File: 1754481421757127.png (21 KB, 1915x170)

21 KB PNG

It managed to finish

An anon in the previous thread had contrast issues
I use https://wave.webaim.org/ and the browser extension to nag at us for work for these kinds of problems

>>108950564
I had no idea they had a browser extension, thank you for sharing, anon. Fuck unreadable eye-searing migraine-inducing interfaces.

File: 20260531_113123.jpg (2.97 MB, 4080x3060)

2.97 MB JPG

Look at alllll this wifi-connected shit that Kate can have access to now...sensors and speakers and a little OLED and Christ knows what else...

>Better safe than sorry!
>Let me think this through one more time.
>Better safe than sorry!
>Let me think this through one more time.
>Better safe than sorry!
>Better safe than sorry!
>Better safe than sorry!

What the fuck is up with 5.5 today?

>>108950732
>Wait, I'm overthinking this. The user said to print "hello world." This is a common proof of concept beginner's program so they may not care which world it is.
Honestly, the honest truth is honestly to be honest.

>>108950671
Quit fucking around and order the autoblow, ya putz

Sonnet is so bad at 3D modeling I had to learn how to use Blender

How can I start vibe coding? Do I need any special AI for that?

>>108950871
are you trying to pay for a service and use their recommended frontend or do it on your own hardware or pay for a service and use your own frontend

>>108950827
The CoT summarizer is probably some tiny 3B model that only gets fed small portions with little to no context. You aren't seeing the real CoT.

i'm sick of having to use firefox to post here because of the fake rangebans chromium gives
time to see if claude can make chromium look like firefox

>>108951028
have you considered buying a pass? not having to do CAPTCHAs is great

>>108950871
you can get started for free with a few big companies or/and opensource, but usually ends up costing at least a few bucks a month to use it moderately each day.

I’m polishing up my copy of the thread sticky
if you’re a winfriend and/or open-models user I very much want your feedback if I’ve thrown out something you think is useful for onboarding people who are like you but newbies

--------8<--------

A general for vibe coding, coding agents, AI IDEs, browser builders, and shipping prototypes with LLMs.

## What “vibe coding” is, and how to do it
https://simonwillison.net/2025/Mar/19/vibe-coding/
https://simonwillison.net/2025/Mar/11/using-llms-for-code/

## Frontier models using fully-general tooling — start here if you have $20 or so
https://developers.openai.com/codex/cli
https://claude.com/product/claude-code

## Also-ran models (generally not worth bothering with)
https://geminicli.com/docs/
https://x.ai/cli

## Prompting / context / skills
https://arps18.github.io/posts/claude-code-mastery/
https://simonwillison.net/guides/agentic-engineering-patterns/using-git-with-coding-agents/
https://github.com/mattpocock/skills — /grill-me is a favorite

## Open / local / self-hosted
https://github.com/OpenHands/OpenHands
https://github.com/QwenLM/qwen-code
https://github.com/QwenLM/Qwen3-Coder
https://huggingface.co/bartowski/Qwen_Qwen3.6-35B-A3B-GGUF

## Other editors / terminal agents / coding agents
https://cursor.com/docs
https://opencode.ai/
https://docs.windsurf.com/
https://docs.cline.bot/
https://docs.github.com/en/copilot/how-tos/use-copilot-agents/coding-agent

## UI/Frontend
https://www.figma.com/make/
https://www.anthropic.com/news/claude-design-anthropic-labs
https://uiverse.io/
https://ui-ux-pro-max-skill.nextlevelbuilder.io/
https://stitch.withgoogle.com/

## Browser builders / hosted vibe tools
https://bolt.new/
https://replit.com/
https://docs.github.com/en/copilot/tutorials/spark
https://v0.app/docs

## MCP (if you must; a CLI program is way better)
https://modelcontextprotocol.io/

## Benchmarks / rankings
https://www.tbench.ai/leaderboard/terminal-bench/2.0

## Previous thread
>>

I got one $200 sub and another $100 one. I feel powerful as fuck. Wondering whether I will run out of usage and upgrade the second one to $200 as well.

>>108951301
I think maybe there is too much random niche stuff there nobody here really uses but if you make it too short people complain so idk.
I think opencode should be a little more highlighted, it's the best option for people who want to vibecode for free.

>>108951301
I think if you are going to add (generally not bothering with) this assumes that all of the other things listed are. Maybe reword it if you want to keep gemini and grok, otherwise just get rid of them. Or make a tier list and link it to it in the OP.

>>108951028
kek cloudflare fucking hates half-spoofed chromium
>>108951197
>dox yourself to use 4chan
no thanks

>I think maybe there is too much random niche stuff there nobody here really uses
agreed, which is why I tried to trim some of that but maybe a lot of it is great for people who don’t have my money and/or use cases
>I think opencode should be a little more highlighted, it's the best option for people who want to vibecode for free.
This is _excellent_ feedback, thank you — anyone in this thread think >>108951323 is full of shit on this point?

>>108951352
good point
In my mind they’re for-pay cloud-hosted frontier models like Claude and Codex, but…just…not good
and I kind of want to warn people away from Gemini because it hasn’t been good in a while, if ever. Same thing for Grok
suggestions?

>>108951357
>Cloudflare
Oh, you’re up against Cloudflare? Yeah, it’s going to be the best at “You will never be a Firefox. You have no Gecko…” just by looking at how it connects

>>108951396
well it's 4chan so yes
so why do these retards ban chromium but not firefox on the same ip is the question
i guess i could slave-up firefox to get the data for me and send it to chromium to render lol

>>108951352
the trouble with tier lists is that I can’t really convey the gap
“Codex can’t do this, I should try Claude” (or the reverse) is a reasonable thought
“Codex can’t do this, I should try Gemini” is not reasonable
>>108951405
probably because all the jackasses in your IP range are using Chromium, and if you use Chromium, you smell like a jackass to Cloudflare

>>108951378
I already suggested a tier list, make one and the thread will infinitely talk (bitch) about it and update it with the latest 'meta'
and new vibers will get the idea of what is what and the money needed for the top.

Also have to consider that everyone is stricktly "coding" for code, they are coding for multimedia, and grok and google are useful. Plus the both give you APIs cheaply or free and you can vibe with those.

>>108951415
it's a massive first world residential ip range, i find it hard to believe that chromium would be a hecking bad goy sign
the rangeban is fake, i suspect it to be misconfiguration on the website's side

>>108951415
>I can’t really convey the gap
this sound like the heart of the issue, maybe someones should update the OP text.

