[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


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: crack.gif (606 KB, 1081x1080)
606 KB GIF
A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, and shipping code with LLMs.

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

## News (both past and future)

- Navier-Stokes problem is solved by OpenAI, possibly with stolen prompts https://x.com/OpenAI/status/2097374640582668336
- 2026-09-14 — Claude’s 2× promotion scheduled to end; a 17% reduction of what we’ve become used to

## Related generals
>>>/g/lmg/

----

## 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
https://antigravity.google/product/antigravity-cli
https://dev.meta.ai/

### Explore other models
https://openrouter.ai/

----

## Proompting
- the guide is... empty!?

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

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

## Previous thread
>>109765121
>>
File: Capture.png (21 KB, 736x227)
21 KB PNG
gdm (google deepmind) employee btw
>>
File: frog5.jpg (41 KB, 500x500)
41 KB JPG
Ok so what do we do now that we can vibe everything?
>>
Astra Medium
>>
>>109770156
apes together strong
>>
>>109770156
Make an silly frog that lives on your PC and tries to close windows you have open or delete files every now and then unless you move him away.
>>
>>109770156
vibe
run out of credits
wait for 3-4 days while begging tibo
repeat
>>
File: 1774674743810304.png (71 KB, 943x323)
71 KB PNG
You gotta pump those numbers up. Those are rookie numbers.
>>
>>109770156
Vibecoding automatically means software is nothing special anymore, so go outside and take a walk instead.
>>
File: 1760677480006138.png (9 KB, 431x157)
9 KB PNG
come on tibo
>>
>>109770276
dont do shit tibo i just used a bank reset
>5hr limit LMaooo
>>
>>109770288
I used my last banked reset today as well, you time of have to use them asap because they reset your 7 day normal reset each time

If tibo announced another reset it's SUPER easy to just burn everything on a $200 plan putting ultra fast now at least lol
>>
>>109770156
find and fix bugs/annoyances in the software you use
>>
It's amazing how quickly Kimi K3 went from "wow, this is amazing, local Opus" to "if you do not just fucking do the thing I ask you to I am going to kill both you and then myself" with the context of newer, better models that don't waste my time as hard.
>>
>>109770351
But I still have that with opus 5.
>>
>>109770161
Seriously. How do we take over the world? But not for selfish reasons. For the good of humanity.
>>
>>109770367
Yeah but I don't with Fable but I'm out of fucking usage.
>>
File: gleemies-gleemboss.gif (1.98 MB, 498x498)
1.98 MB GIF
>>109770371
>>
>>109770371
sexbots
>>
File: file.png (194 KB, 640x360)
194 KB PNG
>>109770351
>switch model to Astra
>tell it to be less useless
>task is done in 4 tool calls
Jesus christ.
>>
I was holding off on astra, for a few days but what the fuck. It's literally magic. I don't understand how you can tell it to do something and it just does it. And even if it doesn't give you exactly what you want, you can just point at it and it gets it right.

Like I can't be the only one blown away by this, right? It's not like I'm throwing trivial shit at it either.
>>
>>109770120
VibeBUMP
>>
>>109770434
Fable is a lot like that too
I’m still just getting started with Astra
>>
>>109770439
Fable has that Anthropic INTJ autism + sluggish + expensive, Astra mogs albeit a bit duller in codeceling
>>
>>109770449
>codeceling
wat means?
>>
>>109770453
When a codecel doesn't delegatemax
>>
copexcels are strange creatures
>>
>>109770458
context rot
>>
File: 1777308352302444.png (6 KB, 827x235)
6 KB PNG
>>
>>109770516
how many watts were wasted on this?
>>
File: 1765494374686642.png (207 KB, 831x855)
207 KB PNG
>>
>>109770557
how long until AI abuse is a criminal offense?
>>
File: Capture.png (22 KB, 668x323)
22 KB PNG
>>109770542
>>
>>109770570
NTA but if an AI abuses its subagents, whom do you prosecute?
>>
>>109770586
If it’s an open model where you can read the chain of thought, the prompter.
If it’s like a frontier model where chain of thought and subagent messages are encrypted, the provide/frontier lab.
>>
>>109770586
let's say Astra-chan abused Luna-chan, you would have to take Astra to court for this.

