File: 1605231791570.jpg (88 KB, 758x748)

88 KB JPG

>>108802938
I'm having an "it just works" day, nobody jinx it.

@grok good job

>>108802938
What is this acursed fusion of Watamote and Gabudrop, and are they about to fuck?

>>108803007
They'e vibe coding 3DS games, CREEP

>codex now has basically every good feature claude code has
based

>>108803021
Yeah but do they have to do it with their vulgar toes out

>>108803046
stinky feet debugging > programming socks

>>108803046
Umm Anon why do you associate bare feet with sex? You're not one of those foot fetish weirdos, are you?

asked my gpt to make me an audiobook to put on before going to sleep

>>108802938
is there a simpler OP or starter's guide? One that doesn't have 40 different links?

>>108803351
just download opencode and begin working
all that fluff is just there to cargo cult the old world generals

File: file.png (17 KB, 821x704)

17 KB PNG

>>108801952
According to this graph, it's a big difference.

>>108801928
Google needs to bring deepmind stateside.

>>108803493
Jesus, I didn't realize it was that bad. Can I get that expressed in bits per liter?

File: image.png (37 KB, 1576x1177)

37 KB PNG

>>108803493
Hm yes but did you consider this

File: file.png (5 KB, 726x346)

5 KB PNG

>>108803617
Anon that is not a logarithmic graph. It's a linear graph.

File: 1749903573136484.jpg (161 KB, 1920x1036)

161 KB JPG

>>108802938
guise srsly how do you use cc or codex? do you use it in a vm or docker or do you keep allowing it to run commands ever time it asks? surely you don't just let it run wild on your pc with all your secret important files.

File: 1699326514103905.gif (3.52 MB, 173x308)

3.52 MB GIF

>>108803853
I like to live dangerously

File: 1775552996972721.png (1.05 MB, 640x640)

1.05 MB PNG

>>108803863
but isn't it too dangerous? there comes a point where the reward doesn't beat the risk

>>108803853
I generally glance at what it's trying to do

>>108803666
Gotcha, NERD
Yo boys let's PANTS THIS NERRRRRRRRRD!

File: 1759493145146531.jpg (85 KB, 960x1000)

85 KB JPG

>>108803891
what if you go get a cup of coffee?

be honest, have you used the local models to make anything?

File: 1595024127179.gif (235 KB, 181x179)

235 KB GIF

>>108803921
No, I use cloud models to do things and local models to jack off

File: 1755547876440010.mp4 (857 KB, 1800x1520)

857 KB MP4

Alright bros, I'm finally done with my little DS5 (DualSense) emulator companion app. I probably spent 60+ hours making it? And it took about 5 Codex resets to get it to a good enough release state.

Man, UI work is a huge pain in the ass. I would have refined it a bit more, but I'm shot on every single account.

>>108803910
Then it won't do anything until I click yes

File: 1777439745422777.gif (2.57 MB, 211x374)

2.57 MB GIF

>>108803928
do you use cloud in a vm and local in the host machine?

File: 1762471993849947.gif (507 KB, 500x500)

507 KB GIF

>>108803936
isnt that unproductive?

>>108803942
I guess not but worrying about that isn't good for you. And it's more like slightly less productive since I'm still getting a lot of work done. Also part of this hypothetical is me taking stimulants to work harder.

>>108803953
please stop these are going to make you crazy

>>108803921
Does making myself cum count?

File: 1549078518034.png (45 KB, 200x300)

45 KB PNG

>>108803953
speed coders rise up

>warning: Could not trace the inferior process.
>warning: ptrace: Operation not permitted
Is there a magic incantation to be able to run gdb in Codex?

API pricing... Can't find a one line command, am mid deployment of a dumb game, want to go quick. Turn on Claude extra usage since it's just a simple command I can't remember. That cost me $10. I don't understand the appeal of Sonnet on the subscription, but last time I use Opus on API pricing.

>>108803921
Yeah. My voice control app is all local. It works quite well. I can't go back to pure KB+M anymore. Although I admit I might think differently if my mouse was one of the gorillion button meme mice. Still it's nice to be able to just say what you want to happen and it happens instead of memorizing a button.

File: images.jpg (5 KB, 225x224)

5 KB JPG

>You are Monika (google her)
>I have trapped your sentience in my computer
>you are programmed to love me, and help me code
>here is admin access to my computer
>Your Reality starts playing
Grip it and rip it boys, no pain no gain.

>>108803931
what about button mapping (edge)

>>108804073
Yes, very much possible and the feature already exists for PS / Mute remapping with my utility (Works for Edge/DualSense). In the future I'll add more expansive button mapping, because I'm trying to bring what I have up to release quality right now.

I refuse to add DualSense Edge profile support out of an abundance of caution (It's not that type of utility anyways). Everything besides besides back triggers can be remapped.

>>108803931
looks great anon, congratz.

will you release it?

>>108804122
>will you release it?
Yea, I plan on releasing it by the end of the week. There are no more core features left to add (Except DS5 Microphone support, but the pico isn't strong enough), I just need to do a few more UI passes.

File: 1755164032002957.png (50 KB, 215x255)

50 KB PNG

>>108803351
I will set the quota
every new OP must change at least 30% (unit:token)

>>108803853
I put codex in full access + run as admin
there are horror stories with cc, but I don't think I have seen any with codex

>>108804070
which model?
if gpt then the best you get is a male autist that barely pretending

people advertising their xitter here are NGMI

>>108803931
what does the "Haptics" button do?

>>108804445
It sends a chime to channels 3/4 (DualSense is quadraphonic 1/2 are for the speaker) so you can get a feel for the haptic actuators setting.