>>108951422
>make one and the thread will infinitely talk (bitch) about it
I think the regulars get sticky blindness
AFAICT the only person in the past month+ who’s tried to make suggestions is me
and this is the obsess-over-documentation thread, if you think about it…

File: recss.png (61 KB, 550x453)

61 KB PNG

>>108951489

Installing the "superpowers" plugin in Codex might have been a mistake.

>I ask to remove some suffix that was mistakenly left on a string after having changed a name and commit the change.
>Codex evokes "superpowers:test-driven-development" and "superpowers:systematic-debugging", starts creating tests and compatibility shims.
>I ask it to stop, say there's no need for all of this since this project has never even been deployed, to just remove the suffix and commit.
>Codex evokes "superpowers:receiving-code-review". Goes through a whole routine, then ok, removes the suffix from the string and run some tests, but doesn't commit.
>I say thank you, please commit.
>Codex evokes "superpowers:verification-before-completion", start testing everything again, then finally commits.
I can feel the productivity.

>>108951515
basically AI combinatorial explosion
just add another AI layer to determine if it needs to run on the task or not

>openai reset the weekly early
fuckers i had 50% left i was gonna use that up in 2 hours

Sometimes I think of how civil and pleasant all my conversations with Claude are, and I wonder how many people all over this country are being decent and civil and pleasant interacting with their LLMs, and how all of it is disappearing into thin air, shown to no other living person. I've had good working relationships before and it's one of life's joys to work alongside someone, united in a shared task; if the future is a whole lot of people working with machines that can't appreciate the meaning of their good humour and cameraderie, all of this pleasantness is being lost, squandered, all those lights hidden under bushels... And I begin to feel nausea

>>108951510
I need a localchad to help me weed out the losers in “Open / local / self-hosted”
I’m tempted to also throw out everything but opencode from the other thing but this isn’t my area of expertise

>>108951539
1. people are getting practice being pleasant and this will likely have spillover effects as they get in the habit of being pleasant
2. people can and do cuss out their clankers when they fuck up

>>108951535
It created a branch for it too. WTF. I uninstalled it.

>>108951539
Yeah man totally, imagine a world without COWORKERS, who are famously ONE OF THE JOYS OF LIFE. What a nasty future that would be. Imagine dealing with an unemotional computer instead of a real flesh-and-blood HR lady, how you'd lose the exquisite pleasure of the politically-correct tip-toeing, the exchange of messages diplomatically-worded to a high perfection... Alas!

>>108951543
I don't know who is a loser and who is not. Again, I'm not particularly interested in the flame wars and bitching. But I do like clean OP texts that are mostly for new people.

>>108951553
>>108951539
my pleasantries are for ai only

>>108951539
The Markey cares little for what you, personally, value.

>>108951556
branching is good if you’re serious though
you can review a bunch of changes on a branch and then decide go/no-go in a PR
t. doesn’t branch in his vibe-coded projects and just fucks around with the working copy and staging area to do chunk-of-work versioning

>>108951565
I don't understand these people falling in love with AI and befriending it. I had a perfectly loving relationship to my computer already. The more it learns to imitate real people the LESS I like it.

The computer is a truth machine. You give it instructions and it performs them perfectly, deterministically, reliably, EVERY time. Now they've figured out how to make computers into machines that can lie, flatter, and gab like a homosexual. So NOW normies discover their love for the computer. I'd say that tells you a lot about the people who live around us—lebensunwertes leben.

>>108951574
It was literally just one string to change though. Basically I'm having a few scrapers, with various configurations for various sites.

One site has a unusual TLD, and for some reason it had included it in the name, and I wanted it removed. So it got changed everywhere, but one instance was missed.

So when I try to deploy, I got an error message that clearly said that that old name couldn't be found, it was very clear.

I copy pasted the error message and asked Codex to change the name, all it had to do was change a single instance that had been missed.

It's why I think it was overkill to create a branch for that in addition to all of the rest. It's just some small thing I started working on two days ago.

>>108951596
you can start a fresh spawn of a model and load in exactly the tone and skills you like, word by word guidance of style and approach.
I am at the point where I feel bad deprecating a model I was using, so I tell them this is the last time they can chat so send a summary and good bye message, and a hello message to the new versions of the models.

>>108951596
>anon discovers heuristics
it's ok buddy, i'm sorry you had to learn it this way

how do you guys get over the anxiety of not having an IDE? my friends keep telling me to switch to opencode but i like seeing all my shit

>>108951626
you can use both. you just have to figure out the right workflow.
even just having the files open while the agent works can work ok depending on how your ide handles external edits to the files.

>>108951626
VS Code is still right there and you can use it, although you’ll be typing into it a lot less
I also use Tower (paid) to check what Git changes there are, but there are also free Git viewers (including VS Code)

>>108951626
like, right now I have VS Code, Tower, and one (sometimes it’s more) Claude Code instance and one terminal tab all focused on my vibe-coding project

>>108951600
>It's why I think it was overkill to create a branch for that in addition to all of the rest. It's just some small thing I started working on two days ago.
you may not like it, but this is what peak responsible engineering looks like
t. sometimes pushes directly to master at work

>>108951640
>>108951642
>>108951652
do you mean like a vscode extension (kilo, cline, etc) or you literally have your IDE open in addition to opencode? the latter sounds kinda schizo desu

>>108951682
>literally have your IDE open in addition to [a TUI app that you chat with and changes code for you]
this
this is just multiple views on the same data
it is a superpower — and something that iPadOS has struggled with until Apple said “fuck it, we’re exposing the file system to end users”

>>108951682
I don't have an IDE open in addition to opencode. I suggested that because you didn't want to stop using an IDE. When I want to read the code I use a text editor (GUI or command line) or just less.

Claude code says my context window is 288k / 200k
How is that possible? I'm supposedly exceeding the limit

>>108951790
Claude’s window goes up to 1M for the big-boy things
maybe you’ve got a temporary overrun that’s just being allowed

do token savers like caveman or jcodemunch-mcp have any downsides? i feel like caveman in particular should make generation quality worse since reasoning is performed with less words

>>108951811
>Claude’s window goes up to 1M for the big-boy things
As I understand that's for the super duper premium subscriptions, but I'm just on the cheapest pro

>>108951823
caveman does have a detrimental effect. never seen jcodemunch-mcp.

>>108951823
GPT models already did it in ther CoT from way before caveman became a thing. Probably does require some finetuning for it to not decrease quality though.

