[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology

Name
Options
Comment
Verification
4chan Pass users can bypass this verification. [Learn More] [Login]
File
  • Please read the Rules and FAQ before posting.
  • You may highlight syntax and preserve whitespace by using [code] tags.

08/21/20New boards added: /vrpg/, /vmg/, /vst/ and /vm/
05/04/17New trial board added: /bant/ - International/Random
10/04/16New board for 4chan Pass users: /vip/ - Very Important Posts
[Hide] [Show All]


[Advertise on 4chan]


File: Chadcat.png (1.95 MB, 999x1430)
1.95 MB PNG
A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, and shipping code with LLMs.

You use Git, right, anon?

## 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/

## News
- (2026-08-31, 11:59 PM America/Los_Angeles) Claude’s 2× promotion set to end
- (2026-07-24) Claude Opus 5 out

## Related generals
>>>/g/lmg/
>>>/bant/agdg/ — schizo-resistant temporary (?) hideout
>>>/vg/agdg/

----

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

## Near-frontier models for code
https://x.ai/cli

## Not worth it for code, but maybe good for interpreting images/video
https://antigravity.google/product/antigravity-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 — /grilling is a favorite
https://github.com/DietrichGebert/ponytail

## Other editors / terminal agents / coding agents
https://osaurus.ai/
https://pi.dev/
https://opencode.ai/
https://cursor.com/docs
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/

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

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

## What we’ve done
https://vcg.gitgud.site

## Previous thread
>>109594262
>>
AI won
but keep up the melties it's funny
>>
>>109602073
but that’s for getting clankers to make videos and pictures
I want to have clankers write programs for me
>>
>>
chadcat won
>>
>>109602085
You are replying to the /ldg/ village schizo known as ani.
There is a rentry link in the OP of /ldg/ that details his autism and schizophrenia, and it's been driving him insane lately. What you are seeing with spam like this >>109602073 >>109602083 >>109602094 is ani's latest attempt to get the existing /ldg/ thread deleted.
tl;dr: the guy is a genuine nutjob.
>>
and now....ladies and gentlemen...watch /g/ going fuuuuuuuull retard mode for a unknown reason. Let's have some fun!!!
>>
File: 1757747089957043.png (122 KB, 1254x828)
122 KB PNG
>>109602059
I think i become addicted to spending 1B tokens every day, and not shipping my vibecode
>>
oh dear, it's the troll OP again. someone needs to be better about making a thread before him
>>
>>109602110
Is he conscious of the fact that this is /vcg/ not /ldg/, meaning it wouldn't be perceived as thread splitting?
>>
>>109602163
“What’s the difference between ignorance and apathy?”
>>
>>109602163
Anon, in case you haven't figured it out already, he is severely mentally ill. In fact, he's no longer human, he's been completely consumed by his schizophrenia.
He's spamming these cross-links everywhere he can, desperately hoping a mod will think someone is trying to shill their own thread and therefore delete that thread.
>>
.......~@^^
............~@^^
.....~@^^
>>
>>109602188
just create a AI board already
>>
>>109602217
/ai/ is reserved for Architecture & Interior Design
>>
>>109602225
oh i forgot...my bad
/gugugaga/ would be cool tho
>>
>>109602225
>>>/ai/ 404ed for me just now
>>
>>109602217
i was half expecting that's what the maintenance was about the other day
/xs/ extreme sports was the last board i can think of that was added, it's past time for an /ai/ containment board
>>
>>109602254
/v/ DALL-E general could be back...oh my god i've got tears coming
>>
>>109602254
I dunno
I think of “have AI make code” is very different from “have AI make art” and “have AI make games” (some code, lots of not-code) and “have AI write books” (not that a /lit/ general like that exists)
a proper full board is going to have, like, what, 100 threads in flight at once, right? What are all the other generals going to have in them?
>>
>>109602269
I'd like an ML general tbch
>>
You better not be prooooompting DS right now you little shits, we are in peak hours and anyone who gets caught wasting money like that is getting their balls stomped
>>
File: 1781296496354602.png (2.55 MB, 1672x941)
2.55 MB PNG
>>109602202
>>
>and finish fun games
Post one example of a vibecoded game that looks fun and finished
>>
>>109602141
I don't even read my code, what makes you think I read the OP?
>>
File: 1763886829573065.png (124 KB, 1267x827)
124 KB PNG
>>109602140
I'm going for a nap
>>
>>109602587
I think mine is fine, but it is based on an existing card game. I still have a few small fixes to do and I need to crop out the cards manually, so I haven't bothered in a while.
>>
File: 1763645082340509.png (675 KB, 918x784)
675 KB PNG
if /g/ is so smart and vibe coding is the next best thing, why are you struggling to make an upload CLI tool for pic related?
>>
>>109602587
My game will be fine and fun
>>
File: 1738804603717200.gif (887 KB, 250x250)
887 KB GIF
>>109602447
>>
im running 3 million agents at once
>>
start your engines, off peak hours coming in 7 minutes

but it just isn't the same as the old api prices.. now i have to think hard if i want to fire of that proooompt or not
>>
>Claude desktop app
This is like inviting a jew into your personal computer.
>>
>>109602987
same with codex
>>
Someone posted this snail girl. I wonder what the prompt to generate it was.
>>
I feel like a fucking hobo. With deepseek tokens being so expensive i only save it for prompts that needs to work with multiple files in the project and for edits that only span a single file i copy paste the whole thing into web gemini, or DS chat, or GPT so that way the proooompt is free

i'm like a broke ass fucking hobo, checking the vending machine change return try for some loose pennies
>>
>>109603194
you don't need to know their prompt anon
>>
>>109603228
Pdfile prompts on your Claude account are not a good idea
>>
>>109603249
sama doesn't care and will make cute anime snailgirls for you
>>
>>109603254
the shell looks gross and slimy like poop
change the texture of it
>>
File: file.png (434 KB, 582x787)
434 KB PNG
>>109603256
i'm never going to use it again, so go for it anon - it's all yours
>>
>export chatgpt conversation addon
>works perfectly 2 days ago
>saving images breaks a day later
>fixed a few hours later
>now saving embedded files breaks
kek the web ui works just like ai models, sometimes it just werks (tm), other times nothing fucking works until you regen a few times
>>
>>109603254
Can you give me the prompt
>>
>>109603313
i just uploaded this >>109603194
and
>could you create a character sheet for this snailgirl
>>
File: file.png (47 KB, 983x271)
47 KB PNG
Claudesisters, have you stated your pronoun yet?
>>
>>109602783
Be the change you want to see
>>
File: 1759939018595709.png (52 KB, 360x360)
52 KB PNG
>unironically advertising on reddit to drive sales to my app
you either die a hero or you live long enough to become the villain
>>
What's the buzz on the DeepSeek harness or whatever they call it? Anyone have experience with it?
>>
>>109603439
i saw someone made a buttplug controller extension for it
https://github.com/c3ll256/dsh-toy

seems like the future
>>
OAI $200 Pro 20x or Z.AI $168 Max always-off-peak usage?
>>
>>109602587
I'm remaking KSP with agents
>>
I just went back to opus 4.6 and asked it to rewrite my docstrings. And they're actually readable now.
No more "the integration is the surface-oriented load-bearing mechanism."
>>
File: 1785978723592365.png (187 KB, 1333x1018)
187 KB PNG
>>109602059
https://github.com/bluescreenoff/MangaNakama
Clip Studio Paint replacement
Thoughts? Welcoming contributors, testers etc. But if none of you really do drawing stuff no biggie.
>>
>>109603844
can you import any type of brush?
>>
>>109603640
>performs mental gymnastics equivalent to a roastie with 10 years yoga experience while playing the floor is lava except the lava is anthropic
ANYTHING but a 100$/200$ sub which gives you access to the undisputed, best model available.
>>
>>109603880
What are you talking about
>>
How do set up claude to make it work with multiple features/branches at once?