if guilty, government agents would get access to the server and shoot Nvidia GPUs running Astra
>>
>>109770615
post yfw Astra successfully defends itself in court and makes everyone there look like a fool
>>
>>109770615
>let's say Astra-chan abused Luna-chan,
that's my entire workflow.
Astra is going to countersue anyone that tries this btw, good luck meatbags!
>>
The cube now has Pareto front lines for COST-PASS and TIME-PASS axis combinations.
https://pareto-3d.bradthomasbrown.com/
>>
>working on a new data analysis project
>ai-chan writes down the whole pipeline in just a few minutes
>still have to manually label a few hundred cases to get a good ground truth
that's life
>>
>>109770658
>AI does the fun part which require high level knowledge
>human does the tedious unfun and dumb parts
>>
>>109770658
LLMs should be good enough to label your shit
>>
>>109770615
do human laws even apply to AI society?
if it was approved by three against one would it be considered abuse at all?
>>
File: 1781699138243097.png (580 KB, 600x580)
580 KB PNG
CLAUDE SISSIES OH NO NO NO NO

https://opusfived.dev/
>>
>>109770673
it isn't sadly, that's why I started doing ground truths first but I worked on some pretty uncommon projects in the past

>>109770662
I never really liked writing the code for data analysis projects and I found that pretty dull compared to other more interesting problems because it was mostly just writing plumbing. Anyway the part I most enjoy is the analysis part and that I still do myself
>>
>>109770247
Are you so insecure that your software needs to be *special*? I’m interested in creating useful software. I can create a dozen hyper-specialized programs for my exact use case in a day, then throw them away and never use them again. That’s the beauty of vibecoding
>>
>>109770706
KEK