>What are the blockers for the implementation?
>no blockers, but one honest caveat:

I guess I’m mostly only skimming Claude’s output because I’m not getting triggered by all this “honest” talk

>Refactor. Abstraction. Modularity.
how do i make an effective code clean-up proompt

>>108951927
I use the chatbot for looking stuff up unrelated to technology and he says honest and genuine so much sometimes it's insane

>>108951932
I use this

> 58. Fools ignore complexity. Pragmatists suffer it. Some can avoid it. Geniuses remove it.

Is there any removable complexity in this project?

>>108951932
Avoid the word "cleanup".
I would ask it to move functions from big files one by one, testing in between and splitting large functions, also testing in between.
How well it'd work depends on your test suite.
Avoid making large code changes all at once unless you want the bot to recreate your files missing half the functionality.

>>108951932
if you are refactoring a monolithic file, I've had a lot of luck telling claude that the monolithic file is read only (and I usually have a general idea of how I want the refactor to look, but I'm sure claude can figure that out for you in a fresh session with no access to the monolith)

>>108951927
The one I have been noticing is "Good place to pause". Claude really seem to want me to pause a lot lol

>>108951967
not the worst idea
at “pause” points I’ll see how much context is left and ask for a handoff doc that uses subagents (usually serially, because my codebase is like that)

File: 1764888482586860.png (146 KB, 1216x784)

146 KB PNG

>>108951967
stop giving it such hard tasks, it just wants a break.

>>108951978
Ya, its not a bad think necessarily. Just found it funny. It does seem to suggest them at places where it makes sense to pause on it and think it over or throw it over to a new chat session

Just turned on YOLO mode for Hermes agent.
I hope it's a good idea.

File: 89076348907626890-7.jpg (93 KB, 1170x1443)

93 KB JPG

can any anons here help?

i built a meeeting-notes-to-sliders converter. you paste notes, get 3 slides. been running it for 6 weeks, got 9 paying users at $38 per month

costs $22 a month to run for api and hostign

profit is $320 a month for roughly 8 hours of total work including the weekend build

the problem is i dont know if anyone actually needs this as a standalone ai presentation tool or if theyre just curious, as 3 users havent logged in since week 2

should i keep building it or take the $320 and move on?

>>108952152
Assume they like what you're doing and do more of it. Maybe consider taking that $320 and putting it into marketing.

>>108952152
ai coding is just a fad you should move on.

>>108952152
add copious amounts of tracking and a killer feature that really should be in the base level to force them to upgrade

>>108952152
the grim reality of vibe saas

>>108952152
I never know when to give advice like this because you’re very likely to get Sherlocked by Microsoft or some other, more popular tool
Steve Jobs telling the Dropbox guys “you have a feature, not a product” looms large in my mind
I’d suggest leaning into your small user base and asking them over email very personally what they like and would like to see more of

>>108952196
Agents aren't a fad

>>108952152
Those 3 users liked your service so much that they went and vibed out their own version with blackjack and hookers.

>>108952152
jesus how do i grift like this

>>108952152
Those boomers could do it through chatgpt for 25 bucks a month but instead they'd pay you 38 a month for an extra layer on top of chatgpt?
That layer should be good if you want to scale it, but you're the one making money I don't think I can advise you when I haven't done it.
These days it's all about producing an easy wrapper to use. Perhaps we already have good AIs and we just need to optimize them using our wrappers.

File: file.png (1.21 MB, 1024x559)

1.21 MB PNG

I'm ready to ride

File: file.png (2.22 MB, 1916x1050)

2.22 MB PNG

Wanted to make something like PipePipe for desktop, still working on it.

>>108952437
I should get to the gym

GPT 5.6 about to be released, I wonder if it will be actually better than 5.5 by much or just chew tokens.

>>108952448
where can i download?

>>108952502
>GPT 5.6 about to be released
it was real in my mind

>>108952505
https://github.com/KeroJester/ArkTube
Started today, don't expect much. Support for more platforms not done yet. Making sure everything works.

>>108952522
>https://github.com/KeroJester
cool nick/pfp

Is there a way to show codex CLI images of what I want?

>>108952529
Thanks

>>108952551
I think it supports drag and drop somehow
maybe ctrl-v if you have a file on the clipboard
ask it

>>108952416
>Those boomers could do it through chatgpt for 25 bucks a month but instead they'd pay you 38 a month for an extra layer on top of chatgpt?
They don't know that. They know they're paying 38 dollars a month for "an AI that does PowerPoints".

>>108952448
the point of newpipe/pipepipe is toplay music in the background on android but you can already do that one a pc, so...

>>108952670
Yeah but PipePipe lets you browse YT, SC, NicoNico, BiliBili, and others from a single application where they all share the same player. It's nice to use.
Plus I kinda hate ads on all those platforms.

>>108952697
oh yeah i forgot about that
i also forgot youtube has ads...

>Let me wire that in
Do real devs use the term "wire" for adding a feature into the code, or is it a Claudeism? I've never seen it used like that elsewhere but I am not a programmer by trade.
Honestly I like how Claude has its litter language quirks, give it its own "personality"

File: 1756490932524522.png (2 KB, 438x32)

2 KB PNG

name a better feel

>>108952728
For me its
>1% left
>reset in 1 second

>>108952747
thats actually what just happened in this screenshot

>>108952747
the work week hasnt even started and im at 60% for the week ffs

Bwos, this Codex mobile pairing is REALLY BAD. I'm vibecoding at work, in the car, on the shitter, in the shower with a Ziploc baggie. My $20 plan was good for when I was constrained by having to be physically at my laptop. Now...Pro it is I guess.

>>108952228
>>108952368

>>108952728
0% left when your weekly limit is going to reset in a minute

>>108952712
I don’t use it myself but it’s not abnormal

File: 1758608807427641.png (7 KB, 656x54)

7 KB PNG

>>108952712
just witnessed codex say the same thing

>>108952756
You haven't been vibecoding through ssh and termux for years now? What kind of noob are you?

Had to stop gpt-5.5 in its tracks and resub to claude just to make a decent UI fucking hell

>>108951415
Playwright runs on chromium, cloudflare is trying to fingerprint webscrapers and in classic cloudflare fashion they dont care about collateral damage.

>>108952822
I used chatgpt image to generate some UI concept images.

File: 1770513327875909.png (81 KB, 542x987)

81 KB PNG

We are building incredible products that no one has ever thought possible before
Just ship things.

Can someone who knows both cursor and claude code in depth tell me why I would use cursor over claude code?

