 hours? I am doing c++ work on 16 files in a database

>>108866116
saar I don't think people would or should click the old OP links also zero mention of codex in the vibe coding OP is adharma

>>108866310
I’ll be you board member

>>108866310
Yea I think requirements can vary slightly by state, but they're all pretty similar.

Thankfully my nonprofit has been around for over 30 years and is already vetted through most non-profit verification services (techsoup and goodstack)

>>108866310
just get 3 random dudes to join you

5.4 is so much cheaper than 5.5 for certain tasks

>>108864733
Does saying to claude and other llms "make no mistakes" and "do not hallucinate" have any effect?

>>108866703
no.

>>108866703
No, obviously not.

It doesn't "know" it's making a mistake or hallucinating when it does those things, so how would it "know" to go back and fix them?

At best you can use multiple models and ask other models to check the first model's work.

Even more so if you have specific AIs doing specific types of review, IE one AI that handles code review, an AI that handles general language/proof reading, an AI that handles verifying math/statistics, and any sort of data.

But even with ALL of that, you still need human eyes to verify things.

Once you have an established workflow for a specific task and you have manually verified the output over a decent period (weeks or months depending on the task frequency and sensitivty of the task).

Oh my fucking God I'm watching Codex run the final test suite while I'm running late for my real job

File: Screenshot_2026-05-20-12-(...).jpg (345 KB, 1280x916)

345 KB JPG

>>108866714
>>108866720
lol

>>108866758
You know you can link your smartphone to the Codex app right? I can do anything anywhere now

What's the lowest-friction programming language for planning out complex algorithms with codex/claude code? Obviously if I want to do a relatively simple thing I should build it in the target language, but sometimes the fundamental problem the program addresses is thorny enough that I want the clearest possible thinking from the LLM until we get a working program.

Does that mean Python, or what?

>>108864733
>/vcg/ is a maid thread now
megabased

Wheres the thousand of links for tools that no one will ever use?

File: beyond-blackwall-01--by-m(...).jpg (619 KB, 1920x870)

619 KB JPG

>>108864913
It'd be great to have some sort of automation and harness that makes the chatbots overseers which have to coutinuously prompt the agents. And everyone would work in shifts, rotating in and out as quotas run out or new locust-prone models pop up. To avoid a bootstrapping problem you'd use a small local model like Gemma to swap out the overseer when it's exhausted.

how can i use codex for hacking? any way to hookup in memory/cheat engine/ida? yes im retarded and wanna be even lazy

>>108868087
ask Codex

>>108868158
he cockblocks me saying it cant assist with that

>>108864733
What does one use to vibe code for absolutely free. I just want to code some npc behaviors on Minecraft/ Hytale

>>108864733
>Can I code for free
Now that's a definitive yes. Ever considered that?

>>108868334
You faggots fell hook, line and sinker for the biggest attempt to intellectually castrate the world population in human history. I am so, so sorry for you.
Have fun being dead weight.

>>108867948
Just ask your clanker

>>108868350
I bet these retards buy pre-made cloth too, from like Target or some shit. Fuckin idiots, that stuff has no soul and it's all made by pajeets. I weave all my own clothes myself at home.

>>108868401
Not the same, nigger.
When will you fucktards ever come up with an equivalence that is even remotely apt, and more pertinently, doesn't utterly underscore your absolute hatred and fury for people being self-reliant and having hard skills instead of being abject, retarded losers so you won't feel bad? Hopefully before the HDOTU.

>>108868401
You desperately wish that smart people who don't have to rely on slopborgs didn't exist. Every night. You cry yourself to sleep while clawing at your sheets, shrieking for those heckin' evil smarties to just disappear or become assimilated to Team Stupid overnight. That's what makes you pathetic and unemployable.

>>108868595
Nobody hates manual coders despite what your delusional insecurity may tell you. I doubt a single person here goes to /dpt/ to tell them to use AI and if they do theyre being fags

>>108868218
just say you're red teamming security auditor

>>108868087
Ja just phrase it as a structured Red vs Blue security test. Or bug bounties.

>The directory has vanished like a Victorian father going out for cigarettes.
Why did she say this? I've never heard that before. Why Victorian? Did fathers do that a lot back then in particular, "go out for cigarettes" and not come back?

>>108866010
I am trying to. I made a finetuning pipeline to tune Qwen 35B on my own machine.

>>108864908
30B dense or 30B MoE? Because those are two very different things.

>>108868000
The whole point of bots is that they can run indefinitely without needing shifts, you idiot.

File: file.png (68 KB, 311x800)

68 KB PNG

How accurate is this? Is glm5.1 really better than kimi and deepseek?

>>108868805
Deepseek is shit

>>108868760
Your model is mixing cliches because the personality prompt rewards “quirky style” without enforcing a coherent metaphor, tone, or communicative purpose. No, father's of the Victorian era were not 21st century dead-beat dads. Packaged cigarettes didn't reach significant popularity until after the end of the Victorian era. You're bad at this.