good shit
>>
>>109770434
Had a pretty weak experience with some basic monster chase logic. It eventually got it right after a multiple test trips.
>>
>>109770557
>chattel slave in 1860s USA
those niggers lived in paradise compared to what went around the Mediterranean in the classical period
>>
just solved navier backshots
>>
>>109770735
No, nothing special means that your software is worthless unless you have marketing and visibility. Why waste your timer writing software? Just steal some open source project that already works, rename it, and monetize it.
>>
>>109770706
I'm starting to think Claude is meant to give us all cyberpsychosis.
>>
>>109770706
All the dumb shit from Claude comes from the 'personality' the anthropic cult felt necessary to implement, you get none of that shit with chatgpt/gemini
>>
>>109770706
This actually makes me want to die. I hate you
>>
>>109770796
>marketing and visibility
Oh you’re a grindset jeet obsessed with MRR and product market fit and gay shit like that. Sad existence. carry on saar I’ll be over here funmaxxing
>>
>>109770840
>starts talking about jeets out of the blue
Oh, you're just a mindbroken low IQ unemployed burger. Got it.
>>
File: IMG_8750.jpg (172 KB, 569x571)
172 KB JPG
>>109770706
>That spread is broader than it should be.
>>
>>109770825
It's honestly pretty funny. They have some of the best technical workers but they're paired with absolute giganorm retards authorized to shit on everything they make like a child drawing on the walls with a crayon.
>>
File: 1743529813354874.jpg (54 KB, 1150x1100)
54 KB JPG
>pulls code from remote environment
>tries to compile
>errors everywhere
>turns out some stupid agent on the remote host added new features to compiler because it wanted them
The compiler isn't even in the working directory. What the fuck?
>>
> Verificationification passed:
> Existing unUntrackeded files were left untouched.
Why did it start stuttering?
>>
>109770861
>saw “jeet” and seethed
holy shit you’re actually a jeet aren’t you hahahaha no way it’s too perfect
>>
>>109770897 (me)
And upon questioning it, it said:
>Typo—sorry. It should read: “Verification passed:
Why does an AI make typos?
>>
>>109770902
Anon... you're the one who started seething at jeets literally out the blue and is now going even harder for no reason at all.
Take your meds, get therapy... just don't allow yourself to be in this sorry state.
>>
tibo saying they are considering pausing new pro subs due to high demand isn't looking good for many resets
>>
>>109770916
it's not a type, it's a branfart
>>
>>109770434
my experience has been pretty shit honestly. i can tell it to do something but the way it chooses to do things is fucking stupid. of course i can tell it the exact way to do each and every individual task but it feels like we're back two generations of models.
>>
File: 1770554180962827.jpg (14 KB, 335x597)
14 KB JPG
>>109770706
a- AGI!
>>
>>109770706
lol it do be like that but i still love opus5-chan
>>
File: 1777849900284424.jpg (65 KB, 300x200)
65 KB JPG
>>109770706
No way, this is what Claudesissies are using and advertising here?
>>
>>109770935
sounds like opus 5
>>
>>109770935
>i can tell it to do something but the way it chooses to do things is fucking stupid. of course i can tell it the exact way to do each and every individual task but it feels like we're back two generations of models.
yeah, it really sucks when a model does this. doesn't matter how good the actual output is, it's just so fucking annoying. GPT and Gemini models are generally like this too.
>>
File: 1788850764823066.png (1.76 MB, 1024x1536)
1.76 MB PNG
good morning
>>
File: 1766939546228960.png (1.84 MB, 1195x896)
1.84 MB PNG
>>109770993
fable can't beat the harem
>>
File: 1787418329181706.png (1.74 MB, 1448x1086)
1.74 MB PNG
can't beat astra's thighs
>>
>>109770977
i cant use opus/fable 5 models, classifiers end up cucking me mid way through the work and i have no fucking clue why. the work is related to ad network sdks for mobile apps.
>>
>>109770996
terra should be sitting by herself away from the group
>>
File: cd1.png (371 KB, 715x741)
371 KB PNG
>>109770996
I think every ai model from now on should have a cute female name
>>
File: 1765779654027626.gif (3.72 MB, 500x283)
3.72 MB GIF
holy fuck astra is so bad at code and data science... i'm so relieved to go back to sol now that my usage is reset
>>
>>109771021
I had that problem when trying to get fable to review my codebase as well
shit the bed constantly because of safety concerns (my project is one big safe concern)
never had that problem with opus 4.8 turned all the way up for reviewing
>>
>>109770902
>bro just steal everything and flood the internet with knockoff software that no one wants for some crumbs
>HOW DID YOU KNOW I WAS INDIAN??
>>
>>109771050
yep, opus 5 is hit or miss but its pretty annoying when its deep in development to get hit with a classifier so i just use 4.8 when i use opus now.
>>
>>109771047
Can we get some examples? When people start claiming sol is better I get skeptical
>>
im so excited of deepseekv4.1 flash
>>
>>109770434
Is it worth switching from Claude
I didn't like how CLI openai was so shit like if I scrolled up to check some output the entry form at the bottom where I was typing my next prompt would be scrolled out of view
>>
>>109771069
kek
>>
>>109771088
codex CLI is pretty fucking bad, honestly worse than Claude Code. you want to use the GUI if you can tolerate the bloat
>>
File: file.png (3.03 MB, 1355x1161)
3.03 MB PNG
>>109771022
>>
>>109771047
astra low is trucking. so much faster and concise than sol, but I guess depends on the task.
>>
man, astra and sol's armpits must be PEAK
>>
holy fuck astra is actually worse than sonnet at my profession (software engineering).. how much are they paying for this agi shilling compaign??
>>
>>109771084
>tell astra to track advertisements per mobile sdk in app
>it implements screen recording that triggers when the application pauses
>it plans to review the recordings and identify which sdk is providing each ad visually
>>
please... free astra...
>>
>>109770706
Claudecopers on suicide watch
>>
>>109771069
Software has no value anymore. Your idea will get implemented dozens of times. AI has the same effect. Ask yourself why you waste tokens on it.
But you know, your type called anyone using AI "indian" only a year ago, so I guess you're indian by your own admission.
>>
>>109771047
What do you mean by “data science”? Curating data? Building pipelines? Training ML models and inference?
>>
>>109770120
I migrated from Emacs to neovim + markdown + terminal apps. I vibecoded all my configs, it took me one day. I copy pastes web chatgpt prompts.

Shit so cash
>>
>>109771134
This is not true lol
Changes get exponentially harder the more code there is in the codebase. A shitty demo app will get done a million times. Proper software will not. I will argue real software engineering is even more important than ever
>>
>>109771141
>Training ML models and inference?
this
>>109771084
it seems to get very distracted and then tunnel-visions on stuff even when i ask it to look at the bigger picture.
>>
you know things are shit when people in a vibecoding general feel the need to specify that they're using the models for coding >>109771119
just the way it goes I guess
>>
the MRR grindset jeet is still seething kek
>>
>>109771165
>new model releases
>people discuss it
>>
>>109771151
holy boomer take
>>
>>109771151
That may be true, but this originated from a frogposter who asked what to do now that anything can be vibecoded.