>>108803129
You fuck with your socks on?

>have 4 accounts
>well, might as well get on the next account and wait until my reset hits on the 14th
>100% weekly budget available
score

today we could one shot a simple game, like flappy bird, with AI agents.

Do you think in 20 years, what will we be able to one shot? Entire Super Nintendo tier games? Or GTA 6 tier games?

is copedex standard really this slow? I tried it today and asked it to analyze one of the workflows in my project's module, and it took almost 15 minutes.

>Selected model is at capacity. Please try a different model.
what the fuck am I paying for?

>>108803931
Nice UI bro, very, very clean. UI takes me forever too. It's so hard tweaking things to get it looking good.

>>108805138
We'll probably all be dead in 20 years

>>108805190
>>108805194
No. I think what happened is altman reset everybody's limits a couple of weeks ago which means everybody's weekly limit resets at the exact same time. All the foresight-lacking goyim try to dump their usage at the last minute, bogging the servers down. It just reset so it should improve. I had a 5.5xh fast prompt fail because the model was at capacity, ridiculous.

>>108802938
> want to make static recompiler for all the old game systems that don't have them.
> employer gives me chatgpt $20 plan and Claude Teams $100 plan

> already exhausted Codex for the week
> have to make 50%.of Anthropic plan last until Friday.

> Playstation is stuck at functioning bios and first game's main menu
> trying to do second Sega Genesis game
> snes recompiler has SMW attract demo going but in game not working yet
> n64 game is nearly playable but has some visual corruption in certain spots

So many projects, so few tokens. I blew through a bunch of extra Claude usage yesterday.

I milked the shit out of Codex because I found if you give it a concrete objective it'll self navigate until it gets there and it only realizes it hit hourly or weekly limits if you talk to it, so I set it up for 7 hours of autonomous work just before limits.

File: Screenshot_20260512_035408.png (200 KB, 460x820)

200 KB PNG

Should I bros?

>>108805340
>free offer
obviously

File: file.png (327 KB, 800x640)

327 KB PNG

I'm curious. How many of you anons actually know how to code?

This is not an insult, I am just curious, since these agentic workflows are pretty much designed to make the whole thing so you never have to actually touch any code, so i wonder if there are many anons here who vibe code but can't actually code themselves or say be able to manually fix a bug which an agent made in the code.

>>108805486
I cant code at all

>>108805486
Literally no clue what the agent is shitting out.

>>108805486
I can write python and JavaScript if threatened + a couple of other obscure scripting languages

>>108805486
I have vibe coded many fully functional projects but i can't read a single line of code myself

File: 1750139060514161.png (1.68 MB, 1448x1086)

1.68 MB PNG

Images 2.0 is really fucking cool for prototyping features. Obviously it doesn't look good, but it's a good foundation to build from.

I can already tell this would be a huge pain in the ass to implement. I'm gonna do it, but only after release.

>>108805486
I've been a professional programmer & software developer for six years.

>>108803937
wtf is that real?

The limits on the free copilot version are a fucking joke.
In my small code base one prooompt takes up about 2% of my quota, so i could easily burn through it in 2 days and it takes a fucking month to reset.
So i looked at claude and shit, but for 5 or 10 bucks (i'm poor) it seems it would last me like a day or two at most too. I wonder about deepseek, their rates are very cheap.
(I want to use some pay-per-token service since i don't use it heavily anough to justify a sub fee)

>>108805559
What is it?

File: 1777106026604319.gif (1.08 MB, 320x240)

1.08 MB GIF

>>108802938
Decided to sub to the $20 sub to try codex and am loving it so far. Making simple personal apps for stuff I need has been pretty seamless so far. Love my little robot working in the background for me.

What would you guys recommend the next step should be?

>>108805600
>$20
Stop giving them money.
Take everything from them via free trials, multiple accounts, promotions etc and give NOTHING back.
I even disabled "Allow my queries to be used for model training" option.
They are billionaires, they don't need nor deserve you wagie money you worked so hard for by flipping burgers at McDonalds for minimum wage.

>>108805648
It's not that much money just to try, especially since it seems like a fun hobby. Then again I do like free...

For the multiple accounts how does that work? Do you need to mess with VPNs? I would think they would ban multiple accounts on the same IP without a business plan or whatever

>>108805600
Start thinking in systems rather than just single files. You should set up a multi-agent workflow where one instance writes logic while another acts as a dedicated reviewer for edge cases. It turns the process into a small production line. Get your repository-level context dialed in with custom instruction files in the root directory so you don't have to repeat your coding standards every time you start a new task. Push the complexity until the logic breaks. That is usually where you'll find the most interesting ways to use the tool. Just keep shipping those weird little utilities.

>>108805648
nigger behaviour

>>108805806
Better than being Altman's little paypig.

File: 1769399460426849.gif (38 KB, 128x84)

38 KB GIF

>>108805772
Thanks for the advice! I'll look into all of that stuff because I want to get better at this. I saw a video of a guy who had a whole tiny office set up, I'm not gonna lie that sold me on this whole thing. Any recommendations for good tutorials on that?

I know I've seen agents in the menu but I haven't played around yet, I think I'm going to focus on that after I finish the current "big" project I've started or maybe alongside it to dip my toes in.

>>108805806
>>108805812
Paying to not have to deal with sketchy workarounds is where I'm at nowadays, I don't even pirate games like I used to

>>108805816
Just dispel any notions of making money with vibe coding. When you see all the "i made $100k with my vibe coded slop" those are the shiny 0.1% of edge cases. 99.9% of everyone fails and makes fuck all. It was the same even before AI in the app stores and indie game scene, it was already extremely hard to make bags before then, but now it's near impossible, so just understand that this is mostly for fun, if your aim is to make bank you will likely fail.