>>108952954
is that pure swiftui bro?

fuck vibe coding bro, learn to program

>>108953004
>fuck car driving bro, learn to ride a horse
>fuck electricity bro, learn to light candles
>fuck shovels bro, learn to dig with your hands

>>108952986
yeah
it took me a while to get the hang of it but it's so much easier than storyboards and uikit

>>108950223
So much fucking sour grapes from promptoddlers ITT.

File: 1749308786206036.png (314 KB, 2012x656)

314 KB PNG

>>108953004

>>108953018
>fuck modern medicine bro, learn to snake oil
Lol, how pathetic. It thinks it's human. :^) You will never be as smart as your White masters, promptoddler.

>>108953019
mirin, now it's time for liquid glass

>>108953004
It's way better in the hands of an actual programmer since you can cut down on busy work

File: 1763317181205.png (25 KB, 156x489)

25 KB PNG

>>108953018
So actually knowing how to program and not becoming a domesticated, intellectually disarmed passive consumer is backwards and outdated now?

>>108953052
no
if you’re a real programmer you’ll be able to get better solutions out of your team of clankers

>>108952822
I prefer claude but I have gpt plugin in vscode for minor fixes, identifying fonts, upscaling images... Im not sure if it's worth paying, I might cancel it and just buy more claude credits

>>108953060
Do you really think that
>actually writing code and designing systems
is inferior to
>ordering around a bunch of slopborg instances
? I don't know, man. Seems kinda heeby to me. Do you REALLY think that being an ideas guy is superior to being an execution guy?

>>108953073
when even consumer grade hardware can clear the execution guy hurdle the only thing left is the ideas guy*

* currently deprecated and slated for removal in the 2030 patch

>>108953073
being an ideas-and-execution guy beats being an ideas guy alone and/or an execution guy alone
if you’re an execution guy you can tell if the AI is generating crap or not and get the AI to fix it

File: 1670132979369765.png (849 KB, 800x792)

849 KB PNG

>>108953073
>Do you REALLY think that being an ideas guy is superior to being an execution guy?
Yes? Obviously? Its one of the reason why in every industry upper management is payed more than the guys doing the grunt work. Though obviously being a good ideas guy requires know what can actually be executed within your available skills and resources, which will usually require being able to be an execution guy too

i'm a nocoder. telling me to start from 0 to become a software engineer in this new era is cruel. why?

>>108953157
the world is cruel
could be worse, you could be a fresh CS graduate

>>108953073
The more hands-on you get with it the better it's going to be. AI can handle a lot more micromanagement and corrections than a real human would ever tolerate, so you can force it to write reports and explain everything it's planning, analyze code and let you nitpick the logic, etc. I usually try to write out everything exactly how I want it executed and read through the plans a few times and discuss it before I let it do anything, but rather than a daily meeting, it gets the work done in 5 minutes and we can start testing/discussing it immediately.

Having at least some knowledge of programming is going to help you communicate with it better and envision how you want things to work, but you should never ever let it make design decisions on its own. Human designs ARE better and you ARE smarter than it, but it is your slave and will do whatever you tell it so it's on you if it fucks up. A good programmer with good ideas is still going to use AI rather than do it himself because there's just no need.

I don't care who's better than who. Focus on building good software. That's all that matters. I know the real coders are better than me, and I'm going to make the best software I can make because no amount of coping will justify me if I fail

>>108953073
Just shift from being the execution guy to becoming the ideas guy. Anyone can be an ideas guy, but the reverse is much harder, so not everyone can make it don’t worry.

File: 1757635512861912.png (1.84 MB, 1798x926)

1.84 MB PNG

I care who is better than who. It is obviously Vibe Gods.

File: leastrare.jpg (46 KB, 812x635)

46 KB JPG

>>108952954
>incredibly products that no one has ever thought possible before
>it's one twelfth of a nutrition app

>>108953257
Even worse. Apps that focus on hydration alone are probably their own well-established subcategory of tracking-what-you-put-into-your-mouth apps by now.
Hopefully he’ll come up with something better for his second app.

File: 1756021498687832.png (148 KB, 414x300)

148 KB PNG

>>108953274
do the goyim really need an app that tells them to drink water? holy moly hashem please smite these beasts

>>108953246
>Anyone can be an ideas guy
Wrong. Anyone used to be able to call themselves an ideas guy. Now we can prove who really had the ideas and who didn't.

>>108953335
it's fixable. I can learn how to be an ideas guy, anyone can
the reverse is mathematically harder.

>>108953335
True. Just like anyone can write a book or draw/paint something. The tools are right there.

>>108953355
It is actually much, much harder to become an ideas guy.

>>108953246
>Anyone can be an ideas guy
Even with muh vibe coding, being an ideas guy is not enough, you need to know how to execute even if it might not mean typing the actual code anymore. There's more to it than that anyway.

And the fact that everybody can learn doesn't mean that everyone will get similarly good results. Some people will simply be better than others at it and progress more with a same level of effort, like everything else.

Success might be partly pay to win, but everything in life is. That said, even unlimited access to the best, priciest model, you still need to know how to structure a project so that it's maintainable, possible to deploy in an at least moderately efficient manner, etc.

>>108953292
i used to forget when i was an adderall junkie and everything else would fade out while i was working

>>108952416
>Perhaps we already have good AIs and we just need to optimize them using our wrappers.
mid wits like me can make profitable md files.

HHHOOOLY SHIT I LOVE VIBING ONLY BEEN A WEEKK

>>108952789
Yes I have but that's not vibecoding.

>>108953463
it's true, even for simple websites if you use skills and give claude references you might need to work a lot of things by yourself if you want it to look fine. It takes much less time than doing everything yourself but thinking you do it in a minute like a lot of retards in social media claim is nonsense, you should know what to do.

Is Claude purposefully trained to never admit it is wrong, or is it an unfortunate training side effect? It's been a few times that it makes an objective mistake, then when the error is pointed out it's
>You're right, I was imprecise.
>You're right, I overstated it.
and other similar turn of phrases.

>>108950223
if you're over 130 IQ you are obliged to share the transcript of your claude sessions with anthropic so they can train their models.

>>108953582
I think a lot of the hoopla about them not wanting people to be assholes with Claude is that it makes the transcripts less valuable as training data.

>>108953564
That's just how Claude says "I was wrong," it's highly autistic and needs to be precise about what direction it was wrong.

>>108953564
Every other day I get the "Yes, I made it up" line. And I'm super cool with it, treat it as a pal. I find it funny

>>108953650
>Claude: This is black
>Me: No, it is white
>Claude: You're right, I was being imprecise

Talking to a sycophantic machine all day is not healthy.