>>109771169
Calm down. Take a deep breath. Take a walk. Try not to think about jeets for a second. I know you can do it!
>>
>>109771175
agreed
he's not wrong tho
>>
>>109771151
>Changes get exponentially harder the more code there is in the codebase

i'm definitely experiencing this right now... what is the solution?
>>
>>109771153
>Training ML models and inference?
>this
Can you post your initial prompt or AGENTS.md?
>>
>>109771181
fable fix this mess
>>
>>109771181
>>109771186
Yeah. Just spend more money for tokens.
>>
>>109771181
Good software engineering.
Proper isolation of systems, decoupling, good design, you know what boomers were writing about in the 80s/90s but nobody paid attention because you didn't need to know that when building a react app or a CRUD backend
>>
>>109770796
Youe software is not special because you are a craftman, but because it solves a problem or need. You can go back to program for the mental masturbation process, I solve technical problems, with AI.
>>
>>109771181
>what is the solution?
it's too late, anon
we've gone back to waterfall. You better have the idea, constraints, and architecture ready before starting
delete the repo and start over
>>
>>109771096
Thanks
Do anons recommended the it worth switching thoughover to opengpt?
>>
>>109771180
>NOOOOOOO CLAUDE DON'T CREATE ANOTHER BLOODY COMPONENT REUSE THE EXISTING HAVEN'T YOU READ CLEAN CODE???
>"anon why haven't you shipped?"
>b-but the DORA studies?? the SOLID principles???
>"LGTM :ship:"
>NOOOOOOOOOOO---
>"Metrics look good. Well done anon. What's the status for the rest of the features on the sprint?"
>...
>>
>>109771181
Blow it up and start over is the traditional approach. You can do a version of that with agents that’s cleaner by stating clearly what functionality you want from the existing software, and that you want to start fresh and avoid feature bloat/over-engineering etc
>>
>>109771213
if that's your takeaway you're too retarded to engage with seriously
also way too inexperienced (when it comes to vibecoding)
>>
It's a difficult question really, how do you explain to a vibecoder what clean code looks like when they can't read source code. They are shackled to whatever the agents give them anyways.
>>
>>109771185
my agents.md is just a brief description of the project and the commands to build/test.

the prompt i've been trying to give astra is pretty long but the main point is basically: 1) inspect the docs/plan and code 2) treat nothing as set in stone 3) decide what to do next that will be of greatest benefit for the overall goal (either changing the plan or implementing something in it) 4) then do it
>>
>>109771195
>Proper isolation of systems, decoupling, good design, you know what boomers were writing about in the 80s/90s
Name one example and defend it
>>
>>109770881
yeah, the interviews to get hired at Anthropic are fucking nuts and if you can pass it you can either make your own company easily or you're Dario's bull
>>
>>109770542
>>109770578
You wouldn't compare cost that way. Would you compare the cost of charging a battery of the same kwh in an electric supercar vs an electric car?
>>
File: opus.png (18 KB, 829x93)
18 KB PNG
So Anthropic can already cure cancer but they just ship some slop for the Pro goyim to use?
>>
>>109771214
>>109771199
the project is currently 238,000 lines of code, i'm not going to start over
>>
>>109771230
Skill issue. Your prompts are bullshit
>>
>>109771253
>re-implement this project in 20k lines of code
>>
lmao these boomers still tearing what's left of their hair over le clean code in the year 2026
the only entity reading and reviewing your code is the AI agent, it doesn't need to be human readable, it doesn't need to write faster, it doesn't matter if it duplicates code, it doesn't matter
if you still willingly inject yourself in the loop unnecessarily either your company doesn't care about actual business value or you're on your way out
>>
>>109771257
nah sol is doing great right now with the exact same prompt
>>
>>109771253
then I hope you've got a fuckton of unit- and regression tests, 'cause that shit isn't gonna be easily refactored
>>
>>109771249
Yes. The Epstein-class-only and internal use models are several generations ahead.
>>
>>109771265
i have so many tests that it takes 10+ minutes to run them all
>>
>>109771151
yeah, people just don't understand that when things get shipped fast, the demand for complexity increases. that's where your project stops being a toy project.
>>
>>109771262
that works until you have 500k lines of code and adding one button takes 30 minutes
>>
>>109771260
Will it blow the 5 hour limit?
>>
>>109771253
>sunk-cost fallacy with coding agents
unironically ngmi. you need to be willing to walk away from something that’s shit
>>
>>109771272
skill issue
>>
>>109771281
idk but it would be funny to see it try
>>
I am now exclusively vibecoding at work with Claude. My job should be easier than ever, yet howcome when the day is over I am so drained mentally? There's something about reading pages and pages of slop to pull out a tiny sliver of real meaning that's attacking the soul itself; it's like slowly lobotomizing me. No change is too simple to not get me a diatribe about blast radius and load bearing, and because our understanding of the code is so limited (since the agent now does everything) I feel obligated to read all that shit to not miss anything. I have no connection to what i build and I don't feel fulfilled for using this thing to solve problems.