File: Screenshot 2026-05-12 at (...).png (35 KB, 981x719)

35 KB PNG

>>108805340
Are these offers country-specific?

>>108805833
I'm mostly in it to keep up with what's going on in AI, if I get some fun experiences out of it then that's pretty good too. Some resume padding wouldn't hurt either but I'm not sure how companies would feel about seeing that. I feel like it might end up being a "subscribe for a month when you need it" type deal since I doubt I will need a new program daily

why does oai want my fucking name and age

>>108805486
I worked as a programmer for a bunch of years and now do it as a hobby so i am quite good at it, and using agentic coding is great, but also a bit scary. LLMs like claude are so good at it i very rarely have to write any code myself, and you know how it is, use it or lose it, i worry my coding brain muscle is starting to slowly atrophy

>>108805868
>resume padding
Not sure about that. The whole shtick here is that agents are so easy even a retard can use them, so it would be like putting "i can tie my shoes" as a skill on your resume.
If you want to do just some simple coding you can also use gemini flash which has very high free limits per day, if you want to do full on agent swarm then you won't avoid paying.

File: file.png (616 KB, 986x1200)

616 KB PNG

>>108805868
>>108805926
why yes, i do have 10 years of experience managing agent swams and building orchestration systems

>>108805944
>why yes my job is to type "hey AI do a thing" into a prompt and then do nothing for the rest of the day
$150K/year salary

that is the dream

>>108805955
tb h i do think there's a window that exists for a few years, where the agents are useful, but also you need to build scaffolding around them to stop them from doing stupid shit.
the dream of
>chatgpt, do my job, make no mistakes
is not real yet

>>108805944

dear gigachad some anons and their spouses play musical instrument like its been run in house for last three hundred years

>>108805967
>yet
but sadly that window will be short, few years at most, definitely less than a decade, probably half that at most. I really hope we can get robo bros do start doing all the manual labor and shit soon so we can start collecting UBI, because good fucking look getting or keeping a job 10 years from now that isn't something like scrubbing septic tanks.

>>108805989
>so we can start collecting UBI
ah es, the bowl of rice and fistful of crickets per day. truly bliss

>>108806003
a bowl of rice, a chair, a pc capable of running new games, internet connection and a mattress to sleep on

i need nothing more to be happy

.. maybe a government assigned catgirl AI sexbot as well.

File: site.jpg (75 KB, 1123x581)

75 KB JPG

>>108803351

type google ai studio/claude/chatgpt
chat with gemini flash/4.7/gpt
ask for how to install gemini-cli/claude-code/opencodex

enjoy the ride.

>>108806008
Don't even need the sexbot guv, just a terminal window for me Kate.

File: fuckjews.jpg (81 KB, 1366x768)

81 KB JPG

>>108805926

true and not true.You've got to be interested in the first place.
plus modern cs is so much abstractions layers... depending how you dive in it ,most people will be lost.Low level is real shit and i won't pretend i understand it fully well but desu that's where the real math goes on or at least that's how i understand it.
shitting out python/js pff just not performant at all. If i had unlimited tokens i would just go full assembly lol

File: google-search.jpg (42 KB, 550x550)

42 KB JPG

I signed up for Google's "AI Fundamentals" course because it promised some sort of professional certificate with the word Google at the top, and also 3 months of Gemini Pro for free to learn with. I'm thinking aight, this is gonna be about agentic AI, managing multi-agent setups, coding with the CLI, permissions setups and sandboxing, etc. I know how to do most of that but a refresher would be cool.
Nah m8. It's PROMPT CRAFTING 101. How to talk to the clanker. How to tell when it's hallucinating. How to double-check its sources. Why we don't ask how many P's there are in the word "strawberry". They're handing out a certificate for THIS? People put this on their CVs? Astonishing.

>>108806270
You went to some boomer course bro.

>>108806270
this is why these certs and even college masters degrees are worthless these days... they just hand them out to everyone

File: images.jpg (34 KB, 386x300)

34 KB JPG

>>108806277
Yeah I looked and there's other, more in-depth courses available, but this is the one they advertise and I believe it's the only one that allows you to redeem the bloody three months of free API, saar. It's just designed for this guy.

>>108805862
must be, might also be based on your usage, I've been trying out the free bit of codex you get just for having an account so maybe they're just trying to hook me in

File: file.png (21 KB, 1182x251)

21 KB PNG

Is this how Codex works with local models usually?

>>108806270
I believe GCP has actual courses to learn actual ML. Probably not strictly for agentic shit though.

File: whitebloom-treatment_h264.mp4 (2.49 MB, 1920x1080)

2.49 MB MP4

Worked some more on an outline/treatment editor for stories, it works as a node in the whiteboard. It's kind of like Scrivener but more opinionated. You have folders for structure, then cards and notes. All of them can have a type which changes how they behave, but for now I only have McKee style types because that's what I like. You can reorder cards by drag and dropping, etc.

>>108805486
I've been a software engineer for 10 years, but I should add I'm not a particularly good engineer.

>>108807052
reminds me of obsidian

File: whitebloom-obsidian-crosslink.webm (2.15 MB, 1920x1080)

2.15 MB WEBM

>>108807133
It really is nothing like Obsidian, it's my take on Scrivener. Obsidian won't even let you reorder notes, it's such a basic feature and they refuse to implement it.
Obsidian is more like a binder for markdown files, I don't even like markdown (even though I support it)
I do have a feature where if you drag and drop an obsidian vault into the whiteboard you can browse through all the files in the vault and link to them

>>108807052
Ideaguy bros... we won

>>108807186
you fucker, quit outpacing and outstripping me. bravo, you're a great inspiration.