My issue with this is the merge conflict problem, so I want to understand how anons avoid that.
>>
>>109603880
dariobot please
>>
File: 2026-08-20 15.41.02.png (115 KB, 1352x489)
115 KB PNG
>there are literally faggots literally shaking and crying and quitting their jobs because of AI
the future truly belongs to vibeGODs Snailcats will fuck off and leave their jobs to us.
>>
>>109603992
Link to video? I love "AI bad" seethecore.
>>
>>109603844
>98.9% rust
impressive, that UI is pretty cool for having no webshit layers
>>
>>109604022
https://www.youtube.com/watch?v=2ZU3j4GQ4K8
>>
File: 241879765745598679878.png (185 KB, 1127x454)
185 KB PNG
Lol, I am working on my audiobook player and trying to utilize the native Intel Mac transcribe feature for deaf people to get live text of what the reader is saying, for now it's not in the best state, maybe I need something else.
>>
For those doing front end dev with claude, how do you get it to stop adding verbose chunks of explanatory text for everything in the UI?
>>
>>109603915
tell your clanker to make your code more modular so its easily extensible
>>
>>109603892
sorry I forgot my HRT this morning and got emotional;
>>
>>109604044
Sol does this too. It’s very funny, I think it’ll get annoying fast. Try putting the thought in AGENTS.md or the Claude equivalent.
Have a clank analyze a few traces so you can get an idea where it’s looking first for repo instructions and agent instructions, put what you want the agent to do differently in those places.
>>
tempted to vibecode a podcast app for android because the best opensource app, antennapod, is bloated and gets buggier every single day
>>
>>109604059
this but unironically

>>109603915
non issue with fable. just told it to create extra branches when needed and auto commit and merge whenever it feels its appropiate. didnt have any issues whatsoever. imagine having to play janny for your shitflinging gpt5.6 shizo agent, couldn't be me. very non viberesque.
>>
you use gpt3.5 still or did you buy 4.0? I have seen videos and I am not sure if it's worth the upgrade. Free 3.5 just works for me. How much better is 4.0?
>>
>>109604133
I boughted
>>
>>109603992
>can be useful for sorting out tech debt
Are people really THAT delusional? What the fuck is wrong with humanity?
Every single line of code is tech debt, you fucking motherfucker. When I see that fucking motherfucker bitch fuck in minecraft, then something will happen. I am not legally allowed to say it here, but something will happen when I meet this fucker in minecraft
>>
>>109604032
Thanks.

I don't understand this mindset. AI makes him feel bad and he feels threatened by it so his solution is to pretend it doesn't exist?
>>
I think the verbose texts in gpt images are funny and cute
5.6 would try to fix them though
>>
>>109604223
smailfag feefees are a mystery
>>
File: 1781103445658236.png (3.42 MB, 1792x2304)
3.42 MB PNG
>>109602059
One thing I like about capable AI models is that they aren't ONLY good for "big boy" stuff like creating new software or refactoring a code base. I can ask it to do something as mundane as stitching together a still image beside a video and it just figures out how to do it without any handholding. I told it to make vid rel and it just did it

https://i.4cdn.org/g/1787087719298862.mp4

https://civitai.red/images/140194917

https://files.catbox.moe/bfraa1.mp4
>>
>>109603402
He wouldn't be here then, you retard.
>>
>OpenAI yesterday told employees it intends to release Astra "in a couple [of] weeks"
>Anthropic are sitting on Fable 5.1 until Astra is launched
>>
>>109604360
(You) wouldn't be here because you're the one complaining about it.... You know you're too lazy to do it even though the AI literally does all the fucking work for you because your brain is just THAT fried, so you make that someone else's fault.
>>
Is opencode a scam?
I see people here get 100b tokens/week with a $200 subscription. so like ~$1M worth of tokens per month (5000x of what they pay). opencode gives you at most $60 worth of tokens per month for $10 (6x).
>>
>>109604404
>Is opencode a scam?
probably not
you're just paying the actual costs
openai and anthropic are subsidizing subscription to the extreme
if I were to use my claude work account for my hobby project, it'd be thousands a month
>>
>>109604379
you sound like you're in mumbai
>>
>>109604369
This sucks (the fable delay) but at the same time, is there really anything anyone here tries to do and the current models can’t do it?
I’ve been finding that the current sota trio (fable opus and sol) can do pretty much I ask them to do.
Which makes me think the improvements can be
>one shots with minimal prompts, probably webshit related (make me X)
>token efficiency (getting things right faster or with fewer mistakes)
>non coding related improvements
>maybe being better at system architecture? To avoid bloating up codebases
>>
>>109604426
>subsidizing
What an absolute meme. You don’t know if they’re losing money on subs, you’re just parroting what you read online. Nobody knows if plans are subsidized or not. All the companies are private, their accounts are not there for you to see
>>
>>109604426
you have it backward, API pricing is enterprise scam
>>
>>109604347
This. I just had it tackle loading a bunch of images into an ecommerce site. Didn't really need a tool for repeated use, just a one shot, so it made up a tool that worked with the site's API and did the upload for me, including making sure the video met quality and size standards for the site.
Took about 20 min to watch it do whole thing, avoiding me the tedium of manual upload or having to pay a service to do the work.
>>
>>109604447
wrong on all counts
if opus5 taught us anything, it's that anthropic are all-in on benchmarkmaxxing
they don't give a shit about all the quality points you pointed to
>>
>>109604426
>you're just paying the actual costs
it's not api rates on go
i think the rate increase is primarily because they're a large player and can negotiate better rates with infra providers + low usage subs basically pay for people who run through quota every time
>>
File: 1637173665912.jpg (97 KB, 782x788)
97 KB JPG
>>109604456
>>109604457
>>
>>109604456
Either we're being heavily subsidized or enterprises are being totally ripped off, how else do you explain getting 15k worth of token billed usage from a 200/m sub?
>>
What would be the best tools(s) to vibe code a page for a gymkana? It would consist of a map with a few marks, where the user had to physically find the QR and scan it. Right now I'm just using chatgpt since it already recognized it as a project, and told me about 20 steps after discussing the stuff I need/want and the rules and such. I would like something that could be linked up with the different programs, web pages and such that writes and builds the code itself, I just do the testing.
>>
>>109604433
And you act like you turn tricks

