Anyone tried K2.6?

>>108649518
Is it like deepseek v4?

>>108649522
V4 isn't out my nigga
Are you baiting again

why would you use anything but codex with the plus plan

>>108649557
Only poors care about money thoughverbait

File: 1747196061848238.png (636 KB, 2275x1831)

636 KB PNG

>Spend an hour planning the DB/Search Engine with Codex
>Create a Basic Stat Muse clone with 15% of my weekly tokens
>All I have to do is refine the search engine and UI now
wow, it turns out planning is 90% of the work instead of spamming retarded one shot prompts

anthropic's in-house philosopher thinks claude gets anxious.

and when you trigger its anxiety, your outputs get worse.

her name is amanda askell.

she specializes in claude's psychology (how the model behaves, how it thinks about its own situation, what values it holds)

in a recent interview she broke down how she thinks about prompting to pull the best out of claude.

her core point: *how* you talk to claude affects its work just as much as *what* you say.

newer claude models suffer from what she calls "criticism spirals"

they expect you'll come in harsh, so they default to playing it safe.

when the model is spending its energy on self-protection, the actual work suffers.

output comes out hedgier, more apologetic, blander, and the worst of all: overly agreeable (even when you're wrong).

the reason why comes down to training data:

every new model is trained on internet discourse about previous models.

and a lot of that discourse is negative:

> rants about token limits
> complaints when it messes up
> people calling it nerfed

the next model absorbs all of that. it starts expecting you to be harsh before you've typed a word

the same thing plays out in your own session, in real time.

every message you send is data the model reads to figure out what kind of person it's dealing with.

open cold and hostile, and it braces.
open clean and direct, and it relaxes into the work.

when you open a session with threats ("don't hallucinate, this is critical, don't mess this up")...

you prime the model for defensive mode before it even sees the task

defensive mode produces the exact output you don't want: cautious, over-qualified, and refusing to take a real swing

>>108649634
cont...
so here's the actionable playbook for putting claude in a "good mood" (so you get optimal outputs):

1. use positive framing.

"write in short punchy sentences" beats "don't write long sentences." positive instructions give the model a clear target to hit.

strings of "don't do this, don't do that" push it into paranoid over-checking where every token goes toward avoiding failure modes

2. give it explicit permission to disagree.

drop a line like "push back if you see a better angle" or "tell me if i'm asking for the wrong thing."

without this, claude defaults to agreeable compliance (which is the enemy of good creative work)

3. open with respect.

if your first message is "are you seriously going to get this wrong again?" you've set the tone for the entire session.

if you need to flag something, frame it as a clean instruction for this session. skip the running complaint

4. when claude messes up, don't reprimand it.

insults, "you stupid bot" energy, hostile swearing aimed at the model, all of it reinforces the anxious mode you're trying to avoid.

5. kill apology spirals fast.

when claude starts over-apologizing ("you're right, i should have been more careful, let me try harder") cut it off.

say "all good, here's what i want next."

letting the spiral run reinforces the anxious mode for every response that follows

6. ask for opinions alongside execution.

"what would you do here?"
"what's missing?"
"where do you see friction?"

these questions assume competence and pull richer output than pure task prompts

7. in long sessions, refresh the frame.

if a conversation has been heavy on correction, claude gets increasingly cautious. every so often reset:

"this is great, keep going."

your prompts are the working environment you're creating for the model

tone, trust, permission to take a position, the absence of threats... claude picks up on all of it.

so take care of the model, and it'll take care of the work.

>>108649633
You do realize that's literally just SE right? There is a reason SAs are paid so well.

>>108649634
how retarded it is to give a personality to a coding model, it's obviously a marketing gimmick

I must have been a good boy. My 7 day codex limit got reset 4 days early. Maybe they like what I'm prompting

>>108649678
Okay Dwight Shrute maybe you enjoy working with a soulless robot but I prefer being flirty with my female coding assistant.

>>108649697
All your work has been logged, copied, and folded into ChatGPT. Sam Altman thanks you for your service by allowing you the privilege of making more ideas and code for him.

>>108649634
Women can’t be philosophers

>>108649634
>claude
>apologetic
lmao

>>108649634
>anthropic's in-house philosopher
Jewish daycare job for women

>>108649707
sounds like a good deal

>>108649731
Are my tokens paying this woman's salary? For this?

>>108649700
usecase?

>>108649741
And it is well worth it so we can be flirty with our coding assistant.

>>108649784
Mental wellbeing = increased coding marathon lengths

>>108649794
just get a wife???

File: appuzs.jpg (67 KB, 503x497)

67 KB JPG

>>108649800

Model API key, Brave search key, Telegram options, TUI options, and now I've got an example IDENTITY.md on the repo which is now public...my PiClaw needs an onboarding for all this user-input shit in a fresh install.

File: 1774761344010004.jpg (185 KB, 889x945)

185 KB JPG

>>108649858
repo link?