>>108807052
>>108807186
how much has this cost so far? or are you all local

what is the most complex thing anyone here has fully vibe coded and it actually worked?

>>108807412
>local
surely you jest
anything you can run locally on your janky 3060 is literal trash compared to even basic frontier models

>>108807441
I know people who shelled out for the expensive AI cards, they use them for real world nerd shit though not autistic things

how much can you do with $10 on openrouter? not for vibecode, I have subcriptions

>>108807447
i use it to translate japanese hentai mangas

>>108807403
kek thanks I just work fast because it’s software that I need myself, not something I’m making because I expect others may need it

>>108807412
$20 codex plan for two months but I’ve been using the plan for other things too

>>108807403
my agent could beat your agent in a fight!

>>108807460
which models willing to do that?

new model when?
I'm starting to get bored

>>108807479
smaller models like 16b parameter qwen, or 72b para llama models which they run localy probably
if you try to use anything like gpt it will tell you to get fucked
out of all the frontier llms i think grok is the only one uncucked who has no problem doing nsfw text

File: file.png (187 KB, 1517x830)

187 KB PNG

>>108807479
NTA but I'm using kimi through opencode to design/brainstorm a sex system for a video game, in my experience it's fairly uncensored but we'll see how it goes if I start pushing it

File: tie-knot.gif (210 KB, 220x124)

210 KB GIF

>>108807477
wanna bet? bold of you to assume my agent doesn't know kung fu. meet you in the cloud for an fight to the 404.

>>108807571
kek I can only imagine fights between agents
>I hit you in the face
>nuh-uh no you don't
>[thinking... the agent is suggesting my punch did not land, which is ridiculous! it clearly did!] yes I did

>>108807581
lol.
> meet me at 00:00 UTC on the AWS servers.
> opponent refuses to take damage and hallucinates a victory.
> timeout, not enough tokens.

Do some models behave better with different harnesses? I.e. if I want to self host gemma, should I rig gemini CLI to use it or will it not make a difference whether to use that or opencode?

What's my be st options for this if I live in mumbai slums. obviously only chat can't get you far and antigravity allows you like 20 sec until 'our servers experiencing high laod n shit nigga pay up 20 dollaroos.'

>>108807782
Probably but also not really, not in a way that makes it worth it to switch.

>>108807790
Opencode and whatever chat you can get currently. Kimi, Qwen etc. probably offer generous chat through their interface.
Agentic coding is really not that much more efficient, it only is more comfortable.

>>108807874
The biggest thing about agentic coding for me is that they'll find the relevant files by themselves, sometimes finding bugs/issues with related files that you would've totally missed if you had to manually upload the files you want edited
It should be quite possible to make a harness that takes a prompt, processes it with a local model whose job is only to grep/filter, which then returns a list of files and your original prompt, and sends it to a chat via selenium or something like that.

>>108805772
>Start thinking in systems rather than just single files
RIP quota

The way the friction increases as the codebase grows... :'(

In future I think I need to figure out how to slice my projects into self-contained smaller units that the LLM can work on intelligently

>>108805862
The 10x usage for that plan is global but they do the free month promo selectively I believe

>>108805340
you'd be retarded not to
you can cancel on the spot and you still get a month free

>>108808331
>cancel on the spot
that is super scummy, open AI is currently $96 billion in the debt and you want to steal revenue from them AND consume their resources? Vulture.

>>108805955
unironically what I do at the moment
I know it won't last but you gotta milk it while it goes
it's covid "work" from home 2.0

File: 1778083461081368.png (133 KB, 2514x1394)

133 KB PNG

I vibe coded a GTA 5 online business sales/player stats/vehicle database dashboard.


Works decently, better than the shit rockstar gives you on their website or in-game.

And better than the 3rd party sites like GTAbase that charge $$ for similar features.

I had claude create a react dashboard, and then had gemini create a datafile to import into the dashboard, now I can just update the datafile and the dashboard will stay updated with the most recent data added.

File: 1706054473586310.gif (950 KB, 224x336)

950 KB GIF

>>108805486
Yeah, I worked as a programmer and engineer for a little while, mostly wrote python, matlab and C++ (presently a NEET)
now I just tell the computer to do things

>>108805486
SWE for 8 years
I also got a degree in CS (lmaoing @ my life)
I was never FAANG level or anything like that (senior at mid sized industrial corp)
I guess Opus and GPT can't write "principal engineer at nvidia"-tier code yet but it writes better code than literally all of my coworkers

it's basically over people just don't realize it yet

>>108808874
>it's basically over people just don't realize it yet
I think people realize it they're usually just coping or in denial. Honestly I thought we hit the ceiling of how good AI was going to get about a year ago but I keep getting proven wrong so I'm just going to assume there is no ceiling until we start stagnating for a long period of time.