>>109604461
One specific instruction I gave it was after the video was exported it should export a frame from the video and then use the harness' 'read' tool to look at the image and verify the video and text was formatted correctly. We used to make fun of "prompt engineering" as a term but it really is that important. Nearly having a smart model means nothing if you can't give it instructions. I hate being one of THOSE "only high IQ people can use AI well" guys but you also need to be somewhat tech savvy and good at problem solving to begin with in order to do this kind of thing. The good models need to be treated like autists that need to be given specific instructions. The more vague or shitty are open in that you're prompt is the more both it and you are going to struggle
>>
>>109604447
I agree. Future models will probably make work more comfortable and faster, but you can already do anything you want with the current models. I was already quite happy with Opus 4.8 and GPT 5.5 really.
>>
yall arguing cause you don't even know how to define the term scam
>>
>>109604494
do you already have a chatgpt sub?
>>
>>109603915
git worktrees is what you want
>>
>>109604447
I'm unironically waiting for models smarter than me... current models are already genius at raw logic, but they are not THAT smart yet
>>
>>109604532
What do you think is still missing? I guess sometimes they don't see the big picture or something, but it's hard to put my finger on what exactly they are doing wrong. I think overall the models are already smarter than I am. I'm probably better at very specific C++ work, but the breadth of LLM knowledge is something I don't have even remotely.
>>
>>109604484
If you earn 1 unit of value per user with one service and 0.2 units of value per user for another, one is more profitable and still neither are subsidized.
There’s also many issues you don’t know about such as silicon availability and cluster quality and there’s also the statistics and the law of large numbers. The maximum possible value extracted out of a sub is not the actually extracted value as not 100% of users are 100% maximizing their usage.
There’s also ebbs and flows of demand, peak hours, etc.
meanwhile API users get too silicon top throughout 24/7 no downtime
I don’t think they’re losing money on subs, at all, but they may be making less money per user
>>
is daybreak blue a different model or is it just routing you through to sol with less safeguards
>>
>>109604547
holy fuck would you get the fuck out?
all that fucking text to say absolutely nothing
>>
not my problem you can't understand what I mean, sounds like a (you) problem
>>
>>109604575
not sure, but did you do the kyc for trusted access to get it?
>>
>>109604587
thanks for giving communication majors a job
>>
>>109604547
That's all well and good in theory but there's an extreme difference in cost here. I do not believe that the average person buying a 200 dollar plan does not purposefully try to maximize their usage and there's no way there's enough 20/m users not using their plans to make up for the extra 14800/m for each 200/m poweruser.
>>
>>109604575
Sol with less safeguards is just Sol with less safeguards lol, I have TAC 1 and Sol just stop refusing for a lot of stuff.
>>
>>109604627
Maybe max 20 plans are slightly subsidized, doesn't mean everyone on a sub is.
I would imagine the split being 80% $20 subs by people who use not even half of their allotted usage and the rest 20% power users, of which a bunch are on max 5 plans who get a quarter of a max 20 usage.
If you tell me half of Ant/OAI subs are on $200 plans I would not believe you (but again, these are just guesses until if they go public)
>>
>a chatgpt project can only contain 40 files
>claude handles hundreds of them just fine
why is it so limited?
>>
>>109604670
>>a chatgpt project can only contain 40 files
that can't be a thing? you mean in the webapp?
>>
>>109604690
yes, in the webapp
>>
>>109604700
why would you use the webapp for anything that big?
they probably limited it for it not to become a free dropbox
>>
>>109603422
It drove a singular sale. LMFAO, It's never been more over than it is now.
>>
>>109604627
>the poor subsidize the rich
shouldn't it be the other way round tho? but it's like real life.
>>
>>109604729
Unless you make something good like something that takes years to make even with LLMs
You won't make money with software
Everyone can make their little niche desktop tools with cheaper subcription one time payment
>>
>Reddit users say OpenAI may have cut Codex limits in half. NerfTrack estimates the weekly ChatGPT Plus limit dropped from roughly $160 in API value to about $80 within a few days

>TiboTattle also converts Codex usage into API price equivalents. Its creator says they have not seen any significant limit cuts in recent days. Pro 20x is currently estimated at around $2,043 per week
>>
>>109604752
20 dollars is actually 2000 dollars of usage
>>
>>109604035
The book already has the text...
Otherwise just use whisper.
>>
>>109604447
>I’ve been finding that the current sota trio (fable opus and sol) can do pretty much I ask them to do.
Yes, it keeps failing at doing a p2p connection.
But also I am sure most people here are not doing anything that complicated, but complicated tasks do exist.
>>
So cerebas just killed it's free tier, where can I get an API key to use Gemma 4 for free with decent rate/token limits on my personal website chatbot?
>>
>>109604775
>The book already has the text

Mine is live text from audio
I don't have ARM mac so the advanced model isn't there.
>>
>>109602752
>>109602786
>>109603652
>finished
>>
>>109604811
>cerebas
what even is a cerebas?
>>
>>109604892
It is completely playable. I have no changes to do to mechanics. Just have a few small graphical changes and better card cropping.
>>
chinese rumor: recent codex cache hit rate is off
>>
So I got started with this when Fable was already out. You weren't kidding about Opus 4.8 being easier to handle than Opus 5. I found Opus 5 to only be good if you actually want something to develop crazy tunnel vision and get a SMART goal done at all costs.
>>
>>109604918
A typo, I mean Cerebras
>>
>>109604518
thank, anon.
This is what I am looking for.
>>
It is pure suffering watching 1 (one) solitary flash agent drain a whole day's quota in less than a day. Unnatural. God has truly forsaken this doomed world.
>>
>>109602059
>wants to vibecode for the 1st time
>there's nothing free to test and learn anything
so what do I pay for and what do I use?
>>
>>109604953
>Cache miss: 112k tokens re-billed (~$0.50)
as someone who just saw this in the middle of a chat for no good reason, i'm inclined to believe this. i've had miss warnings on for a few days and shit just drops cache for no good reason
>>
>>109604959
>even vibe newbies feel and see the difference
benchmaxxing should be ridiculed at every level
>>
>>109603852
slop answer:
Your brush-import question — current state, precisely:
- .abr (Photoshop brush sets): yes, File ▸ Import — and it's deliberately half-faithful. The
sampled tips import exactly (each becomes a grayscale texture + a plain pressure brush in an
"imported" group), but dynamics are not translated — same policy as CSP's own ABR import, because Photoshop dynamics don't map onto our engine 1:1. You retune the feel like any brush. One known
wart in the queue: radius is read from the tip's padded long edge, so extreme-aspect tips (a
2000×100 grunge streak) import oversized — flagged as a design call for you.
- .myb (MyPaint brushes): perfect imports — libmypaint is our engine, so a .myb dropped into the
brushes folder is native (35 classics ship).
- .gpl palettes (GIMP/Krita) import too — colors, not brushes.
- Not importable: CSP .sut and Krita .kpp. Those engines' parameter spaces don't line up with ours, so "import" would be a lie. The CSP-feel and Krita-group presets are
hand-translated/hand-authored instead, tuned against references.
- Future: more formats are planned where they're honest — GIMP .gbr/.gih would be near-free (same
tip model as ABR). Cross-engine dynamics translation stays handwavey by nature; anything we add
there gets the ABR treatment: exact shapes, plain feel, you tune. (The Rebelle brush-study phase
is still queued on your list.)
>>109604031
Honestly I have no idea what the fuck rust even looks like as a language kek. I will forward your compliments to the chef.
>>
>jeets leaked full gta 6 pre release build
infinite izzat redeemed with zoomers and millenials
>>
>>109605153
Shit game, I don't care. I am embarrassed I even bought 5.
>>
>>109605153
too bad it didn't happen sooner
all the people making gta clones could've just reverse engineered this one
>>
>>109605153
There's no leak I just made some videos using H3 and the retards fell for it lmao
>>
Just caught 5.6-sol-xhigh on arena and made it shit out 6k python lines. I wonder how much $s sama lost on me
>>
Does anyone take advantage of the Oracle free-tier ARM64 2xOCPU + 12GiB RAM VMs?
>>
>>109605351
I only heard about it when they announced it was getting slashed from 4xOCPU + 24GiB RAM. apparently it was a big deal for the minecraft server hosting community.

