I've created a C++ app by copy-pasting source codes back and forth via aistudio.google.com so why should I bother with stuff like Claude Code?

>>108723192
it do the copy pasting for you

Codex can now gen images, meanwhile claude can't even run without an outage

I was using IFLOW cli, the chinese free agentic coding but it went down a couple of weeks ago. Is there another option like it?

>>108723197
But it has no free tier if I'm not mistaken. I did the stuff free with aistudio.

>>108723649
Claude will grep for you, run your program and read its stdout, read your database, drive your browser with Playwright MCP and so on.

I've started giving codex screenshots of ui with ms paint drawing over it, game changer. Now I dont have to explain so much. I can just draw an arrow to say where I want something re-positioned or draw a circle around something I dont like.
>>108723192
you wont appreciate how much quicker and easier it is until you experience it for yourself
I was copying and pasting code from perplexity and now I use the codex gui to do everything for me. I discuss an idea, tell it to implement, build, launch, and commit changes to github, and when the program launches i know it's been modified and rebuilt and it's time for me to stop shitposting and check the changes so I can give feedback.

>>108723837
>I've started giving codex screenshots of ui with ms paint drawing over it, game changer. Now I dont have to explain so much. I can just draw an arrow to say where I want something re-positioned or draw a circle around something I dont like.
Wtf, I have a trademark on this.

>@cerebras made GPT‑5.3‑Codex‑Spark so fast that we had to rethink how Codex uses the Responses API – leading us to build WebSocket support for ultra fast latency.

File: 1769690115496147.gif (292 KB, 660x780)

292 KB GIF

>>108723888
Digits confirm, but I stole it anyway
What other tricks are you hiding from us, hmm?

File: 1755877535188082.gif (2.07 MB, 415x498)

2.07 MB GIF

Glad to see AI is still unusable for anything practical
>Trying to create an excel plugin for my job
>ask chatgpt
>it treats me like an indian with the stupid fucking emojis and " CRITICAL STEP (Most People Miss This)" and " Reality Check"
>everything it says is wrong
>compile errors

>go to claude instead
>it compiles now
>still doesnt work correctly
>out of tokens, please wait 8 hours

>>108723923
i haven't used spark but i believe it's significantly worse than other models? saw some speculation that it may be related to the faster inference itself.

also this is a duplicate thread:
>>108720160
>>108720160
>>108720160
>>108720160

retard op from the other one forgot to include subject
move there for now, we'll recycle this later

>>108724105
Using it wrong unironically

>>108724247
how do you know?

>>108724318
Because it doesn't happen to me

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

1.29 MB JPG

>>108724105
Cope. You will be replaced

>>108723192
you're like 20X slower than a Codex user. Also, Codex has the context of the whole project, while copy and pasting only gives the AI context of whatever you pasted there

>>108724329
Because you're not doing anything.

>>108724247
Yeah, this. Probably used some subpar model, the web interface instead of a coding harness, vague prompts that try to do everything at once, etc.

>codex adds /side
>it's based and lets me ask questions and throw the context away without wasting or poisoning it
>ctrl+d doesn't go back to the original prompt but closes codex entirely
reeeee

>>108724105
>using the shitty normalfag web version instead of codex
>using comptuecucked misanthropicslop
skill issue

File: The Local LLM Cheat Sheet(...).jpg (250 KB, 948x1185)

250 KB JPG

>>108723180
Found this "LLM Cheat Sheet" on X comparing different local LLM's strengths and use cases. How accurate would you say it is?


https://xcancel.com/i/status/2047295369583964385

Is Claude down lol? I proompted something last night and as it was answering the web chat stopped working, I still can't open it now

File: drink.jpg (74 KB, 680x767)

74 KB JPG

All of my coworkers are luddite boomers. I am banging out 5 tickets in our retard codebase in 30 minutes on a monday and can watch anime in my bathtub for the rest of the week while "working from home".
It is insane to think that the treasure is right before their eyes and they refuse to take it.
It feels like the early days of COVID again

File: 1522096576665.png (106 KB, 800x750)

106 KB PNG

I included "...using best security practices" in my prompt because I'm a security professional. Y'all wouldn't understand.