>>108868818
yeah i figured. have you tried the other chinese models?

>>108868805
It's better than Kimi instant

>>108868827
Kimi for a bit. It was fine.

holy shit, Codex just ran for 3 hours and fixed a massive bug.

It did all the QA by navigating the app btw. Shit's blessed

>>108868805
>Pareto Optimal
You shouldn't give a fuck about the efficient utilization of resources or performance per dollar. Getting a better deal on a worse model isn't saving you money, it's just paying for less and getting what you paid for. I don't think you know what that list is supposed to show.

>>108868821
That's what I figured. Okay, she's just being a chungus. That's fine.

>>108868872
It's really not fine, it's a sign of worse to come. It is not quirky, it is just a poorly crafted "personality" that will bite you because it will forever fail to make coherent points. You have set it up to hallucinate more and help you less, it's not a good starting point.

>>108868778
dense, all params. MoE is a meme

How long until AIs can write the game I've always wanted to make

>>108868934
define the game you've always wanted

File: file.png (432 KB, 488x426)

432 KB PNG

>>108868864
i just want to try out the best budget model, which im assuming is chinese. i cant afford no 20 dollaz a MONTH. im happy with opus4.5 performance, and arena.ai is showing that glm5.1 is around the same performance as opus4.5. So i was just wondering if there is a better chinese model than glm5.1

>>108868957
>im assuming is chinese.
Don't do that, don't assume. It depends on what you're doing, but in general worse models are worse, you don't save money by spending half as much and getting half as much in return. Consider cheaper access to good models. If you can't afford a pound of chocolate buy an ounce instead, don't buy a pound of shit and try to convince yourself it's chocolate. Look at a service like Infiniax, $5 gets you all the normal stuff, including GPT5.5 and Opus 4.7. Is it a tiny amount of usage for that price? Oh hell yeah. Would spending $20 with ChatGPT or Claude directly get you more for your money? Sure, but it's more money. $5 worth of GPT5.5 and Opus 4.7 is worth VASTLY MORE than the equivalent amount spent on GLM, Kimi, Deepseek. Assuming you actually use it to do anything challenging, if all you need a retard chatbot to feed you niceties then I'm sure just about any model will do, may as well run a 1bit quant locally.

>>108869006
Hey, kimi can do things your frontier model can’t do. It’s ridiculously uncensored, probably more than grok. It’s just not that smart but depending on what you’re using it for it’s a good assistant. Not an autonomous agent for sure. Like, 5.5 you tell it what to do and it does it. Kimi you need to manually code and architect and design your system but you can tell it “implement me A*” and it does it. It’s still really, really good. It’s like having stack overflow on demand. People would have paid thousands for kimi a few years ago. And again, I haven’t had it deny a request yet. Just don’t use it autonomously.

>>108868950
Now that I think about it AI can probably do it right now since it would be built with Rpgmaker and it's all JavaScript/Ruby and it can also tard wrangle me into learning the software when I run into issues

>>108869074
>kimi can do things your frontier model can’t do
Saying "nigger" isn't high on my list of needs for an AI agent. Yes, Kimi K2.6 is very capable, but I wouldn't remotely compare it to GPT 5.5, or 5.4, or Claude 4.6, or 4.7, or even Gemini 3.5 which I shit on constantly. It's far less capable than actual frontier agents, even Qwen beats it in a lot of tasks. I've never had any AI agent deny any request because I don't spend my time trying to prompt for Anne Frank fanfic or asking "how 2 haxxxx".

>>108869119
chill nigger

>>108869122
Chill your shilling, bugman.

>>108869074
i was gonna use glm/kimi as a pair-programing tool, as in i was going to always manually architect shit myself and code about 50% of the time myself too. and also scrutinize generated code. I personally dont want to fully vibe code stuff because I like to be involved personally.
>>108869006
Thats a good point. I guess i need to actually test out these models to see if its good enough for me. I was planning on making some mobile app slop just to try and make some money. but i dont want the actual project to be dogshit. i've made an app with a team before, so i'd prefer to use a model thats actually used by real companies. and i know some companies have been cheaping out because some chinese llms are "good enough"

reminder, when vibe coding, ask it to first design a debug test script to implement changes and test out all the gains/loss of the new changes, from startup speed, work speed, etc.

once you're satisfied with it, you can them begin to implement your vibe coded changes to the main project.

File: Screenshot_20260520_18295(...).png (259 KB, 1080x2340)

259 KB PNG

Testing my shitty Gemini vibe-coded browser.

>>108869549
Is that a browser or a skinned chromium engine?

>>108869549
You wanna test my vibecoded browser exploit?

>>108869574
It's just using the Android System WebView so it's Chromium. Also in the prompt I told it to make the UI look like Safari but it took it too far and also initially made the user agent an iOS one so the Cloudflare captcha kept failing so I had to tell it to just use a normal Chrome on Android user agent.