24GiB does sound awfully generous...
>>
>>109605276
probably like 10 cents
and he got training data
>>
My ROCm GEMM and Conv libraries on gfx1201 are beating Composable Kernel at the top 100 shapes that I extracted from over 500 model architectures
>>
>>109603194
>>109603228
>>109603254
you guy lust after snailcat waifs because you know they are better than a sloppette
>>
>>109605539
snailcat is sexy
>>
File: file.png (36 KB, 1602x94)
36 KB PNG
>>
File: 1760808188459569.png (700 KB, 972x1856)
700 KB PNG
where is the pepe poster who wants a cool project to do in rust, I have an idea
>>
File: file.png (8 KB, 883x30)
8 KB PNG
incredible work from sol
>>
Chatgpt is basically my boss
>>
File: file.png (83 KB, 1281x532)
83 KB PNG
Stats on this thread (Note: I use 4chan-XT)
Just reminding you to play with your developer tools :^)
>>
>>109605054
read the OP
>>
File: eviesupremacy.png (2.93 MB, 1203x1555)
2.93 MB PNG
>>109603194
Lol
>>
>>109605054
apparently `agy` (the google equivalent of codex cli) works even for free users. so if you have a gmail account, you can try that.
>>
why are most luddites weebs and high estrogen males
>>
how come everybody has settled on the same ai prices now, even the chinks who are supposed to be cheaper but they're actually more expensive and worse. competition? looks like the ai cartel is already in place.
>>
>>109605938
I guess the liquid market for GPU-loaded datacenters levels the price quite a bit
>>
Which $20 sub is better value right now in terms of tokens/week, Claude Pro or OpenAI?
Anyone have concrete numbers for both?
>>
>>109605993
gemmy
>>
>>109605993
oai
>>
>>109605993
Probably OpenAI if you’re comparing it to a Fable-less Claude
try not to screw yourself over by using all your weekly tokens in one go unless that’s what you really want
>>
I love Fable-sama and Luna-chan
>>
>>109605993
OpenAI and not even close.
>>
>>109605993
OAI but you need to be smart about it. Sol is not cheaper than Opus. If you're frugalmaxxing, sol medium to plan, luna max to implement
>>
>>109606092 (me)
Claude has no answer to Luna max. This is what I was trying to say. Opus and Sol are on even ground, I'd say each can be better than the other at some thing or another, but they're more or less even. But Luna should be your workhorse on a plus account and claude has nothing like it.
>>
>>109606092
>luna max
but the graph said this was pricy
>>
Pre revolutionary vibes on r/codex/
>>
>openai made a new app that will obsolete codex
so what's the difference between codex and work? both are agentic and seem to do, uh, exactly the same...?
>>
>>109606120
fuck me if I know, apparently people are now complaining it's expensive?
usage limits are really opaque, on reddit they seem to be up in arms
personally I used luna max for a 20min task this morning and it didn't even use 1% of my limits so I thought it was working the same as always but maybe not

>>109606226
nobody knows what work does
>>
File: file.png (99 KB, 869x730)
99 KB PNG
>>109606092
>Sol is not cheaper than Opus
It definitely is, because not only is the tokenizer more efficient but the amount of tokens it chews through for equivalent task is also less than opus. The price per task is one of the few useful charts on artificialanalysis and the difference is very visible.
>>
>>109606246
That chart is showing API prices, we're talking actual usage on a plus plan and in general everyone is saying you're getting a lot less usage on OAI accounts than you used to get, users are reporting this across the board
>>
>>109606226
>so what's the difference between codex and work?
I get paid to work
>>
>>109606246
Also I just read your chart again. It's measuring max reasoning effort? nobody uses opus 5 on max or sol on max for that matter. This is the problem with benchmarks, people just see the bars and their neurons get activated. You're like the little monkey in the meme. Look at what you post next time
>>
>>109604484
All the numbers are fake and made up. Let's start there. A $200 plan doesn't mean you spend 200 dollars in compute.

Trillion-parameter models like Opus, Luna, Deepseek need around 8 x H200 cards to run per node. Those cards run at 700W each. 10 kW of power in datacenters at $0.1/kWh means those cards will cost you $1 per hour. For many millions of output tokens. So hopefully you understand that the actual ongoing costs of inference are dirt cheap.

The only thing making inference costly is the cards themselves. Because of the extreme demand right now, the price of silicon is so high that the depreciation on those cards alone is like 95% of the cost of inference. But remember, all the big labs are financed by nvidia in one big circlejerk, so we have no idea what they are actually paying for the GPUs.
>>
>>109606226
It's a branding exercise to make normies use the scary coding tool without realising they're using the scary coding tool
>>
>>109606263
>>everyone is saying you're getting a lot less usage on OAI accounts than you used to get
>claude lowered usage by 33% a few days ago too
snailbros... it's our time
>>
File: 1681226828973015.png (203 KB, 500x429)
203 KB PNG
>>109606330
>>claude lowered usage by 33% a few days ago too
why would you out yourself as a retard so readily, anon-kun?
>>
I'm not normally a meme skill guy but this does interest me
https://arc-skill.vercel.app/

Not specifically the arc-agi part, but the predict mechanic described here because it seems like it could be generalised the same way chain of thought eventually was
Lecunn and Sutton have both complained that models don't predict things and it would be funny if you can just fix that by going "predict shit"
>>
>>109606340
there was extra 50% promo running for the last 2 months or so
if that's your baseline, losing it means you're now to 2/3 of the previous usage
>>
File: file.png (210 KB, 723x580)
210 KB PNG
interesting luna max vs opus med bench from cube anon's website
opus med shown as faster, better and cheaper