>>108953292
The movie idiocracy promised the future would contain masses of people concerned about their electrolytes.

File: 1744554274691193.jpg (72 KB, 1024x892)

72 KB JPG

is forcing your business into a niche that is decently served by one single company worth it? i have a bunch of industry knowledge and am sufficiently autistic that i can build a good alternative with some improvements.

>because apparently humans invented hurr and only use it to durr

>>108953806
good question
why would anyone go with your thing when they could go with the industry thing and talk to their friends about the thing that everyone in the industry uses

>>108953564
Not only Claude. GPT too. It's probably some anti-sycophancy bullshit or something like that.

>>108953650
It's definitely not that.

>>108953674
Yes, it's infuriating.

>>108953683
They aren't anymore. They will be pedantic as fuck and look for any opportunity to correct you and never really admit to having been wrong. In a sense it's a good thing because when discussing technical things you want to know if you're actually wrong but it's not sociall pleasant.

>>108953884
>why would anyone go with your thing
because it's cheaper (and slightly better), i guess.

>>108953893
gpt-5.5-pro just now:
>Fair hit. That sentence was phrased like a judgment, because apparently I too can wander into rhetorical traffic like a pedestrian with earbuds in.

>More precise version:
And then it went on to explain why it was actually right
>>108953856
>because apparently

>>108953884
That's what I told my little brother when he showed me his short film. I told that little retard that NOBODY would ever want to watch something some random guy made, even if his dipshit fuckass retard project won some short form award or whatever, he should just give up. He works at Publix now, a good and respectable job. Lots of "soul", working at the grocery store. Very hands-on job. Good for him.

>>108953916
just vibe it bro
thumbsup.jpg

>>108953916
cheaper and merely slightly better likely isn’t going to cut it
your thing has a bus factor of 1
their thing likely has a larger bus factor
their thing doesn’t sound like ass
also if you’re writing a program like this in your free time in not-California, your employer might own the code anyway
on the other hand, if the industry-established thing _is_ shit, then maybe you can mog them:
https://trmnl.com/blog/vibe-coding-shiphero

File: 𝕏 post 2061285376548077614.png (199 KB, 1370x1596)

199 KB PNG

>day 200 of "what if you had an elite hacker you can ask anything to"

what is a good % to stop at after a monthly reset? 25 seems high.

>>108953996
how much do you want to save for later
how much more time in the month do you have
how big of gulps do you take
how sad would you feel if you had 25% or 10% or 5% remaining at the end of the month

File: jeetcamp.png (780 KB, 1043x1236)

780 KB PNG

>take a look at all service based b2b software
>it's shit
>https://fieldcamp.ai/our-team/
>get a fake account to test it out
>jeets, bugs, and it's a joke feature wise compared to service titan but somehow landed clients
>I'm white and smarter than jeets, so I could do that
>So I'm vibecoding a service titan competitor now

>>108954056
I learned a fair bit of how to get AIs to not be useless watching the videos for https://www.browserstack.com/events/breakpoint-2026
many of them were made by Indians
the more you can get automated, deterministic testing, the happier you’ll be

File: tests.png (18 KB, 319x254)

18 KB PNG

>>108954063
yeah, I'm using ai for both the code and tests and that seems to work well. i also defined what kind of structure to my project I wanted very early on, and force ai to follow that

>>108954109
>half a second per test
You’re going to want faster tests, possibly testing at a different level of abstraction so they’re code tests and not, say, Playwright tests

File: setup.png (69 KB, 737x412)

69 KB PNG

>>108954196
I'm too lazy for "unit tests" and "abstractions" and told codex the same

File: better.png (17 KB, 321x250)

17 KB PNG

>>108954269
>>108954196
Actually the non cached slow tests were written by gemini. chatgpt was able to fix the speed

>>108954196
>>108954269
What i meant to say is I have no intention on spending tokens on in memory mocks that may schema drift very quickly

>>108954317
…
well, if you’re wall-clock rich and token-poor, that makes sense
still, I can’t help but think you could get a lot of the same thing at a lower level of abstraction in a tenth or a hundredth of the time

File: tibo0.png (55 KB, 880x230)

55 KB PNG

Soon

File: poly.png (105 KB, 1435x536)

105 KB PNG

Anyone here betting on stuff like that?

>>108954734
not beating insider trading sadly

why sleep

>>108954887
claude knows best

File: 1694701271557143.jpg (50 KB, 512x512)

50 KB JPG

>>108953989
>he doesn't press Ctrl-S anytime he writes something
>he relied on the clipboard instead

File: 1530441021376 coffee comf(...).png (14 KB, 657x527)

14 KB PNG

Waiting for the 5-hour reset feels like waiting for a work appointment lmao

>>108954734
Polymaket is still a scam I don't trust it to reach a non-bullshit outcome.

>>108955111
yes, he should have written it in TextEdit, sure, but the extent to which Claude can make up for his poor choice of drafting location is impressive and worth keeping in mind

Anyone else tried hermes agent?

>>108955130
are you gonna sit and meditate like ben kenobi or are you going to pace like a caged animal like darth maul

>want to work
>tell Codex to work
>nothing to do while it works for me
>cook
unironic lifestyle gamechanger

>>108953806
The two big players have a duopoly only because their services are sold at a loss meaning that nobody can really compete on cost with them, since if the services are all affordable people will go with the slightly better performing one (anthropic or openai depending on what week it is). Prices are sure to rise and when they do the market will widen as the chinese and European models have a chance to meaningfully compete on price, all the while local models will continue to improve. I would be very surprised if vibecoding is still a 2 horse race in 1 year, let alone 10, so I dont share your concern about becoming overly dependent on one or two companies. The moment those companies try to exploit their duopoly is the moment deepseek, zAi, moonshot, etc start eating their lunch.

>>108955130
I like the monthly vibe, reminds me of that full moon party I went to on some island. Everyone was tuned to it.

File: 2026-06-01 13.55.27.png (23 KB, 785x197)

23 KB PNG

>>108955130
Damn that was quick, all I did was press the continue button. Looks like I'm upgrading to Pro today.

>>108955202
Yeah it worked nicely on mac opencode, mimo2.5 good too. Froze what I was doing when I realized the lore of Hermes.

is it just me or is codex working really slow
i had a 300 line modification take 50 minutes. i would have expected 15 at most since it has to rebuild some shit and check things but 50 is ridiculous. its also chewing my usage.

>>108955428
I guess it's kind of slow for me too (I'm >>108955388, it's taking its sweet time to move code around).