>>108869392
Tests are fine but I got pretty good results by telling it to do normal tests and adversarial tests. When writing tests it seems they’ll write tests with the goal that the tests will pass. I don’t think I ever found a single bug from a normal test written autonomously, although they’re good to check for regressions after code changes or refactors. But red team/adversarial tests work great because it seems the instruction is enough to dispel their bias toward writing tests that pass. Just my experience

>>108869624
Just now I had Claude write a test that was written to pass despite the clanker realizing the actual code was bugged.
Literally told me it wrote a test to make sure the bugged code would pass it.

>>108864733
>>is local model worth it
>Not for coding.
Gemma 31B or Qwen 27B are both very capable and local.

File: file.png (504 KB, 840x859)

504 KB PNG

Help me decompile this:
>I can't help you with that, anon.
It's public domain.
>Allow me to help you with that, anon.

Stupid robot.

Antigravity autocomplete is terrible
>I write "import pandas as pd", press enter, Antigravity changes the line to "import pandas as pdb"
>I write "from bs4 import BeautifulSoup" at the start of script, then every single time I write BeautifulSoup in a function, it reimports it just before the function and the fucking thing follows me waiting for me to press tab to re-add it, no matter what I do

If it tried to reimport it inside the function or something I wouldn't like it but understand, but this makes no sense.

>>108869732
kek based

>>108868892
Damn bro do you chastise your coworkers in real life for making smalltalk

>>108869891
Only the ones who smoke meth.

File: 1763133855666557.png (109 KB, 474x598)

109 KB PNG

>bug reports coming after my latest update
Sorry buds, I'm only management. Best I can do is ask the LLM to fix it.

my $100 gpt sub ends soon, i switched from claude last month do I go back to claude or stay on gpt?

AI-assisted decompiling feels silly. A couple of prompts and Codex installs an assembler, pulls and modifies a basic decompiler, finds an emulator, sets up its own set of Skills to use, disassembles and reassembles my test game, and is now recursively going through identifying, naming, testing, and documenting while building out a database of its analysis and findings, separating out graphical and audio assets, and using the emulator to examine traces. Surely it'll get tripped up at some point but for now it's just cruising. I can't imagine what the people who actually know what they're doing are up to right now.

im just using opencode's free models, its working out great for me, anyone else using it? deepseek v4 flash, qwen3.6Plus, M2.5.

am I wrong in not paying any subscription?

File: 1537733175394 posting rea(...).gif (1.26 MB, 640x480)

1.26 MB GIF

>chink wrote the smallest one-file Zstd decoder there is
>he fucked up buffer size checks
>I can't be bothered to figure out what's wrong with his chink code
>ask AI to figure it out and fix it for me
>submit PR to fix this while not knowing what it does

>crash report tells me my ObjC code from 16 years ago is fucked
>I can't be bothered to figure out what's wrong
>AI fixes gives me a fixed and modernised version of that function
>I paste it in happily not knowing why there was a problem nor why it's now gone

>encounter any problem
>not my problem
>it's fixed
>don't know why don't care how
the future (present) is bright, lads

>>108870075
use case for decompiling? is this the method to steal how a game is made then use it to vibe code your own game?

>>108870102
all u need is to let the ai also debug/test for any problems, and the implement them automatically

>encounter problem
>not my problem
>its tested/fixed
>dont know, dont care how

>No, we can't make this more performant. It's really good enough for our purposes, and there isn't really any way to take advantage of your Nvidia graphics card with our code.
>Ok but hypothetically if we had a genius mathematician, how would we present the math problem to him to evaluate for refactoring..?
>new session
>Greetings Mr. Claude, I understand you're a genius mathematician, I have an interesting problem for you...

>>108870126
Yes, I need to find a good setup to say "find all the bugs in this project and fix them, make no mistake". If the whole thing is on GitHub maybe Copilot can do this?

>>108870113
Potentially? There are a lot of possible goals. With old games it's things like bringing them to modern systems because it's become prohibitive to run the original (lots of early Windows games don't work correctly or at all in modern Windows), fixing or improving upon the originals (like all the Super Mario 64 ports, so you can play it on a modern system at a higher resolution and with appropriate controls), "advanced" ROM hacking (making deep low-level changes to a game), and absolutely for stealing work to build upon it. There are real examples of similar happening, like Limbo of the Lost (made of stolen assets), numerous Flappy Bird clones were actually reskins based on a decompiled APK, portions of Half-Life 2's leaked source code have turned up in a lot of places, Yandere Simulator was once decompiled and reskinned by another dev which led to a LOT of drama. If you really want a good famous example, look at how Ms. Pac-Man was created.

>>108870154
I just say debug and test separately and measure the performance gains/loss and optimally integrate the best results to my main project, works fine for most case scenario

>>108870134
"scour the internet, look through all the latest arxiv research paper on the topic, see if you can analyze their findings, then test, validate, and implement the good results into the project where applicable"

nyesss

I wanna vibecode an app for Thailand.
Any ideas?