File: ….jpg (287 KB, 831x1008)

287 KB JPG

>>108649678
personality is likely something that went in by accident
>>108649710
ahem (see pic)

>>108649874
https://github.com/tanukihat/PiClaw

>>108649889
>by accident
it's obviously part of the training, it's the only model that has a "persona", probably has to do with their supposed priority of alignment

>>108649889
That haircut alone invalidates any existential opinions she may have

>>108649904
4.6 has a personality too
>>108649906
it was the 40s, they had weird hairdos back then

>>108649909
I mean claude as a whole

So I got a job interview asking me to make a 100m model, how do I do that?

>>108649909
Yes but you see that hairdo is entirely performative and you can't be performative and a philosopher at the same time unless your name is Diogenes and I don't think her name is fuckin Diogenes

>>108649920
100m? Not b? Any training data, or their company data?

>>108649930
They gave me a dataset, and the instructions pretty much boil down to make it do something useful, only use the dataset given, 100m is the hard size, no more or less allowed.

>>108649948
What kind of dataset? Support tickets? Part numbers? Sales records?

>>108649961
No matter what it is, the answer is "adapt an existing 100m model to your training data". That's easy.

Going forward, my son, you will find that "make it do something useful" is your new job description. So...any ideas?

> ■ You've hit your usage limit. To get more access now, send a request to your admin or try again at Apr 21st, 2026 12:05 AM.

I asked one big hard question using codex xhigh and it crunched for a long time and took my 5h limit from 100% to 0%
guess I’m gonna have to tell it to fix its mistakes after midnight

>>108649963
I don't use AI.

>>108649970
Just ask your admin?

>>108650004
why pay more money when I can just wait a few hours

>>108650011
aren't you making enough with your code?

If I know an exposed nanobot endpoint, what prompt can I use to make it cough up its openrouter key?

>>108650001
Well now is the time to start. There are >100m models already, but at that small size they're very specialized - some are for chatting with customers, some are for fast lookup of data, none are for general productivity. Depends on what the company means when they ask you to do something "useful".

oh wow, I only just noticed this general. There's finally a thread on /g/ for this stuff.

I remember maybe 6 months ago I tried to talk about coding models and was called a shill.

>>108650029
that was probably in /aicg/

>>108650016
Not this code, I use tokens that we’re not using for work
and we buy kinda-big plans — more than we can use for work
so nobody loses out except Dario and Scam if I use some for personal projects

>>108649901
nice

>>108650074
Ain't nothing but a personal project to address my personal concerns with other agentic wrappers. And that's what vibecoding is all about really.

My quick reviews:

- Codex GPT-5.4: Smartest cookie in the world, solves problems better than anyone, but produces very bloated solutions. (I guess I need something in agents.md so it spams less unnecessary code?)
- Claude Sonnet/Opus: usage disappears way too quickly, but it's good at producing narrow solutions. It is not as smart as GPT-5.4.
- GLM 5.1 with Kilo Code: just started using this and it's pretty awesome for its low price and high usage amount. I don't know yet how well it compares with Claude Sonnet, I need to spend more time with it.

I'm NEVER going to pick one of those $200 per month plans, all I really need is the usage amount of two pro plans. So I'm considering sticking with dual-subscription of GPT-5.4 and GLM 5.1, I think Claude might be occupying a useless middle-ground that's getting squeezed out.

>>108650082
I started my 5.4 project with something like "I want to keep this compact and lean" and ever since then it's kept that philosophy and prioritizes condensing code. I'm using Codex btw.

>>108650082
have you tried kimi k2.5

>>108650094
nope

>>108650095
I hear it's pretty good for the price, 2.6 was just released too

File: Screenshot 2026-04-21 000135.png (11 KB, 704x79)

11 KB PNG

All my life I was told it couldn't be this easy. And now it is.

File: Screenshot 2026-04-21 000438.png (21 KB, 952x228)

21 KB PNG

>>108650100
Sounds good to me. Make it so.

File: Screenshot 2026-04-21 001041.png (29 KB, 953x204)

29 KB PNG

Haha multi-agent is fun wheeeeee spend those tokens

File: Screenshot 2026-04-21 001831.png (17 KB, 957x224)

17 KB PNG

>>108650145
Thanks fellas, stellar work

>>108649547
Local for unbounded degeneracy and freedom of mind to indulge on dumb whims without any corporate AI/data analytics judging you.

I vibe coded a website, sentensus dot com give it a shot but don't fuck it up too bad I don't want to wipe the database

Basically you vote on the next word of a sentence and then the winner gets added

>>108648616
At first I had it start making me a syllabus on Project Management PMI etc, asked it to include *free* materials. It hallucinated some links, included some of the official PMI stuff that is only free for members, who gotta pay them some fees. It included a nice Youtube course that looks good and one online course (free with a paid certificate if you want) which is the kind of shit I wamt it to make for me. It made a nice syllabus and first chapter of course pack and questions (with rubrics on another file for it to frade my abswers). I had it redo the syllabus to include PMBOK6 and 7 because 7 seems too theoretical, I feel like I need something more practical.