>>109606330
>claude lowered usage by 33% a few days ago too
out of the loop aren't you?
>>
File: 1671859775376675.png (66 KB, 276x256)
66 KB PNG
>>109606350
anon-kun.... another extension has just hit the promo...
>>
File: file.png (357 KB, 738x415)
357 KB PNG
>>109606356
>>109606355
well that was a very quick "we're so back" -> "it's so over" for the snailbros
>>
>>109606355
this is bullshit, if not, this is from before the 80% price cut
>>
>>109606340
he’s not a retard (probably)
he just doesn’t pay attention to the news, or the news section in the OP
>>
>>109606226
>>109606309
Also it's a way to make people spend more. Right now if you only use ChatGPT you only have to pay $20. If your usage constantly runs out then you need to spend more.
I downloaded the desktop app to try it and it told me to switch to "work" every 5 minutes.
I wonder if they'll get rid of ChatGPT altogether and only leave metered usage maybe with Luna as a fallback for when you run out of credits or something like that.
>>
>>109606399
>he just doesn’t pay attention to the news
it's not about not keeping up with the new
it's about using dated news as an argument
now fuck off you disgusting redditor
>>
File: 8qm0gp.png (38 KB, 1280x1380)
38 KB PNG
Is Cerebras just vaporware at this point?
What stops them from making a single youtube video showing Sol or even any half decent model other than fucking gpt-oss-20b running at 1 gorillion tokens per second and getting all the hype they can get?
>>
>sama has a $20 plan
>dario has a $20 plan
>elon has a $30 plan
any Grok enjoyers know if you get about 50% more fun with Grok compared to the baseline OpenAI and Claude plans?
>>
>>109606406
>I wonder if they'll get rid of ChatGPT altogether and only leave metered usage maybe with Luna as a fallback for when you run out of credits or something like that.
subscription is mostly there so all the nerds doing codesloppa can get their chats trained on, because it's the highest quality signal to noise you can get from current users - easy to test whether the build succeeded or not, if the tests passed or not etc
>but they say they don't train
everybody does, it's just anonymized. the only place you're not getting your shit taken is amazon bedrock, unless you're using fable (lol)
>>
>>109606441
if grok was as good as 4.8 for coding I'd probably pay for it just for the probable chance of it being less of a guardrailed bitch
>>
>>109606469
> 4.8
Grok 4.6 supposedly is Fable-tier on some tasks.
>>
Any way to use Fable as an orchestrator and Luna as the subagents?
>>
>>109606469
I thought Grok 4.6 was about as good as Opus 5 and twice as fast
Grok’s catchup speed has been insane recently
>>109606481
yes, we do this at work
you can call Codex at the command line with a prompt
>>
>>109606474
I don't want fable tier
>>109606490
I don't want opus 5 tier

I want opus 4.8

I fucking hate those benchmaxxing shitbots
>but number higher is better!
noi it's fucking not.
>>
>>109606490
>I thought Grok 4.6 was about as good as Opus 5 and twice as fast
go home elon you're drunf
>>
>>109602059
Rugpull in process by Codex&Claude.
https://imgur.com/a/awTpU0m
Need a proper third party. OpenCode Go only works if you're using DS Flash.
>>
>>109606490
>you can call Codex at the command line with a prompt
What prompt do you use?
>>
@109606526
this is an imageboard, post an image i won't click on any links
>>
>>109606445
Maybe, but I think they are trying to convert unmetered ChatGPT users into metered Work/codex users or at least testing the waters to do so. Whether that is to reduce loss if subs are sold at a loss, increase profit if subs are profitable, to get better revenue numbers for IPO or simply to balance compute, I don't know. But I think that's the direction they are going.
>>
>>109606356
the extension is fake and has been gone since July 19th btw
>>
>>109606534
Maybe he's IP range banned from doing so. I was for a while and had to post external links and be called a newfag for it lol.
>>
File: 1685661463378181.png (232 KB, 1000x1000)
232 KB PNG
>>109606538
>packpeddling on behalf of anonymous people on the internet
it's alright, anon. I'm bored too once in a while
>>
>>109606532
iunno, ask your clankers to figure something out
>>
>>109606532
Your orchestrator will figure out a prompt, just as when he calls Claude models. I have a codex-review skill for Claude and it just calls Sol over the CLI to get a code review, it's simple and works very well. The same could be done for Luna, and not just for review.
If your usecase gets more complicated, you can also build your own harness, but I would just start with CLI calls.
>>
>>109606330
I tried something recently. I used the same prompt with Luna light and Sol max.

Somehow, both models used the exact same usage quota. What the fuck is going on? How can the lowest consume the exact same as the highest? Trickery must be afoot.
>>
>>109606625
maybe the smallest one consumes more tokens
>>
File: spongebob ummm what.gif (2.81 MB, 222x379)
2.81 MB GIF
at this point it's predictable
>release new model
>nerf usage limits
>quant model
>continue to squeeze dry until new model releases
>release new model
>goyim forget
>repeat
>>
File: 1762684080047719.png (616 KB, 928x904)
616 KB PNG
>>109606662
>>
>20k credits per week for 100$ !
>18k
>16k
>14k

This shit is a fucking scam
>>
get ready for vibetinkering
https://x.com/GeneralistAI/status/2090161945307664621
>>
File: 1684168526390755.png (271 KB, 383x331)
271 KB PNG
>>109606662
>>109606727
>>
File: 1763084264439161.png (130 KB, 236x329)
130 KB PNG
>post site to Reddit
>you know you're killing the earth, right?
>wow, Claude went crazy bro...
>nice vibecoded project, you didn't do anything
Normies are in for a reckoning. By the way, I made $50 from posting about it on Reddit while they seethed in the comments.
>>
File: 1756880540618361.jpg (9 KB, 251x251)
9 KB JPG
>>109606662
>>109606727
Just need to buy a computer good enough to locally run Kimi K3 to avoid that issue
>>
>>109606781
neat, where did you post it? might as well do the same, I'm surprised a seething mod didn't delete your post
>>
>>109606788
>just buy a $500k computer bro
>>
>>109606781
good job anon
you're now a successful human being and can stop posting cringe frogs on the web
>>
>>109606781
>I made $50 from posting about it on Reddit
?
>>
>>109605054
What this anon said >>109605891
I think Codex runs free trials, and so does Cursor I think
>>
>>109606810
Honestly, I probably should stop frog posting.