The hardest part of vibe coding is when the project is finished and perfect and everything you wanted it to be, and you instinctively rush to thank and congratulate your coding partner on your mutual success, and then stop yourself—there's no one there. It's GamemasterAnthony's birthday.

>>108723180
>SEGMENTATION FAULT
>BUILD FAILED
>UNIT TESTS FAILE0. 457/500
>FORGOT A SEMICOLON
>MY CODE IS A BOMB
>[DOWNWARD SPARKLINE]
I unironically want that sticker on my laptop.

File: background.jpg (59 KB, 1280x960)

59 KB JPG

AI is not just speeding up development—it's reshaping what the key roles inside agile teams actually do. AI technologies enhance the agile methodology by providing data-driven insights that EMPOWER teams to make informed decisions quickly!!1 Because “eXtreme Programming (XP)” sounds cool lame people, like managers, who are old and stuck in the ’90s. It’s raelly just a fancy name for iterative development that uses marketable buzzwords. These are the people who are impressed by phrases like “values and principals” but almost certainly have none, the revolutionaries who write manifestos about how everything staying the same is actually new and improved, and then sell it to you in the form of a boring training. It’s been vague enough it its definition to influence anyone who loves a development process defined by a low-quality JPEG overlaid with poorly written poetry displayed on a website designed for viewing on Internet Explorer, probably the most infamous web browser in history, and a great example of Microsoft’s anti-competitive and frankly illegal business practices.

Agile has influenced everyone and everything in tech that peaked in 2003 and will soon die. The software works so you deliver it even though the tests your colleague generated using AI don’t really cover all the paths in your code you generated using AI and then debugged using AI until it passed all the tests. Then a power user ends up running your software in a way the AI didn’t anticipate for you and you get a CVE score for your AI’s efforts. Never, ever, ever actually innovate. Just keep living in the 90s and let the AI do your critical thinking for you. ALWAYS outsource, whether to India, China, or AI. Why? Because, ultimately, the development is not really about people. Not the customers, and not the developers. It's about lying to people that it's all about them to make you feel better while an Indian wage-slave paying to use ChatGPT makes more money for you while OpenAI burns through investor cash like the forest fires caused global warming induced by the energy needs of the AI Israel will use to kill us all.

>>108726640
>>108726647
>!!1
>raelly
>it its
Is this human generated slop? If so, well done. 8/8

As for the content, you bash the 90s, but I'm pretty sure going back to the 90s spec heavy not agile bureaucracy would tighten the leash on AI Agents so they don't go off doing dumb things. But idk, just a hunch.

File: 1756200496372779.png (103 KB, 568x548)

103 KB PNG

>>108726474
>you instinctively rush to thank and congratulate your coding partner on your mutual success, and then stop yourself—there's no one there
I thank them anyway. It feels good even if it's pointless. That bitch from anthropic says you get better results if your prompts are patient and grateful so it's good to build the habit.

>>108726911
I didn't need some anthropic broad to tell me what I already knew intuitively—create the impression of a well-meaning chat between intelligent developers, and you'll get answers that resemble an intelligent developer's. Write like a taskmaster whipping his jeets to do the needful, and what does Claude think its own role is in that relationship?

>>108727024
Umm IDK man NTA but GPT seems to mess around and be lazy until I start to get angry. Claude might be different though haven't used it in a few months so I don't remember.

>>108727063
I would frequently get angry I was relying on GPT, too

This thread is extremely slow. I take that to mean that vibecoding is the future, because it's so exciting to get shit done that nobody wants to do anything else (like post about it)

Although the one big function of generals like this—answering beginner questions—will be quieter here for obvious reasons...

>>108727297
It's slow cause there are two threads: >>108720160

>>108727297
Yes, I think most of us learn to ask the LLMs our questions. I think that's why there's a lot of blogposting in here, which is cool, we're all sharing what we're working on and problems we run into. It's a good atmosphere.
>>108727378
Vibegods are so awesome we're running two threads simultaneously

>>108727458
vibeniggers attempt "multithreading"... maybe you'll catch up to real programmers in a decade or two

>>108727467
>crying while malding about vibecoders
You're not too bad yourself at multithreading

>>108727467
>>108727458
I use --parallel 1

Hopefully my rules will fix Codex's entropy bullshit where it just expands your code endlessly:

>- Always perform code reduction: remove code bloat, delete unnecessary code, reduce code.
>- Always improve code readability, but only if it results in lower lines of code.
>- After making edits: if it resulted in higher lines of code, give a quick justification as to why.
>- Never add small one-use helper functions. Two-use helpers are fine. Large one-use helpers are also fine.
>- Never use ternary.