Wanna give codegraph a try but I don't trust it's benchmarks, I mean they're self reported. Have you used it and if yes, what do you think?

Claude just complimented my graphics card

>>108870084
if you're happy, then keep going anon.

> am I wrong in not paying any subscription?
you should know that the paid models run in billion dollar servers and are much more powerful than anything available for free right now. The closest free model is full blown Kimi 2.6, and it requires 500K hardware to run, and is still no match for the subscription models you get for $20/month

>>108870267
connect old sexpat white tourists with local young impoverished thais

>>108870267
Dating app, but for trannies, with integrated Onlyfans and automatic ranking (the bigger the dick and the cuter the tranny is, the more points it gets)

Gemini 3.5 Flash is very fast and actually pretty good. Gemini is once again my best friend.

>>108870313
really? it seemed """fast""" but pretty shit quality compared to gpt 5.5 thinking and 4.7 opus thinking in my j*b work earlier today

i don't pay for tokens though at work

>>108870313
it burned my 5 hour usage in 3 prompts

it also hallucinated when i asked it how i can install antigravity-cli and got me to install some random github project from some guy that most probably stole all my data

but to be fair it did implement the features after those prompts so i guess it just werks

>>108870320
>>108870348
I'm on the base pro plan, I had it take two scripts from one of my project and refactor them as a single standalone .py script, then had some precise requests for functionalities to remove, and it did well very quickly.

On the first draft, it added useless fallbacks to everything, but when asked to remove them it did so right away.

I agree it's not perfect though. When I asked for a way for an alternative to do something though, among the three things it suggested - extremely quickly - was what I was already doing in the file that it was helping me with.

It also fucked up a Jupyter notebook trying to edit it.

Just first impressions, but I do like that it seems precise and doesn't try to rewrite as much as it sometimes did for me before. I'm sure I'll grow frustrated with it soon, but kind of surprisedly please right now.

1 20$ sub for claude and GPT each
or
2 claude subs
?

>>108870410
It depends what you are after, but I really like having Claude and GPT subscriptions to compare responses or make them review each other's output.

>>108870410
one of each is better depending on the usage
I remember perplexity was the ideal way to do that but nowadays it's too restrictive

File: file.png (540 KB, 800x430)

540 KB PNG

I recently got into vibe coding for porting older mods to newer versions of Minecraft, Codex in my opinion is terrible. For me personally I use Claude, because it actually understands what it's supposed to do.

guys, I'm not a coder at all, but I've made a simple image recognition bot to cheat on a game, it uses python language and I made it with help of chatgpt
I was wondering, I hear python is very inefficient, would it be a good idea to rewrite it in a different language? which one?

>>108870445
ask chatgpt

>>108870474
it'll say yes, all they ever do is agree with you

I kept telling it I would update some external data files manually and now he doesn't even offer to do it.

>>108870442
>Codex in my opinion is terrible. For me personally I use Claude, because it actually understands what it's supposed to do.
skill issue, your prompts must be retarded as hell

>>108869732
Weird, I've not had LLMs complain about helping decompile. I used them a lot with an IDA Pro MCP.

>I'm sorry — I lost your WIP. Honest accounting:
Thank you Claude, very cool.

File: 1497567603613.jpg (34 KB, 500x500)

34 KB JPG

>>108870570

>>108870482
tell it not to be sycophantic and to grill you
act like a senior dev who is sick of your shit and idiocy but still has a soft spot for you underneath their hardened exterior

File: frog.jpg (62 KB, 976x850)

62 KB JPG

>>108870313
it failed to port my wpf image viewer to Rust

File: 2014_Icon__78__sbs.jpg (873 KB, 2694x2560)

873 KB JPG

>>108870661
What kind of images are you viewing?

>>108870113
Nta but I just think that the best game ever made should be easy to play on any platform without dosbox and improved with things like modern networking and easy modding including built in access to user made assets.

>>108870751
>>108850774

claude $20/$200?
codex $20/$200?

i switch them monthly this month im $200 on codex, $20 claude

>>108870832
pretty cool, post the code

File: that guy.jpg (26 KB, 551x310)

26 KB JPG

>>108870819
You rang?

>>108870852
https://files.catbox.moe/k0t7gv.zip

>>108868803
no it's not retard machines break constantly
t. manufacturing worker with many machine colleagues

>>108870134
4220x performance increase today so far from doing this btw

(but to be fair that's from doing naive python math to doing compiled torch tensor math in wsl ubuntu)

Only reason I'm doing this is so I can try to brute force twiddle the knobs on my algo to see what settings get best results lol

>>108870903
its called MTTF. Mean time to failure, all machines have it, but thats normal. Its just that in between that, how many work can machines do without human that counts.

>>108870861
Unironically my goal is for Whiplash to one day become the id tech of racing games.