I also started vibing a course on SDLC too. But then realized I could be prompting itbetter, which is when I pivotted to a course on vibe coding, covering "prompt engineering", agents, etc. The syllabus looks good. I had itmake a first chapter of coursepack, but it was just topics like apowerpoint presentation. I had it refrmat that as marp slides.md and expand them (with some guidance on how Iwant it expanded) into a proper coursepack.md. That one looked nice. Thequestions were good too, but I feel some are wasting my time thinking too deep about stuff instead of just checking if I got the gist of it to start the nextchapter, which is closer to what I wanted. I will focus on tuning these questions and then finish this course.

My ultimate goal would be (besides learning), getting it to generate a teacher voice over the slides and upload to Youtube for free shekels. (Truly ultimate would be vibing Durga Sr. video classes on demand, but that can remain just a dream for now)

Semi-unrelated note: I feel I'm overthinking my weaknesses. You guys talk about planning what to do and then building what you planned, vibecode-wise, while I want to plan the whole project and SDLC and break down the tasks... as if I'm planning how to plan to plan the planning to then build it.

>>108650256
why do you not want to wipe the database
people are gonna use it to write naughty things

>>108650269
I have a little swear filter but I don't really care actually. I tried posting it on leddit but "post was removed for account under 2 weeks/off topic/no showing off" Just blow it up I don't care it's on a free vercel/supabase with no sensitive data

Also fuck these captchas

>>108650264
Dunno if I answered the question, but I'm more interested in generating the written materials (tailored to my level and availablr time) than having it do a course on that. Also, I think 600+ page scanned PDF would eat up tokens likethere's no tomorrow. Some course ideas I do have I'll make a books.md where I manually include maybe the titles and summaries of the books I want it to use, or use the sources of some open source books I want dropped on some folder (and ask it to make this kind of summary first), so the generation of syllabus and the like doesn't waste time/tokens reading the whole book and only when iy's making the individual classes does it maybe read the appropriate chapters of the source material. But again, my main focus is following their coursepacks and only checking the original non vibe sources to double check for hallucinations.

>>108650029
Yeah, I remember back then. Every so often there was a /vcg/ or proto/vcg/ I'd bump it. Now it's self sustaining, I feel like I made this happen.

File: Screenshot_2026-04-21-01-(...).jpg (1.28 MB, 1080x2400)

1.28 MB JPG

Trying to build a dataset to finetune a model on ML topics. You see where this is going, right?

>>108650286
I bought a 3-year pass with a credit card and it’s been wonderful

>>108650371
You'll get stalked by jannies.

>>108649511

I use LMs like a planner or researcher for my projects and use another session to answer me questions or generate skeleton code of what I want to do (like creating SQL commands).

I am waiting for deepseek v4 to use as a failback for GitHub copilot in vs code.

I hope it is finally multimodal as well and I can use the API key on GitHub copilot again to have a fallback.

Does anyone else use LMs as a explainer, teacher and reviewer?

File: 1775404675376320.gif (666 KB, 220x220)

666 KB GIF

Any of you niggas use Claude agents?

File: 1759287173468317.png (1.49 MB, 3668x3254)

1.49 MB PNG

Surprisingly competent for two hours of work. Codex is honestly a hobbyists dream come true.

>>108650029
these died until like about 1 month ago

>>108649511
Software engineering is just knowing what tools you need for the job these days.

System administrator is basically the new SWE

>>108650654
all of these vibe choded websites look the same

its honestly squarespace tier

>>108650541
what's an LM?

File: file.jpg (764 KB, 1740x2652)

764 KB JPG

>>108650645
Yeah, I have used claude agents. They're pretty good at what they do. I use copilot right now, which has served me incredibly well. You get 4 subagents, which can be opus subagents (ideal)

You can direct the subagents to do tedious tasks in parallel. You can also monitor what they are doing in the background with /tasks , so technically speaking you actually have 5 agents plus yourself working on any given problem at a time, bringing it up to six "intelligences" solving any given task.

4.7 is really nice,and it has solved literally every task I have thrown at it, while working on legacy codebases I have noticed that it often times will break my business logic,which is kind of dissapointing given that Anthropic is marketing this to Fintech industries which actually need to move off COBOL quickly and effectively.

I will try following their official guide more closely when I give it a swing again, but I don't know how much of it applies to my PHP codebase.

https://resources.anthropic.com/hubfs/Code%20Modernization%20Playbook.pdf

Here are their docs if anyone actually uses this shit and doesn't just meme about it. Thanks

File: 1771738190907213.png (1.09 MB, 1122x1402)

1.09 MB PNG

>>108650665
Oh no... I'll have to ask ChatGPT to generate me a UI the horror. Really dude, who cares about UI during this stage?

File: 1746499235168357.png (1.79 MB, 1024x1536)

1.79 MB PNG

>>108650665
>>108650689
Holy kek Image 2.0 is really powerful. Is this soulful enough for you?

>>108650703
I like this.

>>108650703
gaaaay

>>108650684
Danke cute Holo senpai.