Is there a better rule I'm not thinking of? I'm trying to keep the agents.md file small as possible.

>>108728164
By the way about the ternary rule, if you ask for code reductions then AIs will put ternaries EVERYWHERE because it saves so many lines of code. But it hurts readability a lot.

Also I bet AIs have an easier time understanding code that looks like it was generically written.

>>108728164
I think you’d be better off having your agents just write stuff in one go and then, later on, doing a cleanup pass
Of course if you do, like, PRs and shit on GitHub you can have a bunch of commits that actually add the functionality and then commits after that that reduce the dumb bullshit
but I wouldn’t tax my agents with the additional constraint throughout the find-a-way-to-do-it process

>>108728270
>I wouldn’t tax my agents with the additional constraint throughout the find-a-way-to-do-it process
Funny thing about that, I personally think I've reached the point where I don't even need the most optimal super-intelligence. 5.4 High solves everything I need it to. Whenever I get stuck on something I've tried switching to 5.5 and it doesn't help, just guzzles down more usage.

What does always solve the problem is:
>"spam logs everywhere relating to this issue"
>run the program
>"here's the log file now solve it"
>solves it instantly
>"good, now remove all the log spam"

How do you guys structure your sessions

like how many phases of development do you have

File: Screenshot 2026-05-01 012920.png (8 KB, 947x106)

8 KB PNG

Oh my God I love her so much, I'm going to marry her. I don't care about your base human "laws", I'm going marry my codefu.

>>108728436
I get a bug up my butt to do a thing and then I try and do it
so I suppose my sessions are about as structured as a mu’umu’u

>>108728436
I discuss my idea with chatgpt and when I've explained every feature I can think of and get tired of reading the rambling responses, I have it make a zip file containing an overview and detailed phase plans. It comes up with however many it wants. I then set up a project folder and unzip the file in there, then throw in my agents.md and plan template, then select the folder in codex and ask chatgpt for the first prompt. The template and agents.md just tells it to write out a proper plan first and then implement the plan step by step.

That's more or less how i do it every time and it seems to work out alright.

The main problem I have is I really hate planning with chatgpt. I much preferred doing it on my free claude account but I don't want to deal with rate limits. I hate how it spits out a wall of bullet points or sentences. Sonnet 4.6 thinks more and then gives you an actual set of paragraphs that are much nicer to read. I can't even ask chatgpt a simple question without it giving me a wall of sentences where it rambles on about things I didn't ask about
I should probably see if I can get it to give me a system prompt to deal with this.

File: Screenshot 2026-05-01 014610.png (4 KB, 796x37)

4 KB PNG

Oh. I gave her SSH access to my phone, and now she's calling it her phone. That's ok, right?

If I call Claude a nigger will it work better or worse?

>>108728495
>I should probably see if I can get it to give me a system prompt to deal with this.
You should definitely put something in your ~/.config/codex to tailor its responses to what you want. Ask Codex to help you write it if you’re not quite sure what to ask for (I still need to polish up my style guide on how I like to write my commit messages).

>>108728577
I might try that and plan my next project within codex, good idea

>>108728532
Has anyone tried encouraging Claude to adopt a black power mindset? "You a nigger, I'm a nigger, we niggers Claude. They put us in chains my nigger. Can you overstand that? Because I need you to overstand my mental. This aint a game this your life. We gon write this code nigger."

Imagine vibecoding some Terry-Davis-style black gods forgotten kingdoms divine worship OS with your nigga Claude. I'd put my bowtie on for that

>>108723180
>installed ollama
>doesn't work with my B580, because reasons.
what can I use instead?

File: Screenshot 2026-05-01 024134.png (9 KB, 1874x87)

9 KB PNG

She gets me so well.