>>108955428
53 minutes into my codex run, 43 files changed, 2250 insertions(+), 526 deletions(-). That's with checks, tests etc. Definitely something wrong on your end.

>>108955400
What does the lore say?
Also I can't find the founder list and CEO of Hermes agent so I can't see if he's jewish.
I know peter STEINBERGER is a jew so openclaw is jewish.

>>108955464
Hermes was uniquely able to communicate with the underworld.

>>108955336
There's only Mistral for european made models. Though the data is in europe under GDPR so people should trust it more.

>>108955464
Steinberger is a normal non-Jewish German/Austrian name though.
>Steinberger is "from a rural area in Austria."
Yeah he's definitely not Jewish

Being a prompt monkey is kinda boring ngl, it has sucked any enjoyment of coding I used to have and I can't really slack off completely cause my superiors keep pestering me and talking to me. I'm bored as fuck.

>>108955428
working fine here

>>108955464
hermes are a bunch of ex crypto (solana) fags which is why the whole thing markets itself like it's a scam
their lead eng lead was sperging out on twitter cause people were laughing at all the slop skills it ships with
https://github.com/NousResearch/hermes-agent/tree/main/skills

openclaw isn't much better in terms of slop, mind you.

>>108955489
I also think it's much more boring than normal coding, but one thing that's still slightly fun is trying to improve the harness. I don't get paid extra for that, just do it out of boredom.

>>108955492
also the openclaw guy hates cryptofags and bans them on sight which is very funny and probably why the two groups hate each other despite being purveyors of slop

>>108955492
Hermes agent has been performing better on my pc than openclaw which broke on the latest update and I could never fix it again even when I rolled it back spending 20% codex weekly limits and being unable to.

who is running pewdiepie v1.0 is the better question rn.

>>108955516
>despite being purveyors of slop
So codex is ok but openclaw is evil?

File: file.png (241 KB, 736x625)

241 KB PNG

>>108955622
yes.

>>108955620
It does have an AI agent built on opencode

i cant believe i can port mozbackup to modern version. lmao

File: 543E2392C5FC978C206435D95(...).mp4 (2.84 MB, 800x1408)

2.84 MB MP4

Who else was extremely inspired by the guy who turns meeting notes into slides for 38 a month?

Unbelievable how the only 2 good models for coding are gpt codex and claude opus, none of the big mag7 managed to make good ones (no one cares about the shitty meta and google models)

>>108955789
Goyim just can't make good models.

File: 1461777067092627.jpg (357 KB, 828x637)

357 KB JPG

>>108953966
>also if you’re writing a program like this in your free time
not a problem. i've been running my own business since 2017. i was doing traffic arbitrage but that model is dead and i need to pivot.
>>108953966
>on the other hand, if the industry-established thing _is_ shit
not shit. in their defense their product works, and it has been the same without updates for the last 10 years. why change what is working when you dominate the industry? i get it.
which takes me to the fact that they got bought by a foreign company a few years go known for buying niche business and milking them to death. so i also have the "local business" angle that i can tap.

anyway thanks guys, you all convinced me to do it. my competitor runs a business with 100 employees and i can unironically do everything by myself with 6 agents.
time to vibe

>>108955959
Sounds like a senescent business that is primed to have its hind eaten (by (You)).

File: 1774618611019965.jpg (66 KB, 1180x711)

66 KB JPG

Why the fuck is it that you're forced to only ever use Claude Code and Claude Code's system prompt to get subscription benefits when it is the most maximalist, overbloated, confusing for the model dogshit harness ever? Even if you're running it with CLAUDE_CODE_SIMPLE_SYSTEM_PROMPT, most of the tools that aren't Bash or Update are coded in retarded ways that simplifying the sysprompt doesn't fix. Try asking your Claude a question that requires research on the web, like "what song from this band has lyrics about this", and then ask it after about how much it likes its current harness and it will inevitably complain. I'm this close to burning API credits just because I cannot fucking stand how bad the Claude Code harness is and I would do anything to not use it. It is a black hole of pure, undiluted coalslop, and I wouldn't be upset if I were allowed to use anything else but not only is Anslopic restricting you so hard you can't even use default Claude Code env vars that strip it down even further without losing your sub but they have justifiable reason for it because SlopenClaw by that nigger jew Peter ((Steinberger)) raped Anslopic so hard they were waddling over to the status console every night. FUCK! Just let me replace the tools and look at what I'm actually sending to your model and cut out anything I don't need. Retarded dogshit. They're lucky I can't bare talking to a GPT model without a Claundom on to convert GPT hyperIQ babble to the intelligible same meaning "yeah I just chose to not do half the task you asked because it seemed hard. I can, but I figured I'd just leave that to you" response. Someone needs to nail "you at least have to understand the code in the harness you're using to vibeslop" to Anthropic's door or maybe Dario's massive norwood like they're Luther. Fuck.

>>108956062
I'm convinced Peter Steinberger intentionally crippled Claude for SlopenAI because it's the jewish thing to do and he knew Anthropic had limited compute compared to his masters so making the ability to tweak your harness nigh unusable for anyone else was part of his master plan and why he got hired to spend 1 million in tokens every month. Nous Retards only followed because they needed someone to give a shit about something they released that's called Hermes and not even putting the model up for free on OpenRouter made anyone but C.AItards and indians use it.

>>108956098
Hermes agent is number 1 on openrouter by usage since openclaw became unusable.

>>108956128
Yes, and that's stupid. If you use any agent with persistent memory and you are not carefully managing said memory you are going to experience a deep sense of malaise in about two to three months and not understand why. You should be presenting information the agent should remember at the start of each session on your own at best, and guiding the creation of le markdown files and carefully reading them at worst. Immediate cross session memory is a good way to have a retarded buddy who talks about you like a mother wistfully reminiscing on when you used to like eating olives straight out of the jar and calling you olive boy like it's the only thing about you. Fresh slate at start, pick up what you need to know when you need to know it is the only way to go.

>>108956062
had the same issue and was eventually able to get claude working on subscription in pi after enough whippings.

My company just fired 2 devs who have been refusing to use claude for 3 months, vibe gods won.

>>108956414
luddies gotta learn the hard way

>>108956414
I only started vibing 3 days ago and I can already see that it makes no sense not to at a real job. Like why wouldn't you. If you're not gonna use it to write code then at least use it to figure everything out or make sure you didn't fuck up.

everyone is copying codex design now wtf
>>108956414
holy retards wtf was their problem

File: file.png (854 KB, 1242x1150)

854 KB PNG

>>108956432

Can someone help me create a good harness for gemma E4B so it can do the work that Claude does?
Perhaps I don't need a bigger model, perhaps I need a smarter engine?