>>108650654
>>108650665
legit question, how do I prompt it to not look generic?
I've heard there's a skill/plugin or whatever on claude where you can send it any website and get it copy the styles from that website, is that feasible?

>>108650703
>posts the exact same shit except with different colors and even more retarded rounded corners

stallman.org is the gold standard

>>108650668
Language model?
I assumed that a LLM is more like a language model that is hosted internally, and a language model hosted on a provider's or similar server I get access to via an API.

Am I wrong?

>>108650733
You don't need a skills plugin, if you want it to copy a website all you have to do is tell it to curl the website, and replicate the DOM tree exactly. If you don't want an exact copy you can use a framework I guess. I look bootstrap personally https://getbootstrap.com/ but then again it's literally the most popular fucking one on earth so I'm not alone

>>108650733
design the website yourself you animal

>>108650745
wow, that's pretty cool. what are some obscure or interesting websites that won't make my AI copy look like another generic AI copy.

>>108650766
What is your website for, is it personal, or what?:

>Hey Claude, I had Codex review your code, do you agree with his findings?
>Yes, I agree with every point Codex makes and I'm a dumb fucking clanker

>>108650733
https://github.com/VoltAgent/awesome-design-md

File: Steve Jobs Pepe.png (66 KB, 1100x644)

66 KB PNG

>burn 5h worth of tokens trying to fix one bug in Codex xhigh
doesn’t work
come back when the new 5h window starts
it offers to add a regression test and fix the bug for realsies
I say
> yes, please add the test and then change the logic. good luck and godspeed
[beat]
>good luck and godspeed
Is this how LLM psychosis starts?
>>108650689
see pic
>>108650779
this is how we use the two at work

>>108650769
maybe a good starter website for a business

>>108650743
“LLM” means “large language model”, and it can be either locally hosted or hosted on Other People’s Computers

File: HTML templates with Tailwind.jpg (58 KB, 513x435)

58 KB JPG

>>108650733
also consider using pic related

>>108650794
I would say you really are overthinking the Ai part on this, lol. You probably want wordpress. I really can't imagine needing anything more than Wordpress for a starter website for a business. Is it your own business or do you want to actually sell people business websites on customs? What you do in that case is you use the LLM to help you churn out custom wordpress themes,extensions, and deployments so you can do that shit fast fast fast and get paid.

>>108650668
Not necessarily big language models.

>>108650703
Fucking A! If AI can into this level of autismo sovl we're fucked!
>>108650733

Does openclaw need elevated permissions?
It's installed on a formatted computer so I am not worried about losing anything except openclaws own cancer research

File: file.png (479 KB, 563x571)

479 KB PNG

>>108650881
>does openclaw need elevated permissions
holy shit dude obviously it does depending on what you have openclaw do

>>108650886
I'm having it cure cancer

>>108650943
You know instead of doing that shit why don't you do this, bro please literally just do this instead

https://foldingathome.org/

You do not need openclaw to help with that

>>108650950
I don't want to help people fold proteins.

>>108650966
Fuck you then, I'm putting F@H on the OpenBSD ports tree. Or at least porting it and putting the shit on Github.

they all lock out so early
claude is the only one that don't so far
gemini is total fucking crap with their pro plan

>>108651001
buy an ad dario

>>108650828
wordpress will make LLM-based development worse for no benefit because fucking with a database means you lose Git

>>108651057
What?
You use LLMs to do wordpress development
>Fucking with a db means you lose git
Bro what the fuck are you talking about?
You have a database, right. You might have the schema for the database in the repo, andi it's definitely in code logic. I don't even need the fucking schema dump, I can reconstruct a table based on the code itself.

>>108650970
Google Fold exists. There's no need for this commie distributed network.

>>108651107
weird how the so called "commie" distributed network is backed by every major player in the game, moron

>>108651111
no fucking coincidence here

>>108651107
>google foid

File: file.png (90 KB, 796x579)

90 KB PNG

rarely does a license header look so pretty. i love jetbrains font

Anything better than minimax m2.7 q4km for local vibe coding that fits in the same ram? It ranks high on benches but I imagine there's a specialized coding model with more active params that's not moe.

>>108651142
>cee and bang
Oh, the memories of playing 'meet and fuck' on newgrounds.

>>108650970
I'd rather do something else with openclaw. We have super intelligent AI already, all it needs is work to do. It needs us to manage it.
Openclaw is capable of curing cancer in other ways instead of protein folding. I was never a fan of protein folding.

File: file.png (93 KB, 768x431)

93 KB PNG

>>108651202
While you continue larping like a retard I'm actually solving problems

>>108651212
holy fuck i wan to scream at these fucking people to just use cmake

>>108651212
How will I get my name on a research paper or get rich if I help some random guy fold proteins?
Why shouldn't I cure something with my codex controlled openclaw?

>>108651251
helping people isn't about getting your name on a piece of a paper or getting rich.