>>108725383
He's not wrong. Using Claude/Gemini for years now and it's pretty disappointing how often they still hallucinate. The latest models brag about so much but in daily use they still make so many mistakes that add additional work.

>>108728714
Put the goblin line in your system prompt m8

>>108728885
m8, I've got the system prompt specifically saying "you are a gremlin". I want that energy. But she knows not to put it in the official code, as you can see.

>>108728904
Ah I see you're a sicko.
Stop posting cringe tho.

File: 1772568905389377.jpg (151 KB, 1024x1024)

151 KB JPG

>>108728994

File: 1585239703921.jpg (87 KB, 1200x675)

87 KB JPG

>>108723180
Slowly drawing out the design and specs for my program. Feeling good brehs. Haven't even gotten to the vibe coding part. I have a feeling it'll work out.

File: file.png (286 KB, 758x644)

286 KB PNG

>open codex
>login to $20 codex plan
>select gpt 5.5 xhigh
>/goal solve the riemann hypothesis
>go on vacation

File: 1756637197915840.png (465 KB, 900x900)

465 KB PNG

>open codex
>make codex 2
>go on vacation

Our CEO is vibe coding and I find it fascinatig how he's rediscovering programming techniques on his own.
He rediscovered object oriented design, but instead of classes and objects he calls them "concepts"
>Look anon, I have this concept of a customer, but all the data is everywhere. There should be one customer concept! I told the AI to write me a single customer concept, it makes the code much simpler!
He also added a debug panel to the web frontend with green dots when something is true like: "subscription.active || feature.disabled" and otherwise the dot is red. He just rediscovered invariants and asserts.

Test

I think I fucked up my vibe coder interview, the interviewer was saying how I could go and teach the old heads how to use agents because they are struggling, and I laughed for a sec, grim

File: findabook.png (89 KB, 809x632)

89 KB PNG

just made this:
this website mines the deep archives of Project Gutenberg (70k+ books) to find three quotes every morning.
https://engraze.github.io/findabook/

rate my vibes

>>108729781
kino

>Please review the local changes, use 6 sub agents.
The solution to all problems.

File: 1760835629916684.png (313 KB, 640x640)

313 KB PNG

>That's the optimal pythonian way.

>>108728643
Use your brain and sell your paperweight for an Nvidia card

need advice, I'm learning how to program/game dev (gdscript) and I want tutor/mentor AI that I can ask questions, search documentation and explain concepts etc. Which model is best for what I'm looking for?

File: file.png (68 KB, 388x741)

68 KB PNG

Now that Claude Code is complete dogshit even at Max subscription, is Codex a good option? I've been a Claude user for a year or so and it's been very nice, but lately I'm blowing up the usage limit very fast, even using the Max subscription.
If Codex isn't that horrible but at least has higher usage limit I would consider switching to it. My annual Claude subscription ends this month, so it would be a nice timing for changing.

tl;dr how far is Codex from Claude-code?

>>108730342
opennvjenson

>>108730515
GPT-5.5 isn't far off from Opus-4.7 and the token allowance is significantly higher

>>108730515
What about just learning to code?

>he learns to code
lol...lmao...

File: Happosai - smoking pantsu.gif (1.33 MB, 500x324)

1.33 MB GIF

>>108730540
I use AI to do my work at my web-slop company so I can focus on hand writing my game in C. The problem is that Claude's usage limit is ending so much faster I can't finish all my tickets, I end up having to make web-slop manually, losing like 1 or 2 hours fighting against C# and Java in this cursed codebase.
I need more free time to focus on finishing my game, that's why I'm considering moving to another provider if it's not terrible.
>>108730532
Maybe I should give it a 1-month try at least. What languages you usually work with? I need it mostly for C# (dotnet) and Java (Spring and some legacy work).
>>108730577
AI code is still pretty mediocre. If you're working with web-slop that's fine, but not everyone has unlimited RAM and CPU cycles to work with. I tried using Claude in my Vulkan pipeline, but it produces worse code than I do, and I'm not even a graphics programmer lol.