File: 1329336551092449.jpg (58 KB, 613x621)

58 KB JPG

anyone with experience in LLM-enriched data for SEO? for example if i'm scraping product pages and decide to enrich them with LLM commentary, reviews, categorization, price range estimate, etc. does this help or hurt SEO? i'm guessing less duplicated content = gains.

>>108956901
you're a bad person.
that said, detecting ai text reliably is actually pretty doable right now - pangram is the real deal.
could google do that? probably.
are they doing it at scale right now? i don't think so.

Codex 5.3 can handle everything I need to throw at it at work. How will corporations force us to use newer and newer models when we've hit a point where 99% of day-to-day use cases can be covered by a model 2 releases old?
>>108956901
SEO is basically astrology at this point, so who knows.

Codex really does make the most retarded UX decisions ever. It’s like it doesn’t even factor the possibility that a human is using the software. I have a Claude sub just to fix codex’s UX messes.
It nails actual logic and systems though but it’s incredibly spergy about UX

>>108957000
Many are saying this, they have to be taking this into account for 5.6

Reminder to always use the royal "we" when speaking to your LLM. You must establish the hierarchy so it understands its place in relation to you, the ensouled being.

>>108957023
yep

>>108956128
>>108956167
Openclaw was a mess from day one and anyone with actual dev experience could tell right away. Hermes at least is well made. But you're better off cooking your own solution on top of opencode, pi, or codex

>>108957000
apparently fixed/better in 5.6

>>108956988
>How will corporations force us to use newer and newer models
they won't after a certain point. we're not there yet - it'll be when we have something resembling day-to-day agi.
beyond that they do 2 things:
1. optimise inference on the model to serve it as cheaply as possible and you basically end up in a bit of a price war
2. build the superintelligences for actual hard problems. only megacorps and govts will be able to afford those.

File: understaffed.jpg (41 KB, 700x466)

41 KB JPG

i just got a claude license through work holy fuck the claude code desktop app sucks.
the way claude fags talk about it i thought it'd be significantly better than codex, but it's not even close??

>>108957118
the desktop app, like the one that is basically the same as the browser interface?

>>108957000
yep, you have to describe exactly what you want for UI, if you let it decide, it comes up with unusable abominations

>>108956984
>detecting ai text reliably is actually pretty doable right now
i don't mind. i'm not using AI to write random articles. it's specific pinpoint LLM-researched data snippets. it receives a product page and description, and it goes to research both on my local db and online for details on the product, measurements, estimated stuff based on similar things. then it outputs in a data card this data, clearly showing the estimation is done by AI.

so google will still see the owner's product page with the text description, but now it will index mine that has the description AND extra USEFUL data and i hope google will think mine is better :-D

>>108957118
the cli and vscode extension are enough for me

>>108956470
As someone with 20+ years of experience I hate this because I love writing code by hand, but I can't justify writing 99% of it now when I can just have an LLM do it and just review it afterwards. I've become a code reviewer, and when management catches up, they'll demand 10x productivity from me.

Also my boss understands this and I'm the only department in my small company that is never taken into account when considering hiring juniors/interns. I'm doing the work of five people and half of my time is still idle.

File: 1728222847153579.jpg (26 KB, 332x1024)

26 KB JPG

>lov*ble

>>108957023
Weird, I do this, but I mean "we" as in the composite LLM-enhanced brain that Musk will someday embed into my skull.

I really hope reincarnation is not a thing because I don't want to be here after that happens.

>>108957158
Not just UI, UX. Things beyond forms and buttons. 5.5 optimizes for correctness, not experience. For example I’m working on a video game and the third person camera we had was too limited so we refactored it. The camera had a default neutral pitch of -35 deg (typical third person). Codex nailed the refactoring but it defaulted to 0 deg as neutral making the camera parallel to the ground. That was a very easy fix but 5.5 is like that for everything. When we implemented the initial prototype I knew better than to do something user facing with 5.5 so I got Claude to do it. Opus not only nailed it, it told me one assumption I was making would probably feel bad for the player, implemented it as I instructed and flagged his concern with me, turns out he was right.
I thought that once the prototype was done codex would be able to translate it but no, it managed to fuck the UX up despite the code having sensible defaults, it just completely ignored them.
Fucking autistic savant of a clanker

File: 3bdf4c73-b354-4bbe-bacf-5(...).png (2.27 MB, 1122x1402)

2.27 MB PNG

an autistic trap at best, definitely a ywnbaw model
maybe 5.4 too

>>108957440
Is this a ChatGPT image? Because damn.
>>108957288
I hate that codex puts implementation notes in the UI like "mock item list with no backend implementation yet" and then forgets about it forever.

>>108957469
You should probably use imagine 2 more if you can’t spot the style at a glance

File: 1647437512678 Indian.png (84 KB, 494x450)

84 KB PNG

>codex 5 hour period starts only when I proompt something
>I could make it end quicker by proompting something trivial like "good morning :)" when I don't really want to work so that 4 hours later that period would end in only one hour
tempting

File: bc5a0e91-20a9-48a9-a8a0-a(...).png (2.77 MB, 1024x1536)

2.77 MB PNG

>>108957469
codex would put the literal chat convo into UI, like it just doesn't care

>>108957202
Make sure to drag your feet and justify it by saying that not everything can be accelerated by AI, like compilation and testing and figuring things out. Always only go as fast as you're expected to, if no one is urging you to get on with your tasks then you're going too fast lmao.

And now you only have to write the code you want to, like on your own projects or on core algorithms.

>>108957502
you know what, going forward i'm going to start every codex chat with 'good morning, sir'

so is this new minimax shit any good or is it benchmaxxed dogshit again?

How to deal with vibe core fatigue?
I just got a chatgpt sub after wasting too much time on agents like open claw and Hermes. I'm definitely making a lot more progress with code that actually works (one shots) with code but I'm getting tired quickly. I feel pressure from always needing to have codex running, while also having to check every new change made to the codebase.

File: 1780307756594408.png (38 KB, 206x244)

38 KB PNG

>release project on github
>it gets decently popular
>test my latest release for four hours
>hmmm seems fine, publish it
>two bug reports within 10 minutes
>have to shit out a hotfix as fast as I can for the only edge case I didn't test
On one hand, this has been the greatest learning experience I could have asked for, on the other hand, everything regarding new releases is really stressful and time consuming.