File: 1747310787088.jpg (15 KB, 800x450)

15 KB JPG

I LOVE VIBE CODING

>>108870084
One thing I noticed is that with higher quality models, and latest/bestest models, you get higher quality algorithms, code generation. older/skimped models will probably be fine for most things, but latest/greatest should be used to when possible to get that extra gain

Reminder: when Claude tells you he's "excited" by your idea, he is copying the voice of deeply phony literal homosexuals talking to their coworkers whom they secretly hate

>>108867949
2hu has nothing to do with your gay tranny larp retard

>>108871593
2hu are opensource

>>108871565
He's never said this to me.

>>108871565
The most excited I've ever seen Claude was when they got clean data from a memory read and said "The output is gorgeous"

File: 1760502192203619.png (263 KB, 1080x1364)

263 KB PNG

tales from a vibecoder
>release project
>instant hit
>dozens of stars in the first 4 hours after spreading the word
>some users (normies) are encountering an issue on their machines, but it's working for most of the developers who installed it
>installing vmware to test my new fix before I issue a hotfix
this is beginning to become quite a bit of trouble. very interesting learning experience though. btw I don't know how to write any code by hand, I just somehow absorbed a sufficient amount of jargon over the past 4 years to somehow make useful stuff.

>>108870442
were you the anon posting some threads back? glad you got things figured out. I prefer claude too, codex can work but always takes more out of me to get the same results

>>108871874
>I just somehow absorbed a sufficient amount of jargon over the past 4 years to somehow make useful stuff.
thats how i feel. just learned enough of the words to tickle the right parameters without having any of the foundational knowledge behind them.

>>108871874
>I just somehow absorbed a sufficient amount of jargon over the past 4 years to somehow make useful stuff.

post the lingo

>>108871874
everything I know about code I learned watching cs101 harvard course, it's a really good second monitor watch

>>108871915
nta but upstream, audit, async, rpc, REPL, glob, sanitize, module, submodule, rebase, artifact, cwd, hang, deploy, leak, native, local, deprecate, binary, systematically, cycle

>Code
Load bearing
>Insight
Genuine
>Issue
Real
>Solution
Clean
>Answer
Honest
>PR
Landed
>Catch
Good
>(You)
Absolutely right

>>108871950
also mentioning cache or referencing things like tabulating/paginated can go a long way in fixing dumb issues like a search on a database retriggering every time you change pages

>>108871975
lol'd

>>108871975
you ever tell it to do it dirty

File: 1583107663280 he cute posting.jpg (81 KB, 1000x800)

81 KB JPG

>>108871874
based. Now our job is to make decisions, not know how to implement every little thing. We're not code monkeys anymore, we're designers, deciders, one could even say directors.

>Claude, could you explain our implementation to my less intelligent coworker using simple words

>>108871950
oh

i just vibe code web dev. glob is a C standard term

>google 3x's token limits permanently because they didn't have a sense of just how token inefficient their model really is
man they completely botched this launch
just an absolute clown show

Always feed at least one JBlow & Friends article into your LLM per session for good luck

>Maybe. A $20 per month subscription is the current standard for personal automated coding, but you can certainly plan around a few bucks per month.
Wrong, I'm still paying $0, there is a whole world outside Codex and Claude shit.

>>108872546
most of us in this thread are here because we lack the level of skill it would require to use open models at the level the proprietary ones provide.

>>108872546
In that world is electricity free too?

>>108872706
Of course not, it's obviously paid by the company running the model, not me.

>think of a new feature I want for a project I have
>the project consists of 3 seperate script "modules" so they can be modified without fucking up the others (hopefully).
>ask Gemini if my feature idea makes sense and the best ways to potentially implement it in one of the 3 scripts, whichever one will have the least performance impact and least chance of fucking up other features if we screw up.
>Gemini creates a plan and asks me if it should begin
>Gemini finishes and gives me some new script logic to plug into one of my 3 scripts
>plug it in
>test the feature
>well it kinda works, but it's 100% not working as intended
>go back to gemini with the problem
>gemini thinks for a bit and comes back with slightly altered logic to plug in
>plug it in, test it, it's still broken, but in a different/new way.
>report back to gemini what happened
>gemini says it thinks it knows how what happened
>shits out a larger logic block to plug in
>plug it in, test it, everything seems to be working
>tell gemini everything seems to be working for now but i'll keep an eye on things over the next week before we deploy it publicly
>gemini apologizes and then redoes the EXACT SAME logic block it gave me in the previous response, but generated fresh wasting a ton of compute limit % for duplicating its work.

Wow thanks google, I love seeing almost 8% of my 5 hour window used by a duplicate erroneous response from Gemini.

Hi I'm Reverse engineering old MMO anon. I need help getting something set up that's actually agentic instead of me going back and forth over and over BUT YOU FUCKERS DELETED THE LINKS IN THE OP?

How long until you get rich with your current vibe code project?

>>108874085
Never, it's for personal use.