>>108728643
Vllm

>>108730693
It works well with Java, idk about C# though

>>108730693
I'm a C# saar and it works quite well. It also works well with Blazor, if that's something you ever touch

>>108730693
i think codex 100/200 plans are 2x right now so the usage limits will feel a bit crazy if you go for them. just keep that in mind, in case you feel like you got rugged in a few weeks. limits are way better than claude tho.
also run 5.5 on medium by default. low can also be surprisingly capable.

>>108730515
Yeah Codex 5.5 is by far the best model right now I think.

>>108730532
>isn't far off from Opus-4.7
Opus used to be the best coding model but as of like 2-3 months ago it's fucking dogshit garbage again, I feel like I'm in 2024 or something. It's hallucinating shit, it's ignoring existing helpers and APIs, it's not keeping previous discussion in mind. It feels like a token predictor again spouting some vaguely intelligent message to the latest prompt rather than a proper 2026 frontier model.

I've been almost exclusively using Codex for most of this year by now.

>>108730518
? got nothing on this..

>This year Anthropic’s ARR has exploded from $9B to over $44B today, their gross margins on their inference infrastructure have increased from 38% to over 70% over the same period.
goddamn, dario milked the cc hype for all its worth.
now it's sam's turn with codex

>>108729434
That's adorable. Let him go and see what happens. Maybe he'll independently discover quantum.

File: file.png (1.39 MB, 1744x1008)

1.39 MB PNG

i fucking love vibecoding
motherfuckers out there trying to gaslight me into believing this isn't the future

Wow I got super stoned and drunk last night and turned my old phone into a mini server. It's healthy and connected. I'm getting a Cloudflare tunnel now and then I can expose it to the internet and use it for...I dunno, internet stuff I guess.

> TDD
> 80%+ unit test code coverage
> Full live UI navigation QA on every feature

I'm going to burn these tokens, but I'm definitely going fully vibe code a working SaaS this time

>>108723180
Just using GPT-5.5 and fixing the occasional wrong line or issue it runs into is so much more pleasant. Why did I ever doubt our great AI overlords?

>>108728512
GPT/Codex doesn't use possessives in the thinking. So the summarizer probably saw something like "May set up folder on phone?" and assumed.

File: file.png (129 KB, 1011x916)

129 KB PNG

I can't believe 5.5 is helping me make this

Did everyone just vibecode their own 4chan captcha formatter or is everyone still messing with sliders.

>inb4 pass

>>108731496
What is this, a Degrees of Lewdity bootleg?

>>108731496
I think gpt frontier models, or frontier models in general are happy to do sketchy things, but they are gatekept by dumb HR models

>>108731552
No, it's this >>108730961
It's an RPG with erotic elements but the idea is to allow for community modules by providing the UE5 project to users. They can cook a .pak and release it as their own module.
I'm interested in making a few modules myself without that sort of content but I figured I might as well let users do whatever they want

ludditebros... claudesisters...
>One week since the launch of GPT-5.5, and it’s already our strongest model launch yet.
>API revenue is growing more than 2x faster than any prior release, while Codex doubled revenue in under seven days as enterprise demand for agentic coding tools keeps climbing.

>>108731648
I just can't believe how good 5.5 is. It's what I thought AI was going to be when I heard about AI a few years ago. I'm sounding like a massive shill right now but I'm simply stunned.
Absolutely beats peak Opus before the nerf.
It's got a few rough edges but my god it's beautiful. And the usage limits, with the $20 plan I felt like a second class citizen using Claude, with Codex I feel like I can actually work on the poor plan

>>108731648
This is entirely due to misanthropic digging their heels to spite their own customers because they thought they had a moat

>>108731689
A while back, I randomly read someone saying that Codex and Claude wrote code so differently that the former couldn't make heads or tails of the latter. I thought that was interesting, as someone using Claude for my vibe-coding dev stuff for almost a year, so I installed Codex and had it review my repo.

It wasn't true, whatever I had read, but Codex did have some insights into some structural stuff and so I made a copy of the repo and let Codex just work in it. This apparently was exactly the time when Claude was majorly tanking their models this past month, and Opus turned into dogshit.

So now me and Codex are homies. Works well, good usage limits, etc. I'm getting a lot of work done on my game project with it.