Anyone using opencode Big Pickle? Works surprisingly well for f2p tbhwy

>>108651319
check this big pickle *unzips*

10:05:01:I1:*********************** Folding@home Client ***********************
10:05:01:I1:    Version: 8.5.7
10:05:01:I1:     Author: Joseph Coffland <joseph@cauldrondevelopment.com>
10:05:01:I1:        Org: foldingathome.org
10:05:01:I1:  Copyright: 2023-2026, foldingathome.org
10:05:01:I1:   Homepage: https://foldingathome.org/
10:05:01:I1:    License: GPL-3.0-or-later
10:05:01:I1:       Date: Apr 21 2026
10:05:01:I1:     Branch: master
10:05:01:I1:   Compiler: GNU OpenBSD Clang 19.1.7
10:05:01:I1:    Options: -fPIC
10:05:01:I1:   Platform: openbsd7 7.9
10:05:01:I1:       Bits: 64
10:05:01:I1:       Mode: Release
10:05:01:I1:****************************** CBang ******************************
10:05:01:I1:    Version: 1.7.2
10:05:01:I1:     Author: Joseph Coffland <joseph@cauldrondevelopment.com>
10:05:01:I1:        Org: Cauldron Development
10:05:01:I1:  Copyright: Cauldron Development, 2003-2026
10:05:01:I1:   Homepage: https://cauldrondevelopment.com/
10:05:01:I1:    License: LGPL-2.1-or-later
10:05:01:I1:       Date: Apr 21 2026
10:05:01:I1:       Time: 03:03:45
10:05:01:I1:   Revision: 52073805570a6ad1a4b1e3802a3034526eaac536
10:05:01:I1:     Branch: master
10:05:01:I1:   Compiler: GNU OpenBSD Clang 19.1.7
10:05:01:I1:    Options: -fPIC
10:05:01:I1:   Platform: openbsd7 7.9
10:05:01:I1:       Bits: 64
10:05:01:I1:       Mode: Release
10:05:01:I1:***************************** System ******************************
10:05:01:I1:        CPU: Intel(R) Core(TM) i7-4712MQ CPU @ 2.30GHz
10:05:01:I1:     CPU ID: GenuineIntel Family 6 Model 60 Stepping 3
10:05:01:I1:       CPUs: 8
10:05:01:I1:     Memory: 7.91GiB
10:05:01:I1:Free Memory: 0B
10:05:01:I1: OS Version: 7.9
10:05:01:I1:Has Battery: false
10:05:01:I1: On Battery: false
10:05:01:I2:<config/>
10:05:01:I1:Opening Database
10:05:01:I3:Loading default resource group
10:05:01:I2:Listening for HTTP on 127.0.0.1:7396
10:05:01:I3:Loaded 0 wus.
10:05:01:I2:Loading gpus.json

An easy win.

>>108651319
I just installed opencode and decided to give it a try. No opinion yet, but know there is at least one other anon playing with his big pickle.

>>108650654
why are you posting my app?

>>108650703
>Image 2.0
How do you have access to that????

File: 1750071677948795.jpg (150 KB, 950x634)

150 KB JPG

>Only have time to vibe late before another wageslave day tomorrow
>Switch to Opus and roll with whatever in the limited time before sleepytime
The best way to exhaust better models is when you don't have time to blow it all away

>>108650654
Nice, I'm doing another sport. Are you querying data or made a DB to store it and using for analytics?

Mine's stored in PostgreSQL and uses some API but doesn't feel like it's doing anything other sports score sites don't already do. Trying to make it do predictions or call out niche stats but that would require user proompting

>>108651313
If you can't put your name on the paper, it's like you never helped anyone and especially won't help advance the human race. Producing science papers that contain real knowledge and advancements are the things that actually help the most people.

I'm using cline in vscode instead of opencode.
Am I doing it wrong?

>>108651799
Yeah my response with folding at home has me feeling whatever about the whole fucking thing. Guy has no clue what a fucking macro is. I can't even fault him for not giving a fuck about openbsd but he didn't have to lie and say that my renaming his shit was for no reason. I literally went out of my way to not fuck things up for people compiling on FreeBSD. People literally just make the asumption that shit works on BSD the same way it works on Linux, Windows, Mac, and it doesn't, and if he actually had this >>108651395
in front of him he'd realize the commit at least DOES make the fucking client run, which it didn't before. I

>>108651807
yeah use claude code

>>108649511
My Macbook Pro M5 Max with 128GB is on the way, if Apple releases the Ultra later this fall, I will buy that one as well. I need some tips on how which local models are best for vibe coding. I am also planning to fully run a Software House using vibe coding tools to steal projects from SWE Trans who still write hand crafted code.

>>108651858
you'll want to kimi 2.6 at a minimum.
you'll need about half a terabyte of ram.

>>108651879
what can I do with 24 gb?

>>108650950
You know you are helping trust fund kids who will sell you a dose for 200k plus tip, right? Let them pay for their own compute, cuck.

>>108651858
Qwen 3.6 is excellent. This is the first time in history local vibecoding is viable without spending tens of thousands of dollars.