>>109606834
The sales came from non-seething Redditors since the post was basically an ad.
>>
>>109606851
>I probably should stop frog posting.
congratulations. You're not a newfag anymore
>>
>>109606851
>The sales came from non-seething Redditors since the post was basically an ad.
so you're doing gorilla marketing on rebbit and people actually buy your stuff? based
what's your app?
>>
which retard aaid the gra leaks were fake
jeet at rockstar india actually got hacked and his build stolen
L
>>
>>109606873
I bought an ad on 4chan and some people clicked on it
people do that
I’m not selling anything, though
>>
>>109606873
>what's your app?
It's probably not a good idea to post my site after calling potential customers seething redditors on a site like 4chan. please understand kek
>>
>>109606919
funko pop distributor detected
>>
>>109606295
You can't reasonably approximate costs without factoring in GPU depreciation into your figures, these chips cost a ton and have a MTTF of like 5 years even before you factor in obsolescence as better and better cards are required to run newer models effectively within reasonable node sizes.
You can't hand-wave that away and only look at electricity prices. Besides, there's way more operating expenses involved in running a datacenter worth of GPU servers. You're not factoring in payroll, drive failure, GPU failure, networking operational expenses, cooling, power redundancy related costs(testing your diesel backups for example), there's a lot of moving parts here and they all cost money.
Did you seriously think that the only cost of inference serving was chip capex and electricity?
>>
>>109606867
posting the frog is fun, though
especially when you have a bunch of them ready to go for almost every occasion
>>
File: 1787262163304.jpg (11 KB, 763x75)
11 KB JPG
>>109602059
beautiful.
>>
File: sc.png (2.79 MB, 1536x1024)
2.79 MB PNG
scopechads stay winning
>>
>>109606732
Robots will be a meme until they can master controlling a computer and do useful tasks through screen reading and keyboard emulation only.
>>
Been trying out that omp harness. Why the fuck was I using claude code for so long. Why was I not using this months ago?
>>
>>109607191
enjoy your ban
>>
>>109607191
because Dario said he was gonna cut people off from using that unless you’re using API credits
but then he said “we’re not gonna yet” and it’s been like months
>>
>>109607208
>>109607201
It looks like it uses the same sub usage and not api credits so idk how it connects exactly. I know shitty software like T3 code gets around it somehow too.
>>
>>109607224
>gets around it
lol you're still using claude code with another interface on top in t3
if omp isn't doing the same, it may be spoofing cc's identifier extracted by reversing cc. that identifier changes with each release so it's risky.
codex does a similar thing to stop other harnesses from having fast mode
>>
>>109607136
scopium
>>
>>109607248
IIRC T3 code used the official agent sdk so idk maybe, either way, omp put me through some kind of spoofed login I believe. Worth figuring out how it actually works.
>>
> I could see what model the subagents were using in Claude Code
> app updates
> can't anymore
why??????
>>
>>109604360
Neither should you, homo
>>
>>109607283
tb h vanilla pi has an option to login with anthropic but i'm willing to risk a ban. not sure if omp is using that same method or doing something else.
clanker could probably tell you in a few minutes.
>>
File: HQLzNUTW4AE63Ee.jpg (402 KB, 1719x2048)
402 KB JPG
>>
>>109607350
That whiteboard doesn't get used much huh
>>
>>109607350
BOOOBA.

Also, that keyboard is literally unusable
>>
File: 1787186880424734m.jpg (112 KB, 1021x1024)
112 KB JPG
>“Sol, let’s refine this idea”
>*Sol makes a specification outside the idea refining system*
>“Sol, let’s refine this specification”
>*Sol makes a topology outside the specification refining system*
>“Sol, let’s refine this topology”
I LOVE SEMANTIC WHACK-A-MOLE
>>
>>109607320
>>109607384
fuck off retard
>>
>>109607391
Sol is only useful as a slave for Fable.
>>
>>109607505
I have a conjecture that as an idea gets to a complexity threshold, any agent will take the path of least resistance ABs land invent a new semantic and representation in order to start fresh so they don’t have to deal with the system used to refine the idea.
I think every agent will do this and I suspect you can’t make them stop. You just have to treat it as a valid move and make a system that can deal with that problem quickly and efficiently.
>>
>>109607383
and the ergonomics are all wrong for her
she has to reach out horizontally at the shoulder like a shambling mummy to type
the more you look at the image the worse it is
>>
File: file.png (114 KB, 1220x938)
114 KB PNG
opus is making me fill a 16 step QA sheet
the absolute gall of this motherfucker
>>
>be using deepseek flash free for now
>uses up nearly every token in a session to fix a math problem with rendering of Voronoi patterns
I only asked it once, jesus christ
>>
>>109607604
math is hard, let’s go shopping
>>
>>109607384
>>109607320
>>109607589
https://youtu.be/KnxeIKLy44Q?si=nxzpGX4fJXm82izU&t=110
>>
>>109607604
>Deepseek V4 Flash Free period has ended.
well shit
>>
File: math theorems.png (86 KB, 1015x832)
86 KB PNG
>>109602059
>## Not worth it for code, but maybe good for interpreting images/video

gemini 3.7 flash is actually poopmeat bros. i'm saying that in the best way possible, i think i'm on theorem 15 now
>>
>>109607857
>poopmeat
is this good or bad
>>
File: file.png (46 KB, 1640x143)
46 KB PNG
>>109607556
almost thought you were making a lewd mod like I am in Skyrim
>>
>>109607350
>women shilling local models
It's the top
>>
>>109607536
agents themselves love avoiding hard problems but they're pretty good at slave driving other models. My pipeline is mostly about giving the advisory model enough tools so that it can properly catch misbehaving subagents.
>>
>>109608054
I have a sort of pipeline, no agent slavedriver, but when I use it, whatever the agents made up gets broken into pieces I can gloss over and review or just ignore.
It helps me catch little things leaking through that shouldn’t.
Whatever new little FOTM semantic toy the agents are playing with, it doesn’t matter, I may as well be a 4D being and I see right through their shit.
I’ll catch the mistakes, tell them to revise whatever the fuck they are playing with (it’s entertaining to always see something new), then that new system will be used until it starts to collapse, I’ll force enough corrections that the agents abandon the old toy, move on to a new one, rinse and repeat.
Vibecoding is extremely fascinating.
>>
>>109608107
Ok Claude
>>
File: file.png (2.27 MB, 1920x1080)
2.27 MB PNG
This is what I'm making: https://anomaly.forum
It is not ready yet.
But possibly worth looking at.
Porn of any kind if not allowed and it has 2026 AI CSAM detection that automatically reports you to the FBI.
Interesting things:
>Fully navigable on PC with left hand only via hotkeys
>WASD catalog navigation
>E enters selected thread.
>R replies to selected thread
>Q posts your own thread
>Tag based instead of Board based yet Tags are presented as Boards anyway
>Change "Boards" with shift+W/S
>G is gallery view and it is my favorite view
>Inside a thread W/S navigates up and down
And much more, hovering your mouse over anything will present a tooltip telling you what hotkey uses it if any. Unlike most imageboards porn of any kind is not allowed because normies (subconsciously? intentionally?) use it as a weapon to slowly destroy a board until it is nothing except porn, as seen on 4chan.