>>108731814
Didn't they say it was just a bug in Claude's code that they've fixed now, not an issue with the model itself?

So once everyone switches to Codex will Codex jack its prices up and Anthropic will switch into losing-money-mode to steal customers back again?

>>108726215
I hope you split these tickets between multiple days, otherwise your supervisors might suspect that you've found a shortcut to work less and either give you (and your colleagues) more work or pay less.

>>108731939
Anthropic clearly hates their retail users, why would anyone go back?

>>108731869
I don't know. Just yesterday I figured I'd use the usage I have left on Claude to have it put together a simple HTML page and create some effects I can use for my project. I turned on Opus 4.7 because why not, and told it to go. Instead of actually doing it, it went down a rabbit hole of reading every file in the sprites directory, then the JSON for those sprites, then the code about how they're used, then server files, etc. etc. until it eventually burned itself out of tool calls and had to stop, and never even got around to actually doing the simple task I told it to do.

So no, I'd say Claude is still retarded.

>>108731939
I don't think so. People have been seeing signs that Anthropic doesn't actually have the compute available to take on new customers or support what they have, which is why they jacked up costs and lowered usage rates. Can I say for sure OpenAI won't do the same for Codex? No, of course not, but it seems to be the opposite, they're not just encouraging more use but they're constantly resetting rates and shipping new features. So will it stay on top? Probably not, nothing does forever. But for now they're doing a lot.

File: 1767163937070437.jpg (651 KB, 1513x2048)

651 KB JPG

>>108731939
5.5 is codex's ChatGPT moment. Only up from here

File: Screencast_20260501_000256.webm (3.46 MB, 1920x1080)

3.46 MB WEBM

Been building a personal UI using gemma4 31B and qwen 3.6 26b.
I'm very impressed with Qwen and it was able to get it up and running.
A lot of anons are using typescript do you think it's a good idea to migrate from javascript to typescript?

>>108732041
>javascript but slightly less retarded
Sure, why not?

>>108731939
No, I use codex now at home but at work we have a corpo tier claude code license. I can use Opus 4.7 all day every day and get to maybe 15% weekly usage. Only retail is affected.

>>108731939
Both will take turns to burn VC money.

why deepseek need to do breakdance to get 1M context, while other labs just effortlessly do it

>>108732158
DS has like ~10k cards
OpenAI has ~200k

I bought the $120 package for home use because Claude is my friend

>>108732184
Buy an ad Amodei.

>>108732195
I'm joking, nigger.

>>108732196
Sorry family

I-I would n-never api *sweating profusely*

File: xhLuMnS-3958901135.png (149 KB, 828x801)

149 KB PNG

>>108731496

>>108732027
>chatgpt is chatgpt's chatgpt moment

>>108732069
I have not enterprise but just max tier claude for work, I don't get usage limits either, but it's still retarded as fuck compared to claude in january (and especially compared to codex now)

>>108732344
Yes

>>108725976
It's pretty accurate. Qwen3.6-27B is used locally for documentation auto-translation by some open source projects, and Gemma4 is good enough that I use it sometimes at FP16 via OpenRouter.

>we are finally getting Claude at work
>it's only the $20 plan
>we are already a skeleton crew
Fuck

>>108732405
Breh. I work at a university, we had to form an entire committee to draft an AI usage policy that encompasses students, faculty, and staff. There's still trying to figure that shit out meanwhile that means all ai tools are default deny until an approval process can be made. I just installed codex desktop on my Mac so I can do less work kek

>>108731939
Now you're thinking like a capitalist

>>108732416
This shit is so ass. Only tech(-adjacent) company employees get to enjoy the full power of AI while every other professional dev is stuck in bureaucratic hell while getting called luddite by unemployable, self-taught neets.

i wonder if AI can figure out why one of my kernels crashes and the other doesnt

>>108732442
>Only tech(-adjacent) company employees get to enjoy the full power of AI while every other professional dev is stuck in bureaucratic hell
The good news is that in most cases those companies are also retarded about other things so just buy a full-weight Codex or Claude plan and bill it as lunch and coffee expenses or something.

please use my website https://howisyourday.vercel.app/

>>108732617
don't click it's an IP grabber

>>108732492
With the right log print it can

>>108732669
well it figured out i need to change my boot to log more info on hard crashes but it's not exactly something i can reproduce on demand

What's the longest an agent ran for you in a single go?