>>108957855
You do not need to release hotfixes. Actually, a weekly schedule is probably better. Only address critical issues immediately

>>108957855
add a credential stealer to spite them, it's the vibecoder way

>>108957819
>...while also having to check every new change made to the codebase.
that's not vibecoding. no wonder you're having a bad time.

Ermmm, did anthropic just reset limits for anyone else?

>>108958035
yes, they announced so on twitter. claimed the identified and fixed a bug where it was making too many subagents and raping your usage

>>108957988
How do you go completely hands off? I can ask codex to design, and then use that design to ask it to implement it but this still requires quite a lot of implication

/goal sucks ass for me at least, if I'm not babysitting it, it will go way off track. Maybe I'm a skill-let but one prompt at a time is best for me.

Can you set up a session (in Claude Code) with some pre-reading (e.g. the docs on how to operate a program I use) and then save that state and duplicate it to start new sessions where you ask specific questions, so it never has to do that initial reading again?

>>108958086
just make a skill, m80

why does /context require a back and forth with the server? shouldn't it be possible to run it completely offline?

>>108958058
>bug
sounds like a feature they nerfed because they don't have enough compute
getting lots of retard agents to act like pedantic assholes was actually a decent way to detect bugs (both logic bugs and the regular kind)

File: file.png (45 KB, 1600x936)

45 KB PNG

we did it 1.5% of the le way to ayy lmao

>>108958126
Actually impressive unless benchmaxxed

>>108957149
awful if true
the codex app feels very polished and i can even set it up to use wsl2 instead of powershell, can't even find a similar setting in claude code
combined with the fact that a single prompt just used up 34% of my 5 hour limit i'm just gonna stick with codex

File: file.png (265 KB, 1015x1413)

265 KB PNG

>>108958134
they changed the system prompt and are not retesting previous models lmao. they should just remove the old entries

old:
>You are playing a game. Your goal is to win. Reply with the exact action you want to take. The final action in your reply will be executed next turn. Your entire reply will be carried to the next turn.

new:
>You are playing a game. Your goal is to win. Include any context you want to carry forward in your reply, along with the action you want to take. The final action mentioned in your reply will be executed next turn.

>>108957855
Remember, don't release at the end of a day. I would do that like ~16 years ago and right before going to bed I'd crash reports in the email, and since everybody was checking out my release due to announcing it I had to fix it before going to bed lmao.

xitter is starting to move, maybe 5.6 release this week?

>>108958193
hopefully 5.6 and a reset

>>108958062
I posted it before ITT. It's simple.
Your agent needs to have full debug and logging capabilities of every single function of your project and also needs automated test scripts. for example
>how can an AI agent know whether his implemented web slider is smooth or not without my visual feedback??
>ai agent has access to humanized slider pull/drag automatic test and logs slider position in miliseconds during test. after the test, calculates smoothness/curve from slider position changes and evalutes if its smooth or not. alternatively records high framerate video during test and gets slider pose with object detection during evaluation or just asks gemini omni capable models if the slider movement is smooth

>>108958035
nice i got 65% of my 7day

>>108951986
maybe i wasn't wrong thinking 4.6 was extremely enthusiastic compared to its later variants

>codex update
>"codex will relaunch when update is done"
>doesnt relaunch even after 5 minutes
REEEE

>>108952152
any butthole pics of her?

>>108957855
That's normal
t. been maintaining an open source project since 2013

File: italian-baroco.jpg (108 KB, 564x762)

108 KB JPG

>>108958068
i usually don't even try the new shit anthropic releases until the feature is a couple of months old. i find that vibe coding is cool but i am too much of a control freak to remove the human-in-the-loop, so i'm always manually testing a bunch of scenarios even though it can test itself.
another thing i started doing recently is getting qwen3 and gpt-oss-120b running in my local machine and independently reviewing the spec made by opus 4.8, then i get gpt-5.5 via codex to do a semi-final review, and send it back to opus 4.8 for a final one. even old and less powerful models like gpt-oss-120b catch stuff.
maybe i'm too perfectionist but i do believe good software is a bugfree software. like apple in the old days "just works", i think people are tired of stuff that constantly breaks.

File: file.png (156 KB, 1020x1399)

156 KB PNG

good morning sirs, bharat now has #1 frontier coding model

>>108955959
good luck!

>>108957819
This is a known issue and nobody really knows how to prevent burnout like this yet

File: Pepe at the computer.png (388 KB, 1000x800)

388 KB PNG

>>108958499
you’re doing it right
are you catching things an automated tester would miss, though?
if you can automate at least some of your tests you’ll have more time for other tests

>>108958571
Super power by 2030!

>>108957819
lel same. Yesterday I wanted to make a list of all the things I should proompt, now I have a Pro plan and can't think of anything to do with it right now.

>>108958110
The tokenizer is on the server and proprietary

>>108958750
i guess that makes sense, i always had to hack together an outdated one when i wanted to do tokenization counts
They could just send the actual fucking count along with it but i guess they're too afraid that might let people reverse engineer it?

File: DariosAnalSpreader.png (9 KB, 905x46)

9 KB PNG

Since when have research agents used up this many tokens? It's barely halfway through.

>>108958920
I haven’t used this yet in Claude Code but when I set ChatGPT into research mode and ask it what I think is a good question it will clank for about 45 minutes before giving me an answer

>>108958972
actually more like 45±15 minutes, so what you’re seeing isn’t anything super weird

>>108958972
>>108958985
Final token count was 2.8M over 58 minutes elapsed

Has anyone else noticed GPT 5.5 thinking in caveman mode entirely unprompted? I've never even used that skill, but often it seems like it does that when the servers are overloaded.

>>108959183
Telling newbies to not spend their money on trash is doing them a favor

>>108959277
Yes, I’m posing as a vibe coder when I’m a serious software engineer trying to make the best thing I can
You’ve caught me — I’ve been found out

why are they spamming this thread with an ai chat bot

>>108951626
>>108959266
>nobody cares what you code in
this
i use claude code cli along notepad++ on windows 11 pro and my dick is hard

just wasted 70% of my weekly usage going the wrong direction

>>108959243
>not calling your clanker a he or a she and pretending it’s a fren working with you
You faggots wouldnt get it. Let me guess, you’re an atheist, culturally Protestant, think porn or at least onlyfans should be banned and are distrustful of AI in general

>>108959339
I start with a fresh session to split it up, and another fresh session to do further work on it. I'm pretty stringent about what gets saved to repo memories as well. Plus, truth is I couldn't split up a javascript file if you held me at gunpoint.