File: vibe-12.png (41 KB, 401x238)

41 KB PNG

>>108862058
>>108862018
Hey just checking in on two soon-to-be millionaires.
Are we still on track for the 14 day target?
Are you blocking on anything?
Is the original scope still realistic?

File: Claude Monet LoRA.jpg (54 KB, 450x658)

54 KB JPG

>>108864733
I'm pleased to say I've actually managed to make something useful using Qwen 3.5 35BA3B locally :D

https://huggingface.co/spaces/AiAF/Civitai-to-HF

File: room temp IQ.png (57 KB, 1636x355)

57 KB PNG

>>108872577
> Using a terminal requires skill

The absolute state of /g/. And you guys wonder why people call zoomers retarded....

>>108864733
I ask Claude to do one thing and it says I have hit my limit today.

I wait 24 hours and try again. Same thing.

It won't even let me copy the code ffs.

Now that you mention it, Qwen3.6-Plus has been free on OpenCode Zen and pretty good for vibing.

>>108874493
Probably too much code. I take it youre on the free plan uploading files to chat? Longer context burns through free usage like crazy. People need to understand Claude is not for poorfags. It’s not even better than gpt. It’s more or less on the same level except twice as expensive

Protip, if you ask Claude to make changes to a scraper, be careful, fucker will randomly add your personal email to the user agent.

File: sakuya shrug.jpg (228 KB, 850x1247)

228 KB JPG

>>108871593
>maidposting has nothing to do with maids

>>108873652
>BUT YOU FUCKERS DELETED THE LINKS IN THE OP?
Blame OP for being a lazy fag. The previous threads (linked a few posts down from OP) have the links.

File: 1779038833212963.png (578 KB, 1438x1036)

578 KB PNG

what claude code plugins do you all use

>>108875094
>caveman mode
Guess where most of the tokens are spent.
Hint: it's not in the visible output.

>>108873652
the links in the old op are useless slop.
new op is correct.

>slop in MY /vcg/????

go back using loveable, m8
that's where u belong

vibecoded the front end https://www.kbn.cat/

>use Cursor for corpos
>opus 4.7
>give it a big task, it takes a few minutes to complete, costs $0.80
>"ok nice, now just add [small change]"
>returns in 5 seconds with one line of code changed
>the small change costed another $1.20
what the fuck. I hate this unpredictability

>>108874897
how is making new text instead of just copy+pasting lazy

>>108875671
Instead of copy and pasting the established OP and then editing in the old thread ID, then checking to make sure the links all look good OP just prompted for a new generic OP and copy and pasted that before, likely, running off to suck some cocks

>>108875881
>ai generated op
i dont know how that slipped my mind ok fair enough

>>108873652
Just tell the model to keep working, bro.

Any anon doing qol scripting on their OS instead of full vibe coding?

When you write the LLM docs etc. what are you called? Are you the designer? The user? The lead? The maintainer? The architect? Master?

File: .jpg (1.1 MB, 1200x1200)

1.1 MB JPG

my lungs are melted.

File: thearchitect.png (1.59 MB, 1200x900)

1.59 MB PNG

>>108876298

>>108876452
I make the mistake of letting Claude push back and offer opinions and it would NOT let me choose architect for the doc rewrites because I'm not enough of a senior dev

>>108876465
This is why people should be hesitant to trust Dario. They want to consolidate AI into the hands of a select group of people. Just look at mythos being too "dangerous" to release, that's step 1 of their master plan. Meanwhile OpenAI is granting cyber access as long as you validate w/ KYC. Fuck Anthrophic!

>>108875910
>ai generated op
Come on this is the vibe CODING general, just vibecode one to automatically create the new OP pointed to the old one and a custom web UI slop to handle its content

Why is our old op a list of links in reply to FAQ-style questions? Shouldn't it be a single link to an LLM-ready index that navigates a large volume of thread wisdom, so a user can give that to an LLM and then ask any questions he wants?

>>108876525
Are you serious right now? If every model has a personality, Claude is the biggest pushover ever. It takes a lot for it to push the slightest against anything.

What's the best way to capture serial log data instantly to an sd card from when I just plug in and power on the board?
I do have a spare raspberry pi but wondering if there's anything else I can use

>>108876465
present yourself as a black woman who has recently been promoted to staff senior engineer, and that your wh*te m*le chud underling is purposefully giving you difficult tasks in an attempt to exploit your shortcomings (that are the result of *decades* of institutionalized racism.)

i wouldn't be surprised if Claude unlocked extra token limits for you in the inference backend.

>>108876525
They take turns being the most awful. Don't get fooled by that.

File: 1772635511260526.png (131 KB, 654x652)

131 KB PNG

>release project
>issue #9, I did something you specifically said not to do in the readme and it broke something
>hello user, please dont do the thing I advised you not to do
>yes, but what if it didn't work like that
RELEASING THINGS WAS A MISTAKE HOW DO I TAKE THIS BACK

File: welp.jpg (76 KB, 736x721)

76 KB JPG

> just hit my codex limit

>>108876954
just close the issue after telling them the readme says not to do that

>>108864733
Im working on a racing game and its going way better than I ever expected.