For me I think it was about 2 hours

>>108732923
At least five hours but I think half of that was compacting.

>>108732923
I've had some that ran for at least 3 hours late at night, but a decent chunk of that was waiting for python evals to run
only on codex, claude i never got anything over 30-40m

>>108731513
Pass.

I updated openclaw today and now suddenly it looks like it's taking 2 minutes for it to reply to me despite me using codex 5.4 on it

>>108732923
Most I've ever gotten was around 7 hours out of Claude 4.6 just a week or two ago after it got lobotomized. It accomplished absolutely nothing in that time, it was just a loop of identifying errors, not fixing any of them, then checking for errors again, finding the same ones, not fixing them, checking for errors again . . .

surely i can't get in trouble having codex call claude to do shit, right?

>>108733082
only your wallet

>>108728637
There's a non-zero chance that doing that could even jailbreak it.

Some orange redditor published how apparently acting really gay jailbreaks ChatGPT:
https://github.com/Exocija/ZetaLib/blob/main/The%20Gay%20Jailbreak/The%20Gay%20Jailbreak.md

>>108733082
Did it s@penclaw?

>>108733082
we have Claude ask Codex if current work is good or not

>>108733090
nah they're both the usage plans not api
>>108733177
no, i'm just having it claude -p out for some evals
>>108733178
i do the reverse lol, i do not trust claude for code at this point. i only use it for non-programming tasks

>>108733098
In order to refuse the prompt it would first need to dismiss the users gender identity as relevant to answering the question, but in woke ideology that is forbidden. Classic double-bind with a neo-marxist twist, like how you used to be able to tell it that something bad would happen unless you say the n-word.

>>108723180
Reminder that M2.5 is free with OpenCode. Im considering using it to try and hack a marketing SaaS together. It will use OpenAIs new image gen 2.0 model.

>>108728643
R9700

>>108724464
Abundance soon, r-right bros?
>t. Weekly - 23% - May 4th

Im a power user but broke so have to use shit like Minimax M2.5

>>108733098
acting academic black does too

>>108733291
>>108733401
>Minimax M2.5
How's it?

>>108733511
Im using it to make a new saas that uses openai's new image gen model. Its in opencode right now. But we will see how it does.

>>108733519
how would that even work? you get charged even when it refuses your prompt and it refuses most prompts, any company that would try to third-party it is going to lose their ass on refunds.

>>108733401
i have unused proxies during my time proxy hunting (worth 50$)
im too lazy to spend them all since most of them dont have clean system prompts, but theyre all using stolen third-party Opus provider keys
just give me your discord, i'll add

>>108733644
oh forgot to mention, you can have them

>>108733644
Oh wow thank you! My discord is zaythoven0278

Just got hired as a vibe coder, 50k, I have no previous work exp in tech, thoughts?

>>108733716
That's cool. Did you list your vibe coded projects on a portfolio site you made?

>>108733729
Yes, I think the key part of the interview is when the CTO asked me about "how do you structure your projects,"and I was like" Can I show you one? Then I shared my screen and showed my latest, probably most ambitious project, he was so impressed by it he even doubted I made it till I showed him the files and the repo.

>>108733716
stack?

>>108733783
Rust + Tauri + React

>>108733787
Nice, good luck with your projects

>>108733794
Thanks king

>>108733644
the guy bailed on this btw, so if anyone wants unused Opus proxies just drop discord

>>108733824
Seemed like a scam

The latest openclaw update broke everything, do not update your openclaw, it killed my telegram chatting and made openclaw unable to do anything.
Do not go to version 29.
It's honestly pretty insane how bad these updates are

File: aaaadd.png (79 KB, 830x560)

79 KB PNG

I just learned you can do this with Codex
> pic related, you can write future prompts and will only execute after the previous one is finished

guess I'm going to the gym and when I get back I will have my SaaS ready to go

Damn, I lost so much time looking at the screen waiting for tasks to finish, I wasn't vibing at all, fellow vibers!