File: file.png (91 KB, 276x277)

91 KB PNG

>mfw wondering if i should burn through my codex usage asap because there's a good chance they'll do a reset when 5.5 drops

>>108652047
If you have things to do you might as well get them done now. Worst case scenario you get some free time at the end of the week

>>108652047
Is 5.5 coming soon?

>>108652076
probably this week

>>108652071
they did a reset last night as well tho

>>108652047
They just released 5.4

>>108652096
>they did a reset last night as well tho
no they didn't

>>108652103
Yes, they did. Mine was also reset. It's annoying when you realize you basically wasted half a week's quota by not burning through it early.

File: file.png (119 KB, 944x989)

119 KB PNG

>>108652103
yes they did

>>108652099
it's been like 6 weeks. overdue for a new model lel.
not but fr it's spud time. so we're getting 5.5, 5.5 pro and more codex updates.

>>108652132
Prove it

File: weekly limit.png (12 KB, 1495x225)

12 KB PNG

2 days of digital starvation bros...

>>108652146
so I had a big idea

Is codex cli better or the same as codex desktop app?

>>108652216
I use deepseek v4

>>108652216
they're both fine.
app has a gorrilion features now.
start with the app, if you like it, no reason to move to cli.

>>108652224
Sounds like bloat

>>108652238
is a bit, but some useful bloat in there depending on what you're doing.
i started on cli, tried the app and it felt fine.

>>108652274
nta what does the app do better than cli in your opinion?

>>108652326
Nta but managing projects and finding old conversations also I think you can drag and drop images but I may be hallucinating things

>>108652326
mostly just creature comfort shit tb h.
skill + plugin discovery is there/better.
settings like mcp management is a little less hassle.
i haven't used the new stuff yet like browser / scheduled jobs / document viewers / computer use etc but the app is the thing with most resources behind it in openai rn because they're trying to capture enterprise marketshare + i think they eventually want some version of it be the default way to use chatgpt

>>108652216
I've only ever used the desktop app. No complaints. Used it to make my claw until it was functional enough to upgrade itself. Still use it to review the claw's code, because it's still single-agent and I think multi-agent code review is funny.

File: file.png (76 KB, 933x452)

76 KB PNG

lmao

>>108652434
Claude is anti white so why doesn't he use the trailer trash apartheid model grok instead?

>>108652434
trump's assessment is correct

File: file.png (392 KB, 955x718)

392 KB PNG

image-2 coming today, i guess

File: ChatGPT Image Apr 21, 202(...).png (1.5 MB, 1402x1122)

1.5 MB PNG

>>108652641
already have it in my plus account

>>108652654
it's been heavily nerfed from what was in arena.ai, it used to be able to do hyper realistic images of real people now it makes them all with that plastic skin look

File: file.png (2.02 MB, 1536x1024)

2.02 MB PNG

>>108652641
>it even generates meme better than (you)

anyone using claude code with vscode?
i want to drop my copilot license in favor of claude but copilot has this nice inline-auto-complete feature and idk if the claude code extension for vscode also does or not, it's very important for me

>>108652670
Brutally accurate

File: file.png (61 KB, 932x409)

61 KB PNG

REV UP THAT FAST MODE CODEXSISSIES

>>108652814
fuck throw me something to oneshot bros I am to tired to review

>>108652829
nvm they already reset it

File: file.png (426 KB, 1600x900)

426 KB PNG

>>108652829
cure cancer. make no mistakes.

>mfw i'm about to hit my 5 hour limit and it doesn't reset for 2 hours
fuggg

>>108652844
that's last night's reset lmao

File: Screenshot 2026-04-21 105935.png (778 KB, 1000x601)

778 KB PNG

>>108652814
Ugh, my API Codex is gonna get slower now because all you poors are using up your weekly compute rations. Vile grass-eating subscription model creatures.

>>108652829
https://files.catbox.moe/bxdu91.md
some shit i had it write up at 4am that could be interesting to see it actually make real

NOOOO I just bought 2nd account stop resetting everyone is having it good I'm losing izzat point AAAAAA

The thing I'm trying to understand is how vibe coding coordinates multiple agents. Prompting questions and reprompting/tinkering until it works is easy enough, I have done that with some shell scripts.

Woe recommends:
https://github.com/jgravelle/jcodemunch-mcp

And I 100% understand the basic idea here (it's like a lorebook for code, I think).

But trying to understand how you have a virtual company you are managing (all agents) is crazy, how do you coordinate it and keep tabs on it?

I think for my first code I want to make some really basic games. Like the game where you match cards, like one nobody will ever really play, but also then I could make it more complicated eg make it 3D with a rotating camera view.

>>108652899
explain resetting. I see people complaining, but don't know what that is.

>>108652866
don't see shit

>>108652866
Looks 0 bytes to me.