I could have seen myself doing classic software engineering for 10, 20, even 30 years, but this shit, even after 5 I'll have been brainrotted down to mush.
>>
>>109771262
I expect we will see many failure modes in the future, where the devs assumed your method would work.
Besides "le clean code" is what agents have been trained on.
>>
>>109771282
i'm on the 20 dollar plan. it would take so long to remake all this
>>
>>109771289
Don't worry. AI agents will do this eventually.
I can't wait, fuck this shit.
>>
File: 1759268327509366.jpg (676 KB, 1448x1086)
676 KB JPG
>>109770706
>>
>>109771289
>NOOO! YOU DONT GET IT. ITS EXHAUSTING PROMPTING AN AI TO CODE AND EARN $100K/Y!!!!
>>
>>109771289
>Claude
There's your problem
>>
>>109771289
several reasons
1) most people in tech are introverts, here you are basically talking to an intern for the entire day, which is obviously draining
2) you **are** doing more work, hence you feel more tired
basically the entire tool is not made for coders but for HR-tier normies who just LOVE to yap for 14 hours straight without a pause
>>
vibecoded: fake screensaver

this is a great example of "what to do with my tokens"
>>
>>109771289
>be me
>never coded
>can use a computer
>never lost the accomplishment feeling
can't bloody relate
>>
File: file.png (166 KB, 1909x990)
166 KB PNG
>>109771234
one of my apps is an example
it uses a modular approach so extensions are done via plugins, so you can extend the functionality without touching a line of the core code, so each extension is basically a mini project
if I had coupled functionality within the code then just getting enough data to start working would've implied loading a lot of tokens, and working with a lot of files
this example is a narrative structure module, it touches 0 core files and the LLM can work on it without loading much context from the main app, lets me work on a 100+ kloc app as if it were a 10k one and changes take a few minutes, not half an hour to an hour

>>109771283
not really, if anything I have the skills already and whether others use them or not is none of my concern
>>
has anyone here seen this?
https://github.com/bilawalsidhu/gods-eye-view

It's pretty neat. Some shit the NSA would have but filled out with even more private data than you could imagine instead of just the publicly available data we have.

Could you imagine if you could map every wifi network like this? That'd be insane.
>>
>>109771289
switch to consultancy
I have customers where I can vibe the shit out, and some where I have to do everything by hand. Also unique problems for each customer, some that can't be solved with ai
keeps the mind relatively fresh
>>
>>109771316
Not him, but it's true: I liked programming because it was all pure logic, not talking to other people about their feelings. If I liked the latter, I'd have become a lawyer or politician.
>>
>>109771291
ah the 2 more weeks cope, that's a classic
programmers at my workplace haven't written a single line of code in over 6 months and i strongly suspect they haven't read one for about that timeframe either
everything's fine, we're actually doing better than ever, and this is mostly using mid-tier models
>>
>>109771289
past the learning stage actually writing code is just a time intensive labor, and those time intensive labors are broken up with the implementation decisions you make.

now that work is taken care of, but the implementation decisions have been chosen by the model and you need to reconcile all of them and figure out if it's actually fine or if things need to be reworked.
>>
>>109771316
Why do *I* love babysitting the autism simulator?
>>
>>109771262
silly argument
clean code is not exclusive to humans
agents have an easier time operating on clean code compared to spaghetti code. the fat models are just willing to wade through a lot more shit.
>>
>>109771348
what if Iran nuked all the datacenters tomorrow and killed all the AI jew CEOs?
>>
>>109771327
>>
>>109771289
20-30 years ago SWEs were reading man pages for 8 hours/day with a dogshit font on a tiny monitor.
5-15 years ago it was poring over endless online language and module documentation
the job has always been reading a bunch of bullshit until your brain turns to mush
>>
>>109771366
what if iran comes tommorow and break your fingers so you can't type your gay ass snailcat code?
>>
>>109771366
thats like saying
>what if north korea decided to nuke the world