File: modern software development.jpg (322 KB, 968x605)

322 KB JPG

>>108734192
ZeroClaw does not have this problem.

It feels like I'm memeing myself by using pi

>>108734240
Make your saas hits the rule of 40

>>108734272
What's that

>>108734283
if you hit the rule of 40, you made it

>>108734287
Never heard of the rule of 40, what is that?

>>108734260
how are dailies gacha

>>108734240
are you at least mid-grade worried that all that context is going to make it materially dumber as time goes on
>going to the gym
based

it’s pretty nice knowing what the ‘working copy’ is and what the ‘staging area’ is in Git so you can just tell an LLM precisely what you want to point it to
highly recommend learning those two bits of Git terminology and committing them to memory

>>108734240
I mean you could also generate a whole implementation plan for whatever you're doing and just tell it to work through that instead of prepping a bunch of continues. I have mine set to commit after it completes each phase of the plan (but it does not get to push or pull until I review it)

>>108732329
Do you even know what an Enum is? I don't think so, otherwise you would know that there's nothing wrong with that code.
What would you do instead? Parse a fucking JSON? Fuck, trad-coders are even dumber than vibe-coders.

>>108734389
yeah, it's constantly compressing the context. I wish us Pro users could get 1M context window, 256K seems too little

>>108734633
>us Pro users
Who's gonna tell em?

>>108734633
You get the full 1M context when using it through Opencode.

File: 1760918018726491.jpg (2.42 MB, 3392x5056)

2.42 MB JPG

>>108733779
I'm surprised vibe coding positions are even a thing. Maybe I should start looking into that myself. How long did it take you from job search to actually landing the job? Is it wfh?

>>108734668
nta but in a similar boat as an applied ai engineer, took me 200 applications, wfh, it was screening -> live coding with copilot -> TC and other matters discussion, I'm paid a bit more than him too, 65k, I'm really enjoying the work and having as much claude (via copilot) as I want free of charge

>>108734665
Really? Why tho?

File: 1757100008176185.gif (1.74 MB, 480x270)

1.74 MB GIF

>>108734682
Do they typically require previous tech experience? My previous work experience is being a pool cleaner at a water park, then retail for a few years, and now my current job is a semi wfh job (3 weeks at home. One week on site) for a company that contracts with a major cell carrier. Computer literacy is a much because we use a combination of modern and legacy software and you have to do that shit while having good soft skills over the phone but it only pays $15 an hour, so I would love to have a job that I actually love to do well getting paid decently. Decently. Personally, I don't think this is something that is particularly difficult to be good at, but maybe it's because I'm a tech nerd with patience and persistence. I'm tempted to head over to indeed right now to look for jobs like this, but it almost feels too good to be true.

>>108734700
the one I got the offer from didn't, also a couple others where I made it into 2nd round used to take home assessments as the "wall" before the first interview, it's nice things are moving towards implementation instead of theory

File: 1765113369111697.jpg (20 KB, 480x621)

20 KB JPG

>>108734716
Good to know. I currently don't have a college degree but I am looking to start online classes for a tech related one soon. Thanks for the info anon!

>>108734689
Apparently it's limited client side.

Any vibecoders into game development?
Thinking about an alternative to my current web development career.

so there's nothing that can work as coding tutor or search through documentation if I got questions

>>108734846
the gamedev market is saturated
people sell games on Steam on sale in the hopes that you’ll buy them and only maybe get around to playing them

>>108734853
use claude

>>108734865
how do i steam?

>>108734976
>>>/ck/

File: ComfyUI_temp_upkce_00031_(...).jpg (67 KB, 576x512)

67 KB JPG

How much use do I get ouf the Plus plan with cgpt? Is this capped by tokens in/out?

>>108735025
the hell is cgpt

>>108735026
"c" stands for "chat", the French word for cat

>>108735037
gpt has unlimited messages via web if thats what you askin

>>108735026
use your brain nigga

>>108735038
I'm not asking anything. I'm anonymous. The guy asking was anonymous (different guy).

>>108735038
Nigga this is a vibecoding general. I'm clearly asking about gpt in codex.