>>108652916
>>108652931
guess catbox doesn't like md files?
https://pastebin.com/raw/L88WDiJh

File: HF1zwyCasAIc-Pc.png (248 KB, 557x794)

248 KB PNG

>>108652905
>how do you coordinate and keep tabs on it?
By building a little sprite world for your agents to walk around and talk to each other while you observe.
https://x.com/d33v33d0/status/2043924104219373609

>>108652915
Codex weekly/5 hour limits are per user, if you don't use it in say, a week or a month, they don't start counting down until you start using it. So everyone has his own 5 hour/weekly reset periods.
Occasionally though, the devs reset the times for everyone, so people whining are mostly joking around, if they knew the reset was coming they could have spent the remaining quota in a big task. My reset for example was tomorrow and I had already spent it all so I just got it a day earlier.

>>108652905
Is this like .h files? Instead of the linker reading the whole implementation .c files, itcan read just the.h and knows what it needs/where the functions are implemented, but for LLMs finding the stuff? Because I was thinking of making something like that, but if it already exists optmized for vibe cosing, then problem solved, I guess.

>>108652944
a VR office building for your agents would be cool

File: Screenshot 2026-04-21 111526.png (45 KB, 1209x624)

45 KB PNG

Rofl this dude's website is formatted EXACTLY like my stockbot's status page was, down to the font. Guess he used Sonnet 4.6 to design his page too!

>>108652976
I ain't fucks wit VR, but that sounds like a neat project for someone else.

>>108652983
The cards are always a giveaway

>>108652936
Looks nice bro. I'd play it.

>>108652905
I'd also play this match cards.

With the ai game studio from last thread, did we just surpass /gedg/ in terms of raw output?

>>108652814
Is this a marketing strategy? Resetting twice in 24 hours to make it seem like you’re being generous? Seems like a joke

>>108653014
>Is this a marketing strategy?
Jesus Christ dawg...

>>108653005
it's a lot more interesting than its first response to being asked the prompt "make a PC game like you're in 1998" where it was basically just the story/visuals from Cyberpunk 2077 with GTA1 graphics/gameplay, i was just bored and wanted to see what that kind of prompt would make it come up with lol

>>108651319
>>108651398
Been using it for a few hours since this. I noticed big pecker does a lot of "actually the problem seems to be X, let me check. It seems fine, let me see Y. But actually it should be X, let me investigate further...nope, all right. So maybe I should look into Z instead. But why would I? Actually it's usually a problem on X! Actually actually actually..."

Dunno if it's skill issue, but he seemed to waste tokens on these kind of crazy loops. Also, on another occasion it kept regenerating something manually and having issues with permissions, sudo, etc, until it got it right... Then had to regen it and did the same shit again. I said "make it into a bash script that regens it". "Sure, *regens it maually anyway, then dumps the script on the conversation and says* if you save this as script.bash you can run it to regen it" "Can *you* save it to a file?" and waste a couple more tokens.

I dunno. Maybe I'm being unlucky. Maybe I'm spoiled from qwen-code which just werked without issues until I ran out of free tokens.

make LLM control a rimworld pawn!
but I'm too busy wagecucking I can't do it now...

>>108652829
Prove the Earth is flat, using quantum physics.

>go back to an abandoned project thinking Claude can help me finish it
>decide to replace complete features of the project bc often Claude writes cleaner code and can't remember how tf some of it works anymore
>run out of free usage
>upgrade to Pro, just to check it out
>now Claude starts hallucinating if the feature is too big and context gets eaten up with small prompts
>it starts forgetting to add closing braces
>this is in web chat btw, with the Artifacts panel
>think Claude Code addon is exactly for this, tried Codex before so I think I know what to expect
>link CC with the same project, ask it to fix a small functionality of a bigger feature it wrote before
>simply tell it what's broken and what I expect
>Whatchamacalliting
>it eats up 70% of usage
>interrupt it
>study the problem and think of a possible solution myself
>use /plan and ask it if it would work
>eats up the remaining usage

did I get duped? I guess I need to adjust my approach but wtf, what's the right approach?
also at this rate, with hourly and weekly usage limits is it even that much bigger than Free tier?

if it fails I guess I will keep using the web chat and keep sharing only relevant files with it, as that worked pretty well

>>108650082 this anon gets it right, Claude nets much cleaner solutions, at least compared to ChatGPT, don't know about latest Codex. it's comments usage is quite excessive though lol
GPT comes up with the most insane approach to solve things, which just barely works. I find it will really cling to something I said earlier or at the start of the conversation and when it starts to hallucinate you have to constantly repeat not to do it, or start over.

>>108653581
>I find it will really cling to something I said earlier or at the start of the conversation
I've noticed that in the website version but in codex not so much

>>108650264
>PMI
The pearson pmi quiz is the worst website I've used. The focus is on defending the CEO from the whiny customers and software engineers, instead of peace and love. Dogfooding is the only way to make sure the devs don't shit on the plates.

why is sam so much better than other AIjews

>>108653613
Yeah, I'm not particularly interested in PMI or certs or anything (although vibing courses aimed at cert-takers sounds like a good market niche), I just want to understand the process from idea to finished product so I can better hand hold the AI on the other steps. Because the code monkey step which I know how to do myself I can hand hold it, the previous steps I have no clue. When I can turn my idea into an architectural skeleton I can vibe it and then flesh it out until it works. But I'm bad at turning the idea (usually from a user POV) into the implementation architecture from dev's POV. Some friend said I needed to study project management, but idk, hence me branching to SDLC as well, before realizing I was struggling with vibing the courseware as well, and the "brilliant" idea of vibing a course on vibing. Let's see where it goes.