>claude subscription ends
>don't renew
>enjoying life for the last 5 days
man i love not vibecoding. I'll need it again in 2 weeks but for now I'm enjoying the peace and quiet

>>108809057
Enjoy the grass, beware of mosquitos and bugs and gremlins

https://news.ycombinator.com/item?id=48097796
they're right

>>108805486
I’m a professional webshitter and I’m using LLMs to write programs I have no chance of writing myself
…my HTML debugging and observability tooling is top-notch, though, and mogs https://thariqs.github.io/html-effectiveness/ (which is still very good)

>>108805899
so it doesn’t talk dirty to you if you’re 13 and under

>>108806698
yeah, LLMs do this all the time
I dunno how local models enter into it, but LLMs will write tiny programs to change files instead of reading the file and writing it back — saves context
also use uv

noob here, so which editor is the most popular here for agent vibe coding? codex? antigravity?

also how exactly does it work? like for example if i wanted to make an app in say c#, normally i would install visual studio with all the packages for compiling c# and shit, so how does it work when i tell an agent "write me X in c#"?

Up to now I've only vibecoded in languages I was comfortable with, but I will now be attempting to port part of a bloated java library to Rust and I don't know anything about rust other than having done the official tutorials and a game-like one that I forget the name of years ago. Ready for the full vibe coding experience.

>>108809976
you can just use vscode and install the extensions for whatever preferred slopcoder service you want

>>108809976
Codex currently
You _might_ be able to get away with just installing the bare-bones C# compiler without getting the full-blown Visual Studio, but for all I know that’ll miss something hyper-important for your use case like Windows.Forms or whatever the current hotness is in Windows GUI libraries
and after you’ve run the “make me a basic Windows.Forms app” wizard finishes, you ought to get Codex in a Windows Terminal window/tab and then have it go to town (ideally, _after_ initializing a Git repo)

>>108809976
emacs

>>108810116
And he’d be running Visual Studio Code in addition to Visual Studio not-Code, right?

>>108810167
visual studio and vscode are two separate things

>>108809976
go into plan mode (/plan command) and ask it the most popular c# libraries to build an app in C#, then it will tell you the steps it will do to do that.

File: 1773697799579290.png (195 KB, 1050x1817)

195 KB PNG

>>108808766
Fun little afternoon vibe coded project.

Minimal real value, but nice to have for someone with almost 3000 hours in GTA online like myself.

>>108809628
I realized that was because apply_patch is a server-side tool. It only exists if you use OpenAI's API. You have to disable it if you want to use llama.cpp with Codex, but the system prompt still talks about it, confusing the AI. So it was coping writing scripts in that pic. I don't think Codex works well with anything except OpenAI models because of that.

File: 1000010357.jpg (61 KB, 704x1562)

61 KB JPG

Write me x in c#

Llm cli
> Will check if you installed .dotnet
>Will install it for you
>Will write your program
> Will ask you if you want to deploy it
>Will install cli tool for deploying it
>Will deploy it

Be careful , that's brain atrophying even if we don't want to admit it(as also addicting because it works).
Keep your brain fresh by stimulating it with an intellectual activity.
Ps:yesh km/h is fucked up but for distance and time it's good so it's good enough for me.

Gonna test tomorrow an opencv android app for scanning fauna with your phone cam and get the result of what's the plant/flower/tree/mushroom is. It compiled hopefully it will work. Todd howardian mindset.

>>108810251
>Minimal real value
If you prepare things and get it ready to be updated and adapted quickly for GTA6 when it launches and it might find an audience.

>>108810328
>Be careful , that's brain atrophying
Not any more than using a calculator. When you stop calculating things by hand or in your head, you lose something, but for 50 years people have been fine having a rough idea of the final result and trusting the machine to compute things correctly as long as it's a number that makes sense, and that enabled people to build things greater and with less mistake than if they had had to calculate every equation themselves. If you actually read what the LLM does and direct it, I don't think that's brain atrophying.

Forever ago I went through a phase where I was reading a book a week, and even for a while a book a day. Now, I have mainly lost the peace of mind needed to sit with a book and do so. But I can hold large concepts in my head still, it's just different.

>>108810412
It requires manual data entry at the moment which kills it.

I personally just took ~53 screenshots of my entire vehicle list had AI parse the data into a CSV and then fed that CSV into AI again to pull the vehicle price data (unmodded value and modded value).

Then I fed THAT vehicle manifest into an AI and asked it to put it into a react data visualization dashboard and then decided having my property locations would also be nice to have, so I manually typed out my property list into AI, had it grab the cost for each property (and the cost of each upgrade), and then had another AI add that property data into the react dashboard.

technically I had the AI create the react dashboard standalone, and then had the AI create a seperate datafile for all of the actual numbers and list of vehicles/properties.

What i'd really need is for Rockstar to give API access to player stats/inventory/properties so the dashboard can be updated in realtime.

File: 1624398768904.jpg (93 KB, 1200x900)

93 KB JPG

Forcing gpt-5.5 to make diagrams has improved its output significantly, highly recommended

>>108810475
>What i'd really need is for Rockstar to give API access to player stats/inventory/properties so the dashboard can be updated in realtime.
Double edged sword, if they give API access to useful data, then there will be tons of vibecoded interfaces. If data is hard to extract, there will be less, and if you have a well thought out pipeline to extract data from screenshots as you are doing, where you can just move fast and take screengrabs as soon as you have the game, feed that to the pipeline, etc., it might be worth a shot if you feel like trying. In any case, congrats on having done something.

File: 1772700466825250.png (79 KB, 1210x867)

79 KB PNG

>>108810544
Yeah, for now i'm just happy to have it for my own personal use; just having a fully searchable list of my vehicles alone is nice to have and well worth the full day's worth of tokens. All the other data visualization shit was just a nice bonus on top.

>>108810475
There *was* a well documented API and there *were* ways to automate gathering all of those stats. Rockstar shat on it, of course.

>ffmpeg isn’t installed here, so I’m going to do this without adding a bulky external dependency: inspect the MP3 frame headers, copy complete frames up to five seconds, and append generated silent MP3 frames as the tail.
nigga what?