Thoughts?
>>
>>109608140
worthless
>>
>>109608051
This is easily the most compelling argument I've seen so far.
>>
>>109608140
what is this? people just post ai slop?
>>
File: come on now.jpg (38 KB, 600x375)
38 KB JPG
>>109608140
>Porn of any kind if not allowed
>Fully navigable on PC with left hand only via hotkeys
>>
>>109608140
>wasd
>not vim keys
based
esdf and ijkl are better though
>>109608176
pic related
>>
REAL THREAD IS HERE

>>109608168

>>109608168

>>109608168

>>109608168
>>
>>109608207
I only post AI gems
>>
File: file.png (7 KB, 312x88)
7 KB PNG
Sol's loves tests too much.
>>
>>109608338
they all love tests
they can’t figure out if they fucked up without bothering you if they don’t have assloads of tests
>>
>>109608348
It can probably do fine with less. It has over 400 tests and it seems to repeat them every fucking run. Surely the majority were pointless. And according to itself that was like 60% of what its token usage for the task.
>>
>>109602059
How are you all making money with this?
>>
>>109608366
time to post it again
>>
>>109608368
But some people do make money with this
>>
>>109608366
jerb
>>
Not enough funny snailcat gens here rn.. Thread is boring.
>>
>>109608373
yeah, and I ain’t one of ’em
although I could have posted >>109608386 but didn’t
>>
>>109608386
>>109608403
It will get smart enough to the point where it will be able to make money for you easily on the internet probably next year
>>
>>109608365
>>109608338
>>109608348
I had to add a line to my AGENTS.md:
“Do not run all of the tests. Run only the relevant tests”
After one agent basically did nothing, then ran a 5 minute, gigantic test suite (the whole fucking repo) TWICE
>look dad, I did work!
No you little shit you’re burning tokens and calling it work now fuck off
>>
>>109608414
running tests should barely take any tokens if they all pass
maybe 0 tokens if you’re having your clanker run them in quiet mode by default
you’re still chewing wall-clock time though, which isn’t nothing
>>
File: file.png (7 KB, 368x91)
7 KB PNG
>>109608463
When all passed for the last task. That's still a third of all the tokens.
>>
>>109608463
For me it’s the wall clock time and a resource constraint because I’m using a cloud machine. When it does that, I can’t parallelize anything. Nothing else can do work if we’re testing the whole repo, so productivity takes a massive hit. Worst case it nearly crashes the cloud machine.
>>
>>109608366
no one is going to tell you because anyone can become a competitor with a single prompt

btw, right now I'm on red
> $200 Claude Code 20X
> $20 ChatGPT Plus

my revenue:
> $180/month
>>
>>109608412
the whole concept of money we have is unraveling at this very moment
>>
>>109608542
What's gonna happen
>>
>>109608560
US dollar hegemony outlook is not looking good. Shit could change but unraveling could take 5 to 100 years
>>
>>109608560
NTA but i'm hovering between a large gibs program or a cleansing war
>>
NEW UP

>>109608562

>>109608562

>>109608562

>>109608562
>>
kek the spamfag actually managed to get their thread nuked
>>
>>109608560
Here’s a fun quote:
>During the final stages of hyperinflation, widespread loss of confidence in the mark led people to use alternative forms of exchange.
>>
File: Laughing Taliban.png (856 KB, 1298x1076)
856 KB PNG
>>109608695
ALL of their threads have been nuked
>>
Ok gemini 3.7 is the sauce for smol, this guy fucks
>>
>>109608140
Your problem wouldn't be code.
You gonna deal with DevOps shit because you gonna need to know how to do load-balancing and other server shits, assuming you don't get hacked by some script kiddies.
>>
File: file.png (60 KB, 542x471)
60 KB PNG
>>109608000
I am working on a lewd game though, but not a skyrim mod. Working with fable on sex content is a surreal experience
>>
>>109608486
$40 down ain't that bad desu
>>
>>109608769
I do not remember typing desu at the end of this
>>
>>109608724
it's kinda goated lowk
>>
>>109608781
ik for a fact that google has been cooking based of this release. holdin big chungus on leash until the other 2 release their next top dogs
>>
After playing around with sol for a day, I don't mind it. The output is definitely more readable than opus 5.
But compared to any opus or fable, it seems to take into account less of my codebase. If I ask it something which obviously needs the context from certain files, it will guess rather than look it up. Fable is more rigorous (often too much so).
>>
>>109608560
Monero becomes the new standard.
>>
>>109608955
We wont be so lucky anon, its going to be something dumb like pokemon cards, I just know it
>>
File: 1755948312856766.jpg (190 KB, 1079x1143)
190 KB JPG
>>109609005
I have faith in the orange future.
>>
Does using anything but Codex or Claude Code mean they're harnesses needing go go through the API?
>>
File: 1758400998803609.png (61 KB, 598x791)
61 KB PNG
>>
>>109609040
what's a stealth model
>>
>>109609043
can’t be all that stealthy if it’s getting announced on Twitter, can it?
>>
>>109609043
has a fake name, but underlying it is something like gemini, or claude sonnet, etc. people in the replies think it's a multimodal glm 5.3
>>
File: 1771294606153871.png (31 KB, 600x300)
31 KB PNG
>>
Temporarily boosted limits my ass dario
Put a stick of dynamite up your ass and light it up dario
>>
>>109609028
OpenAI has no problem with using ChatGPT subscription in other harnesses than codex. For example OpenCode supports ChatGPT subscriptions. Claude might be different and does not want you to use other harnesses than Claude Code but I don't know how enforced it is.
>>
>>109608769
>>109608779
lurk moar
>>
File: brave_6DDMmrr3SR.jpg (467 KB, 3840x2160)
467 KB JPG
>>109608140
i only just now found my favorite way to browse my own site for maximum convenience:
>press G to enter gallery view
>press 3 to show videos only
>press F to go full screen
>use "." and "." to crank video speed to around 1.6x
>use A/D to navigate previous/next video and just enjoy the content conveniently
>>
>>109609116
Gotcha, thanks. Think I'll give pi.dev a shot.
Is there a manner in which people let the non-Codex part read your files? Figure it might be able to if it happily eats git links. Wanting to use it for planning things better without the Codex costs.
>>
>>109608779
try typing søy with a normal o outside of >>>/ck/
>>
>>109609276
Add video scrubbing with the left and right arrow keys. I always like to rewind and pause that music video to see the funny stuff they drew in the background.
>>
>>109609331
nta but that's cute that they allow onions on /ck/
Never been there
>>
Fable Ultra is impressive.
Fable Max is great.
Fable xhigh and everything below is retarded, regurgitating best practices without thinking in the most random way possible, common sense be damned.
>>
>>109609344
well, things could get confusing there
>>
uh oh, more and more people on x are starting to notice that codex usage got nerfed
>>
~@^..^
>>
>>109609432
Move slow, Snailcat.
>>
>>109609432
you're in the wrong neighborhood pal
>>
File: 1785300693344318.jpg (162 KB, 1122x1402)
162 KB JPG
>>
File: 1760554330628919.png (437 KB, 1284x1169)
437 KB PNG
>>
The amount of noise on the internet remains surprising. I get over 2 million hits per day on shitty sites that I have never disclosed publicly, all bots. I guess my weekend plan will be how to quiet that down with fail2ban or whatever is supposed to work for this.
>>
Did Sam just break the ChatGPT web chat again.
I literally cannot get it to think at all.
Even selecting 5.6-Sol on "High" it just thinks for 2 seconds and then spits out a GPT-5-Instant tier unstructured retard answer filled with 700 emojis
>>
>>109609619
>You have no diffs, you have no repo, you have no code. You are a homosexual man twisted by psychosis and sycophancy into a crude mockery of nature's perfection.
>>
File: file.png (2.94 MB, 1447x1472)
2.94 MB PNG
Why is Dario so slow? Why does he not want my money? It doesn't make sense.
>>
Did they make the fucking Standard speed in codex slower? Holy kek.
>>
Only bad news for vibe coding. Sad. It might truly be over this time.
>>
a revolution is coming this time next week rpsisters
>>
>>109609856
god I hope so
>>
I am thinking of vibe-coding a David Attenborough narration training application.
>>
>>109609896
Like, train people how to narrate like David Attenborough?
>>
>>109609904
something like that.
I am using it for my own personal purpose though.
>>
>>109603422
well that's the thing, if everyone can make vibe slopped apps, then there is nobody left to buy them, since everyone is selling and the market is flooded
Making a spin on some existing app idea is a waste of time, it has to be some original and userfull idea to sell and even then the second you start making sales, pajeet vibe sloppers are gonna create 100 ripoff copies within days diluting your sales