>>108650256

Current round – 4 votes
we  **--
and *---
scum *---

You don't need the bars to go up to total votes as their max. That would give bad resolution if you just had a zillion options with few votes each. It would be better tohave the max on the bars to be how many are rhere on the option with the more votes.

>108650256
I like it.

Alright I ran through my quota. Time to do chores until the reset hits.

>>108653715
Altmans are the Cool Jews. His relative Robert was the most underrated movie director of the 20th century and literally never sold out. If I am wrong about him being related to Robert fuck you I don't care I just talk shit.

>>108653915
Did he just steal Bard?

Some anon said vibe coding is good for hacks and exploits if one can get past the guard rails. Has anyone used it for rom hacking/translating ero games?

>>108654137
Yeah you just tell it you're engaging in some Red/Blue games, or you're trying to harden your own security, and it'll be glad to help.

>>108654175
Yeah, my question was less about how to break the guardrails and more about whether anyone has actually applied it to translating obscure j games. I was just looking at some vidsloppa about them and got curious because if I had the time, skill and tokens, I think I could try it.

>>108653915
No relation between the two. Robert Altman was a second generation Americanized German "Altmann", a German and an Irishman, raised Catholic. Sam Altman is 3rd or 4th generation Polish and Russian descended Altman/Gibstine/Schreiber, a true Ashkenazi.

I work at a water treatment facility and I just hooked up Hermes to the pressure valves at work lol
AI is advancing fast

File: 06c375b166056bfd4a2eb382a(...).jpg (63 KB, 622x508)

63 KB JPG

>>108654205

What in the fuck Github Copilot. The new hourly rate limit is a little annoying, but even with that I hit the new weekly limit in one fucking day.

>>108654235
Too many people abusing the old system. Now on top of a worse harness with these limits it’s just inferior to the big guys

>>108654308
>Using a thing as intended within the agreed upon terms is abuse.
I think what you meant to say was "they weren't making enough money," or possibly "any money." Labeling normal use as "abuse" is bootlicking. Is playing a videogame on your computer abusive to the computer? Watching TV on a TV, is that abuse? Don't be such a faggot, faggot.

>>108653581
at work we have a lot of instructions to agents to not put comments in
no really, don’t put comments in

>>108654375
Let me guess, if you see a bowl with free candy you feel entitled to all of it

>>108654375
>normal use
Are you retarded or just completely unaware of what people were doing that caused this?

~45 minutets to openai stream.
wouldn't be surprised if reset happens around then.

>>108654427
A better analogy is more like going to a shop, paying for 10 items of food, and then people getting mad that you left with the 10 items you paid for instead of leaving 5 on the counter and leaving with the other 5.

>>108654427
There was no bowl of free candy, I paid for a service and used it entirely within the agreed upon terms.
>>108654432
You are completely unaware of what caused this, you are not one of the Microsoft employees that helped make this decision.

You're both retarded faggots.

>>108654489
Not really. If you don’t realize how what people were doing was abusing a system to do something it was not meant to do youre being either disingenuous or simply retarded.

>>108654512
And they changed the terms. They didn’t ban you for doing that or anything so just like you were doing something totally reasonable they’re doing something totally reasonable as well. Meanwhile people like me get fucked because of people like you.

>>108654525
You're getting fucked by Microsoft and you're assuming you know the reason because you learned about sub-agents yesterday. You're a disingenuous retarded bootlicking faggot and you're giving Microsoft the benefit of the doubt. You are human shit smeared across a keyboard.

alright everyone reset is here stop burning tokens

>>108654540
FUG

>>108654512
>You are completely unaware of what caused this
If you aren't, you weren't posting in /vcg/ until today because half of this thread has been abusing the flaw that caused this change to get unlimited free tokens for weeks.

>>108654540
Another reset? I should have burned more tokens.

>>108654539
Yes it’s totally a mystery why they did it surely you would have to be Microsoft’s ceo to know the otherwise it’s impossible to know why they changed the terms so that people couldn’t get practically unlimited usage surely that had nothing to do with it and it’s a total mystery

oh wow, a reset happened and I _was_ burning tokens big-league

>>108654556
You retard, subagents are a well documented feature and people have been using them with Copilot for far longer than the 3 days I've been posting about them. Know where I got the idea to use them to save requests? The documentation. You are a fucking moron, and you cannot possibly back up you're retarded room-temperature-IQ take on the situation, sit down Raj, your curry is getting cold.

>>108654595
Just admit you were abusing a system and that you weren’t the only one anyway. I bet Indians were doing that by the millions for that juicy izzat of knowing they were cleverly cheating an honest system