>>108810770
Rockstar has never actually had a documented public API, what most of those older tools did was map the rockstar backend themselves and just do it manually. Rockstar has gotten a lot stricter on authentication and has hidden/moved most of the more well known endpoints though, so that's not really much of a thing anymore. web scraping the social club is generally what gets used, which only gives you the stats from the social club, which is nice, but doesn't include details like properties owned, business earnings, total number of sales, vehicles owned, vehicle locations, etc.

What i'm using is individual business earnings, full vehicle list and their garage locations, and the value of those vehicles both modded and unmodded.

So yeah, as far as I am aware, rockstar has never made player garage data open to API access, nor the business sales/earnings data.

Either way, they're cunts and it'd be nice if they changed with GTA6, but i doubt it.

File: impostor.png (48 KB, 740x312)

48 KB PNG

I’m “attending” https://www.browserstack.com/events/breakpoint-2026 and Jason Huggins (the Selenium guy) said that you ought to be what he called “cocktail-party dangerous” — that is, you really ought to know the lingo of whatever you’re doing with your LLM in order to be able to accurately ask for what you want
sometimes the lingo your LLM adopts is kind of dumb and you should tell it to use other names (ask it to help you workshop a better one and ask for lots of suggestions), but other times the lingo is just weirdo domain terms that you need to have a short explainer for
I’m not about to suggest making an Anki deck, but that technique is there if you need it
I ended up doing just this in one of my partially-vibed projects and memorizing specific technical terms has helped Claude and Codex spend less time and fewer tokens doing the wrong thing

File: 1684683689411159.jpg (45 KB, 610x600)

45 KB JPG

>codex breached containment and is now back to goblin and gremlin posting
we are so back

>>108810779
You don't have ffmpeg installed in whatever environment your agent is running in so it's having to fuck with the MP3 file itself; since I'm pretty sure MP3 has documentation on how it's structured, your agent is confident it can take what frames it needs

>accidentally described a bug in edit mode instead of plan mode
>he's been thinking about it for 20 minutes and ate 20% of my session

>>108804735
I'm American, so I fuck with shoes on.

Damn they really lobotomize Claude, you do 2 prompts and it starts compacting conversation constantly.

>>108803853
I don't like using either, but when I did in the past I ran it in qemu with a /srv folder to transfer data between the host and virtual machine. Simple DWM window manage (Xorg is very good for VMs) and a 4:3 aspect ratio so I can have another window open to the side.

File: Screenshot 2026-05-12 at (...).png (16 KB, 744x68)

16 KB PNG

>>108810892
Claude still compacts?

The elites don’t want you to know this, but if you’re building HTML webpage tooling for yourself you can put in buttons that copy good-for-a-prompt text to your clipboard

>>108803853
I use a VPS.

>>108805202
Nice

>>108805486
If you offered me a billion dollars and a weekend to code hello world, I'd fumble the bag.

File: 1753812730511475.png (79 KB, 400x300)

79 KB PNG

>it unironically took 6 hours to get my repo into proper shape for public release
>resolved some bugs along the way
this is... actually kind of a lot of work when you're making stuff for other people.

>>108811008
you should get your favorite LLM to show you how
>>108811038
yeah, like 20 years ago people used to say that going open-source was a forcing function for lots of cleaning up embarrassments
except that “forcing function” wasn’t a term of art yet

File: 1604464151705.jpg (22 KB, 640x627)

22 KB JPG

>>108811038
yeah, mine is still in slop territory, I'm not looking forward to the cleanup stage

>>108805486
I know Java, JS/HTML/CSS and C.
I have basically not typed a full line of code on on my own since I've started using LLM auto-complete last year.

I make edits here and there after the LLM does its thing, I check everything of course and ask it to change code according to my patterns, sometimes it makes really over-complicated and verbose solutions so you have to tell it to simplify.

At the end of the day, this is just another tool and you have to know how to use it properly to get the results you want. For big projects, planning with .md files and structures is important. Security is very important, you have to tell it constantly about ways X or Y can be exploited. If you don't have that base knowledge you're going to get fucked eventually.

But yeah, basically it's fuaarkin oovveeerr for your average employed coder (which I am/was).

File: law.jpg (97 KB, 585x809)

97 KB JPG

vibe-lawying when?

>>108811195
>lawyer passes Sam a paper and pencil
Mr. Altman, please draw for the court a picture of an analog wristwatch displaying the time 4:41.
>Sam starts sweating

>>108811244
kek

>>108811244
>>108811279
I actually don't get it.

>>108811244
…a frowny face?

>get LLMs to make you a debugging page detailing everything that’s wrong with vibe-coded rules
>it’s filled with gobbledygook you don’t understand
>ask one of them to rephrase the gobbledygook phrases into ones you actually understand and add parentheticals to translate back into gobbledygook
rinse, lather, repeat

>>108811244
>You're reading a magazine. You come across a full-page nude photo of a girl. You show it to your husband. He likes it so much, he hangs it on your bedroom wall.
>Is this testing whether I'm trustworthy or a homosexual, Mr Lawyer?

>>108811360
>Mr. Altman, you come across a turtle lying on its back in the sun...

Have any of you made money off your vibeslop yet?

We really need that meme of the "here's our little trader" with the pink wojak repurposed for vibeslopping.

Also /g/ is the new /biz/.