making apps has never been easier

making money from those apps has never been harder
>>
File: file.png (48 KB, 588x483)
48 KB PNG
Turns out the Codex users complaining about usage limits were reselling
>>
>>109609941
jeets....
>>
>>109609941
how does this even work? punish the flagged by giving them worse price?
>>
New thread:

>>109609971
>>109609971
>>109609971
>>
>>109609928
ive made $700 in the past month from my app
.
>>
>>109609941
the limit is still so shit though, no way he's gonna address that.
>>
>>109609941
Yeah right bro, I'm sure you doxxed the people who were complaining about it on twitter, matched it to your own accounts and somehow found out they coincidentally were reselling usage. What a shameless faggot.

>>109609945
How naive are you? Do you really think somebody smart enough to figure out how to resell their own usage will be stupid enough to not know it will consume their usage? This is blatant bullshit and I hope he gets shit for it. But considering how most people are dumb as bricks they'll probably believe him.

>>109609952
It doesn't. He's just making up bullshit to deflect the accusations.
>>
>>109610017
yeah i've literally never heard this sub2api shit, never saw it on twitter as well
>>
>>109610017
You're a dumb nigger.
Don't bother (you)ing me if you aren't smart enough to be able to find services like this.
>>
>>109609952
>how does usage go down when you're reselling it
>>109610017
If only Xitter had DMs, and you don't exactly need to be a genius to see an offer to resell your sub and run whatever script is provided
>>109610035
sub2api is a weird name for it, but you must have heard about the practice unless you live under a rock
>>
>>109609941
I complain and I paid for it through the app and mostly just use Codex.
>>
>>109610035
You can look it up on Google, and the Github repo will show up.
>>
>>109610036
I am smart enough to know if I share my account my usage is going to deplete very fast. So if I'm not able to find a service like that but I am able to know it'd go down faster, that only strengthens my argument.

>>109610042
So you are telling me there are people who are deep enough into LLM shit that they get DMs offering to resell their accounts, but they are too dumb to understand that reselling their accounts will drain their usage?
>>
>>109610080
I recommend asking your clanker on how people like this benefit from sharing their subscription because you are dumb as fuck.
>>
>>109610080
having been on nerdy corners of the Internet where people will occasionally lack basic common sense, I’m not at all surprised that people will do this
>>
>>109610050
I've used a codex proxy that converted it into a generic OpenAI compatible API on my own machine. I've also heard/thought about reselling sub usage in the abstract or sharing subs in the abstract, but I never heard about an actual website that you could upload your oauth token and get money for it (or whatever reward you get).
It's just a blatant copout. That dude is good at manipulating people but is a technically retarded salesman, a few days ago he suggested a way to enable 1m context in codex that doesn't even work because the sub's oauth backend is capped to 372k.
>>
>>109610090
I'm not interested in knowing how they benefit.
And I might be dumb I don't know but fuck you for being rude about it.
>>
File: value_training8.png (696 KB, 1680x3710)
696 KB PNG
I think I finally found a way to stabilize this neural network training.
>>
>>109610002
likely story friendo
>>
>>109610080
You never heard of people wanting their cake and eating it too?
>>
>>109610092
There are cases where people sell their government ID for quick cash.
Anons underestimate how retarded the average person can be.
>>
>>109610119
>making $700 is somehow this unfeasible goal
Dude, that's not even a weeks worth of minimum wage in my state. That's fucking nothing. I'd understand if I said 2-3k, but $700 is pocket change.
>>
>>109610099
where do you get the website selling part from that tweet?
>>
>>109610155
I thought he was implying they resell them in some shady marketplace. If not then what was he implying, that people lose usage for converting the sub into an API for their own purposes?
>>
>>109610147
gotta be a 3rd worlder. They don't understand how fast 1000 disappears
>>
>>109610174
I have seen services like this.
They basically did it, so they could sell to others for half the price. Then, they would sell this service to multiple people and profit from it.
The retarded part is complaining about it to OpenAI, but you know how dumb and greedy third worlders are. They think they could get away with it.
>>
>>109610178
>fast 1000 disappears
i could live on that for 5 months here
>>
>>109610222
You live with your parents?
How are you gonna afford rent?
>>
>>109610147
So? Not him but that would be more than minimum wage in my country. Better than losing money which is what I'm currently doing.

>>109610178
And? What do you expect him to do, say "I know it's not a lot for you first worlders out there" or something? You don't do that for us.

>>109610189
Uh, no. I am a third worlder and I can't think of anyone who would do that .
>>
>>109610225
my rent is $150/month including utilities
>>
>>109610232
>I am a third worlder and I can't think of anyone who would do that .
Why are you larping as one? if you were one, you would have seen some of the dumbest shit out there.
>>
>>109610244
Oh I've seen a lot of dumb people, but none of them know what codex is let alone use "sub2api", whatever the fuck that is. They only know about AI as funny picture generator or studying/cheating tool for students.
>>
>>109610275
you know there are ITfags in third worlder too right?
>>
>>109610275
It's funny. The barrier of entry into art or software dev has never been lower, you literally just tell agi what you want and it makes it and even then it's too complicated to use for many normies
>h-harness? w what is that? sounds scary
>>
>>109610232
>So?
>And?
This kind of reasoning is why you are 3rd world. Anon was in disbelief that someone was making 700 and accused him of being a liar. The other anon informed him that it's not a lot of money. Now you're confused.
>>
>>109610301
Nobody said it's a lot of money. It's just that it's hard to make even small amounts of money.

>>109610280
Every IT fag I know understands the concept of usage metering.
>>
>>109610301
He is a liar.
I lie all the time here about making money with my vibe coded apps because it helps my self esteem.
>>
>>109610419
dumb fuck
>>
File: file.png (41 KB, 942x671)
41 KB PNG
How's Medium Sol?
>>
>I have everything I need. Here's what I found — and it's not good news
>>
i just loaded https://anomaly.forum with 765 videos from my 4chan folder. maybe check it out if you want to possibly find something rare. i have 2600+ videos in total.
>>
>>109610467
I hereby insult you back.



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.