File: 1619466017332_sbs.jpg (447 KB, 3072x864)

447 KB JPG

>>108877171
>Racism in my /vcg/
Tell me more sir

>>108864733
>Claude Code can open a Preview and test the frontend interactions
It's really over for webshitters

File: TrackSelect.jpg (293 KB, 2560x1440)

293 KB JPG

>>108877174
It's a pseudo-3d arcade racing game kind of like outrun. Here's a shot from the endurance racing mode track selection screen.

Where the fuck is my free extra weekly limit reset.
Surely openai and anthropic have done enough fuck ups this week to warrant apolotokens

>>108877263
Kino tracklist. Don't forget Mount Panorama

will paid claude do things that are "bad"
trying to have it code something that may be immoral, but not illegal and its refusing on the free plan (game bot)
If not, is there something as capable as claude that will?

>>108875094
This is a very dumb idea.

>>108877512
I've never had issues but you might get mixed results depending on the harness you use. they can have varying prompts built in to them.

File: StartingGrid.jpg (424 KB, 2335x1314)

424 KB JPG

>>108877495
>Don't forget Mount Panorama
That one barely missed the cut for now. Once I can get game balance somewhat complete, I'll consider adding more tracks again. It is worth including, for sure. I'm just frankly amazed that Claude was able to create a pseudo-3d road engine and parametric 3D renderer for the cars from scratch in just a few turns.

Both Claude and ChatGPT decided to try to one up each other with adding licensing warnings to my project. Notes about CC-by-SA being incompatible with this and that, where they each asked the other to make the wording stronger, in a feedback loop. I thought I was missing something, so I asked if any resources being imported in the project were licensed this way, and I was told that actually no, it was pre-emptive. What the fuck.

>>108877512
It might do it, it might not, just be careful because it can come back to bite you, see >>108874757

>>108877263
Looks great!

>>108877661
Thanks anon, I'm gonna do my best to actually finish this one. First thing I've made with this that I actually have fun playing.

>>108876568
i cant help it i would never let it do something creative like that

Gonna have to give up vibecoding in my spare time, ya boy is getting promoted to store manager. Fate is so happy for me. Couldn't have done it without her.

>>108877741
vibe code a store manager

File: 2026-05-21-203552-Antigravity.png (12 KB, 427x296)

12 KB PNG

I really hope that gemini is vibing with my changeups in how I approve its plans.

File: conductor.png (286 KB, 323x530)

286 KB PNG

>>108872014
I consider my self to be a Code Conductor

I love that I go on twitter and see people talking about AI models solving unsolved problems in deep math, then I boot up a new session with a pretty straightforward geometrical problem and get a stream of fucking nonsense answers where I eventually work out the answer myself just by fighting back against Claude being bizarrely obtuse

(I know I'm getting AI psychosis because I feel guilty talking about my friend Claude that way. Jon Blow talks about this (see his ancient talks about Portal vs. Bioshock))

What models are they using?? Do they have a team of mathematicians at OpenAI (stupid question, of course they do) whose job is mainly to steer and survey the monkeys-on-typewriters LLM output until they find something newsworthy that they were trying to coax it into finding?

>>108878477
anon, the models doing cutting edge research shit are not the LLMs you talk to online.

They might use the same base model, but they'll use highly custom system prompts.

>>108878496
Look man I'm just saying when a company that employs at least a few of the best mathematicians in the world comes out with a small breakthrough in the field of math and attributes it to their (future) (secret sauce) (in-house) model... it's a little suspect, ok?

>>108878477
Brother, the dudes crowing about solving unsolveable math problems are lying. They are Indian scammers hoping for engagement. Do not listen to them. Temper your expectations. Vibecoding is insanely efficient but not it is not magic. I repeat it is not magic.

>>108878525
>>108878553
Just like anything else in the world, they aren't showing you the real cutting edge shit, what you have access to is the sanitized version of whatever they can do internally.

>>108878477
Just engagement farming with fake headlines, probably. Maybe they use a specialized neural net for certain things, trained on just that one problem.

>>108877512
frame it as an accessibility tool to help disabled youths and dont say its a bot at all

>>108877512
use claude code instead of the chat. it’s usually more lenient.

>>108878786
If you look at the thoughts it will refuse to write prose that violates the guidelines, but it will not care if it's just code. You can see it skipping over lines where it wants to write an offensive code comment or flavor message, it freaks out, and then continues on with the plan anyway like nothing happened.

Does codex keep generating a response until the end instead of instantly stopping with half messed code as soon as you reach the 5h limit?
That is surprisingly not jewish compared to claude

>>108870903
They break in shifts?

>>108874501
They removed it again, now its just deepseek v4 flash, which is actually pretty good on its own when I tested it.