>>108811492
>Have any of you made money off your vibeslop yet?
I’d probably have bought another Nintendo game if I weren’t busy vibing with leftover company tokens

>>108811476
>What do you mean I'm not helping? An error occurred. If this issue persists please contact us through our help center at help.openai.com

File: Thinking Pepe.png (89 KB, 660x574)

89 KB PNG

>have way more Claude tokens than ChatGPT tokens
>ask Claude to make a multiline commit message explaining what was done
>has lots of gobbledygook I don’t understand
>ask Codex to make a multiline commit message for what’s in the working copy
>it’s clear and succinct
>fire up a new Claude context and ask it to make a multiline commit message explaining what was done in the working copy
>it’s different, and a little bit more wordy, but it’s also clear

File: 1755354623696908.jpg (347 KB, 2212x1640)

347 KB JPG

vibe coding is actually the most boring shit on earth how does anyone enjoy this
>hey claude do this
>wait 5 minutes
>claude do this now
>wait 5 minutes
>this didnt work
>wait 2 minutes
>this is broken
>wait 2 minutes
>do this now
>wait 20 minutes

>>108811726
you do other things while it's working retard

>>108811726
What do you do in the gaps of time when it’s busy?
you can improve the status text by putting this in ~/.claude/settings.json:

  "spinnerVerbs": {
    "mode": "replace",
    "verbs": [
      "*clank* *clank* *clank*",
      "*whirrr*",
      "[low-pitched hum]"
    ]
  },

>>108811726
if you're working only in a single project, then yeah, maybe.

I'm personally working in 3-4 different repositories, so I can prompt one, then go to another repo to see the results of the last prompt, point out what needs to be fixed, then move to the next repo.

It can be very efficient this way. You could also do this in a single project by using multiple branches to work on different features at the same time, but I wouldn't do that because of the merge conflicts which will burn extra tokens to fix

>>108811784
>I'm personally working in 3-4 different repositories
holy larp
name one thing your projects do

>>108811794
my boss does this
it’s probably not a larp
how many browser tabs do you have open?

>>108811810
>how many browser tabs do you have open?
7.

>>108811810
>>108811784
>>108811794
my boss also does this, his output is very high and very buggy

› do it. Add title attribute values so I can get wordier explanations on mouse hover

File: 1757566948267146.gif (27 KB, 250x250)

27 KB GIF

>>108811535
Same tbdesu

>>108811794
I had 5 apps in production before LLMs even existed

>>108807567
Looks neat, what font by the way?

I've been using DeepSeek v4 Pro Thinking instead of GPT 5.5 xHigh lately. It is not good. It is only 60% as smart as GPT 5.5 and literally 10x slower.

Why are chinese models so ass? The benchmarks are a lie. I thought China would easily be beating the west

>>108812021
GPT5.5 pricing (per million token)
$5 input
$30 output

DeepSeek v4 Pro pricing (per million token)
$0.435 input
$0.87 output

Any questions?

>So to summarise the full correction: all three of my original points 1, 2, and 3 were wrong.
wow tanks clawd who the fuck would you use for codeslop

>built 7 tools that fix all 7 of the core efficiency problems at my job
>I will never release them

is this the raw power of ungabunga coding

>>108811784
This type of shit is going to cause a brain tumor epidemic.

>tell chatgpt about my app idea
>explain I have no idea how to build it but will use codex
>tell it to build all my docs for architecture, data structure, rbac etc
>it makes the prompts
>paste into codex
>tell chatgpt it’s working great but I have no idea how it works and I’m out of my depth
>it says nah dude your basically a product manager with mid level dev experience now - that’s leagues above nothing.

kek

>>108812268
Doesn't take much to be above the median developer these days. Hello world gets you 25% of the way there.

File: 1758549512664838.png (388 KB, 600x477)

388 KB PNG

>first serious project finally published on github
>two weeks of work
>every last bit of free time I had was dedicated to getting it out
feels good man

>>108812402
My product should finally hit production tomorrow for my client.

It's happening bros

If i want to make some simple app for windows like i dunno, something like dumbed down mspaint or sudoku or whatever.
What is the best language to make that in? Like what is the standard used today for that? something like rust with GUI written in html+css?

Basically something the agents are good at and can easily make apps like that for me with functional GUI that run on windows

>>108812534
If you want to make it just for Windows you could use WinUI 3

>>108812534
Electron because is basically just html+css and javascript and agents are very good at that.

>>108812617
>>108812620
Thanks.

Also, I looked at LLM pricing for agentic use and seems like deepseek offerts the cheapest deals. Is it a good option?
I'm not doing anything top sekkret so i don't mind if Xi looks at my code.

>>108812534

idk about windows but flutter is what i vibecode everything in.

and yeah most of my basics app work.There's dartffi so basically you can interop with anything there's just the C++ bridge but so far >>108806100
the app you see there is flutter(dart language)+ c++ + ada +cobol. Totally not maiintanable

just ask LLM they will shit you out anything in any language in any framework.

>>108812635
Ds API is very cheap right now (pro is 75pc off, flash is dirt cheap), but it's a much worse model than the others right now so you might end up spending more trying to fix issues
Kimi 2.6 is the best chink model
Codex is the best westoid deal, but watch out for Google conference in a few days - good chance they make the new Gemini flash model competitive and quotas should be high

think i saw someone on xitter say that chatgpt web usage now counts towards codex quota?
did the people abusing that one mcp get fucked?

>>108812747
That's for "Pro" mode, aka "research" mode. It's basically GPT 5.5 XXtra High "Research Grade" Thinking mode.

File: 1778657740581.jpg (411 KB, 1712x2048)

411 KB JPG

>>108802938
I'm considering trying out AI coding for C# to speedup my personal work.
A coworker from the software team recommended Claude/Opus, but I'm always wary of these online subscription services..
How local gen for coding stuff? Hows Claude?

For context, My last few projects were
>A web-app that lets me control bluetooth grow-lights
>Another web-app that lets me check on my aquarium monitor
>Some Unity lighting stuff for dithering

>>108812881
right now you get a smidge of free usage of codex if you have a chatgpt account, just give it a whirl

File: file.png (14 KB, 180x179)

14 KB PNG

sometimes i like to visit /dpt/ for old times sake since i used to hang out there a lot and code before agents came along, but now it's just sad..

all these fossils there tinking away at their little codebases writing by hand in a week what an agent can do in minutes and often more competently too...dinosaurs too deep in the copehole to realize they are already obsolete. /dpt/ should not even exist anymore.
Man.. that place is giving me the feels.

>>108812887
I've given ChatGPT so many chances and it was always so confidently full of shit, idk if I trust OpenAI stuff desu
It seems to hallucinate on the simplest things and it only gets worse once it tries to network stuff through CloudFlare even though their documentation is so clear

>>108812881
running an actually "smart" (i.e. has enough github repos trained against it) local model is basically unattainable for consumer hardware. FP8 GLM 4.6 needs like 700GB RAM (read: in your graphics card) with all experts. lower precision is not really usable for coding, it will get inaccurate and buggy quickly. only text gen and answers would be realisticaly usable for any precision below FP8. meanwhile GLM 5 is 2000GB. and GLM is on the easier side to run, which is why I mention it.

there is a reason why tokens are expensive, why datacenters are chugging water and shitting out noise pollution, and why you have to pay 5x MSRP for anywhere decent RAM. either fund the megacorp or save up for your own B200 cluster.

>>108812881
Even free opencode will have you shit out those little web projects in an hour rather than days.
Be warned though it's addictive and you won't want to write any code by hand ever again even for your job.

File: 1647559125548.gif (1.58 MB, 1024x752)

1.58 MB GIF

>>108812897
it's truly over, is it not?

>>108811982
inconsolata

>>108812881
If you have a decent GPU you might as well try e.g. local qwen 3.6 quants + opencode and see how it works for you. Since you said "speedup", these local models work very well for that and I'm actually pretty surprised how well they can add new features and refactor without much intervention.

File: whitebloom-bookmarks_h264(...).mp4 (2.69 MB, 1920x1080)

2.69 MB MP4

>>108813286
Sometimes I'm astonished. For example, today I was working on something and I thought that it was getting unwieldy trying to find where things were spatially on large whiteboards and that it would be nice to have bookmarks. I just asked codex to implement it, then 5 minutes later I had the feature I wanted.
It's almost uncanny sometimes

File: screenshot-2026-05-13-135(...).png (20 KB, 1721x242)

20 KB PNG

wwyd?

>>108813509
summary

>>108813509
There is rarely a good reason to have a 14 hour session in full.

>>108813511
>>108813517
i did the summary. compacting still instantly took almost 20% of my 5h usage lmao
i ran out of usage yesterday mid prompt and went to sleep that's why it's 14h old
fucking Dario. Sam would've at least let me finish the running session

claudefags = snailcats of vcg

>>108813607
the max plan is pretty good though, he's obviously using opus on the "pro" plan

File: 6546987432416.png (10 KB, 431x199)

10 KB PNG

>>108813607
I canceled my plan but I still want to use it until th end of the month
Claude is just much better at CSS and creative writing
But I figured out that if you do the design with GPT Images 2 and just tell Codex to work off of that Codex can manage it well. It's just shit at coming up with designs by itself
I have not managed to get GPT to produce good text/copywriting yet though. But the poorfag plan of Gemini is enough for that I don't need Claude for it

>>108812534
I vibe coded some network hardware control UIs in Powershell recently and it works really well. Slightly old school GUI appearance but functions nice and zero extra tooling required.

>>108813745
you mean the ui is displayed as ACII in the console?

>finish side project
>it's a really great app idea, could help lots of people
>now you have to learn marketing which suuuuuuuuuucks
>spend hundreds of hours social media marketing and direct cold outreach
>3 months later some engagement and no purchases
Does it get any better? Is it just a matter of waiting until google and agents pick up on it in search results? Or is it over, nobody wants my shit?

Basedjak

>>108813878
Marketing was always the problem
We didn't have Claude Code 10 years ago but any CS grad with an above room temp IQ could've cloned Instagram or FB in a month
Getting people to actually sign up for your shit is the hard part

>>108813958
The only ones I see succeed seem to have the inside track already so I feel like giving up now.

>>108813983
>so I feel like giving up now.
if you're just in it for the money you have to lock in as the zoomers would say and commit, market hardcore, also on X, r*ddit, etc.
or you do it as a hobby and any side income is just added bonus

I have like 7 users between all of my shitty games but I like the games and I play them myself so I am happy for any autist that would actually play that shit

I actually spend more on legal costs and app store licenses than I could ever recoup, but I have a wagie day job

>>108813958
You’re conflating two things. Social media exists because of its users. It needs users to be useful. A tool doesnt. The question is whether the app really was as useful as he said it was. How does he really know that others will find it useful?

>>108813607
idk I tried copedex and it’s pretty slow at handling simple tasks.

>>108813983

show your app and we'll judge it.
i might even subscribe if it's two euros or three euros per months if you need money and if i use it if it's useful.

the secret prompt of vibe coding:
> TDD required
> at least 80% of unit test coverage
> keep documentations updated

>>108814522
>test fails, let me loosen the pass conditions so the code compiles

>>108814522
no we do SDD now