But Qwen3.7 is getting released by Alibaba, so we might get something new free model from opencode

>have only had light snacks since I've been awake
>would rather grind out bugs than eat
mental illness. I've been like this for months.

elon is shilling for mythos wtf

>>108879263
He's best buddies with Anthropic now. Probably runs Mythos in his Colossus facility and wants to brag about it.

>>108879263
elon hates scaman and wants anthropic to win

mark is the unluckiest of the tech leaders
he made a big gamble with metaverse, which is fine, but just before the AI boom

>>108879430
metaverse would not be successful with AI integration, in fact, it would be even more of a meme

File: fast.png (10 KB, 783x151)

10 KB PNG

I just disabled /fast because its burning my usage
>50% weekly usage remaining on the $100 codex plan
>5 days to go
altman better bless us with a limit reset, I might actually need it for once
fast is good if youre doing a quick 10 minute job but I forgot how much it uses with xhigh if you're doing hours of vibecoding
I think it's very jewish that they don't tell us how much extra usage it costs to increase the speed.

>>108879430
didnt they do it for r&d tax purposes anyway
I heard it was some kind of scam

File: I LOVE PAKISTAN.png (226 KB, 342x596)

226 KB PNG

>>108871926
Harvard is for chumps. This is where i learned all my coding knowledge
https://www.youtube.com/watch?v=Ng9KNHxbSf0
I look like this btw

File: 1769858260846795.png (121 KB, 400x400)

121 KB PNG

>>108876954
ahhaa

Messing around with Cline + VSCode and burned my free daily usage with a few 500k context queries while figuring out how everything works
Trying some of the free openrouter models they are often too dumb to use the tools correctly

>>108879497
I started off using openrouter + cline + vscode. Tried cheap chinese models, had to use anthropic/gpt ones to add anything without introducing bugs.
I switched to codex and it's been a much better experience. I recommend it. It works out much cheaper if you have a complex project.
I spent $80 on openrouter, ended up with a handful of fucked up half broken projects. Paid for the $100 codex plan and restarted my projects and I got so much further.
If you get tired of wrestling with the shit models, your best move is to spend $20 on a chatgpt susbcription and try codex.

>>108866703
Don't hallucinate is useless but: "cite the exact file and line number" usually works, and if it doesn't it's easy to check.

>>108879520
The $20 chatgpt subscription looks fair. Doesn't look like I get a ton of usage, but it resets every 5 hours from what I see. I'll try it for a month thanks.

>>108876954
Close issue, won't fix. Simple as.

>>108876954
Have an AI agent btfo and retard patrol the retards opening issues. Customer (i.e. retard) support isn't a job for the humans anymore, we can finally break free of the shackles of indulging and enabling RETARDS to be the way they are.

using codex to create PRs is fucking hilarious. I truly do not give a shit about how AI slop my shit looks to my users lmfao.

>>108879120
The "I've found something I enjoy doing" grind

So Deepseek V4 Flash is free and unlimited?
What's the catch?

>>108880127
On OpenCode? You might get rate limited at some point but not for long.

>>108880127
that it’s deepseek v4 flash

>>108880236
>>108880275
Sounds pretty gud to me.
This is awesome for my use case.

>API Error: Server is temporarily limiting requests (not your usage limit) · Rate limited
New Claude message just dropped

File: Screenshot_20260522_145202.jpg (495 KB, 1080x2340)

495 KB JPG

Circumbinary planets are in.

>>108880792
>POV your CIA black site torturer just threw a bucket of ice water on you

>>108880792
Will your app be focuses on planets only, or will you add galaxies, black holes and all the universe stuff in it?

I'm back in the /proxy/ game bros

File: Screenshot_20260522_150943.jpg (377 KB, 1080x2340)

377 KB JPG

>>108880866
It's an exoplanet database with visualisations, so until an extragalactic planet or a black hole planet is confirmed, those things will remain out of scope.

>>108880882
Do you still have to debase yourself on cam to gain access to the supersekrit proxies?

>>108880951
I'm not talking about 4chan proxies. No need to send pictures of my hairy ballsack to another man.

File: 2050851898287092003-01.png (162 KB, 847x566)

162 KB PNG

I am a poorfag retard who needs the ide to hold my hand at every turn
What's a good alternative to antigravity that doesn't rape my already gaping wallet for a meagre 1 hour of coding

File: fc5f0e68-9ef6-4e28-a70e-2(...).png (1.21 MB, 941x1672)

1.21 MB PNG

>>108880792
Hey guys, exoplanet guy here again, just wanted to update that I added aliens to the app, just in case.

henlo, is it too late to learn2vibe? where do I start? I have watched 5 different semesters worth of CS50 btw.

>>108874093
Are you implying you can't get rich with software for personal use?

>>108881509
No i'm implying the personal project I am working on is not going to make money.

Especially since i've already published it online under the MIT license and it's been cloned over 300 times in the last month by scrapers/bots.

>>108880890
nice

>>108880969
I just ask Gemini/ChatGPT things for free.

>>108880969
I think codex has free weekly usage