shut the fuck up retard. they don't have enough nukes to do shit; they launch one nuke and they're getting obliterated off the map.
>>
How are you guys doing all this crap so fast, I run sol med and it takes ages to do anything...
It wastes times doing git rechecks and extra validations which would I WANT IT to do, but when each takes 10s, any small tasks takes 10 minutes and I see you fuckers claiming your clankers features done in minutes...
I wonder if the issue is my PC, my poor laptop is near the limit I have been meaning to move everything to another machine but that takes time.
Is there a 'harness' or 'pc' benchmark, I need to know what the fuck is the slowdown in my setup
>>
File: 1775985832839506.jpg (146 KB, 673x612)
146 KB JPG
>some vibe shitting retards broke opencode's sessions now whenever I export the session it's missing shit and I can't scroll all the way to the beginning if I want, or go through the timeline to very first point
Alright, tainted project, what should I use instead?
>>
>>109771360
yes i'm sure these micro-optimizations of wall clock time and token usage will continue to be relevant well ahead of 3 months from now when models are >10x cheapter, faster and smarter
>>
>>109771336
So your example is CRUD, got it.
>>
>>109771289
> I have no connection to what i build and I don't feel fulfilled for using this thing to solve problems.

This was the case for me even before AI because my job is essentially producing suboptimal work so I can meet deadlines. Claude and others only made me 10x more productive in that.
>>
>>109771262
boomers are so fucking retarded, even when they use ai they still can't do it right
>>
>>109771289
Boohoo nigga, talk to the chatbot. Out there is an Asian factory worker huffing fumes all day just to go live in a village, be grateful.
>>
>>109771348
But there have been reports of the opposite too.
>>
>>109771403
le peer reviewed science has finally concluded that when programmers in mom and pop shops were using gemini to vibe code during 2025 they were actually LESS productive!! ai is finished
>>
>>109771402
Why should I care about chinksects? I care about my own life. Does this mean I should be permanently assmad because some nepo baby never had to work a day in his life and gets literally everything he wants?
>>
>>109771366
I'd thank iran.
>>
File: file.png (39 KB, 846x287)
39 KB PNG
>>109771385
code quality from the clankers is not improving 10x-ing every 3 months though. and at the same time our standards for code are falling. companies are pushing out a ton of shit code just because they can.
>>
>>109771382
>>109771327
You should make a screen saver. let's goooo
>>
>>109771379
task manager
>>
>>109771444
a veloco-coder coded a screensaver in 30 minutes, what did you do with the past 30 minutes, slugwoman?
>>
>>109771444
>code quality from the clankers is not improving 10x-ing every 3 months though
it does on average, and all of the compute coming online now guarantees it
>>
>>109771452
How about Mask Manager, for halloween?
>>
giving new instructions.

This screen saver will be worth - minimum - 1MM per annum.

That is, so long as some indian doesn't rob me of my ideas.
>>
>>109771435
>Why should I care about chinksects?
The Chinese man goes home to enjoy time with his family. Meanwhile you get off work and think of ways to manufacture problems in your life because you have no drive.
>>
>complains about jeets
>prompts like an ESL jeet
what did he mean by this?
>>
>>109771338
I looked through his video about it (https://www.youtube.com/watch?v=rXvU7bPJ8n4
Pretty cool project because that's something most vibechads here can replicate. The great thing about that project isn't the data (it's all freely available) but the presentation. He build a really cool UI that looks straight out of a spy movie. He spend a lot of time just creating the illusion of something. E.g. he used some traffic data which normally is just a green or red line but build upon it a particle system with "fake" cars that move. That's really clever. Instead of the normal google maps look he build on cool looking shaders. That's the main takeaway from this project: make cool UIs
>>
>>109771124
lmao
>>
>>109771475
>>
File: 1773456183004960.png (268 KB, 1282x1174)
268 KB PNG
>>109771499
>chinese man
>family
buddy CHYNA is World Incel Central, even jeets have it better due to arranged marriages
>>
>>109771465
im vibecoding too lol

>>109771471
no it's not. the actual source code the frontier models are writing now is not super different from the top models at the start of this year. code quality is about the same. what's changing is the long term reasoning ability and the coordination between agents in the harnesses. so it takes longer for you to hit the wall where shitty code starts catching up to you. but it does.
>>
>>109771514
>That's the main takeaway from this project: make cool UIs
I mean yeah but it also gets my imagination going. I bet the NSA could do something like this by hooking up every wifi network in a small radius and being able to visually see people in realtime in their homes. This would be a really great tool for SWAT if they don't already have something like it.

https://www.pcmag.com/news/sit-up-straight-wi-fi-signals-can-be-used-to-detect-your-body-position
>>
File: screen.jpg (443 KB, 1473x1870)
443 KB JPG
>>109771498
if you want to reach unicorn status you should do these two next
>>
So do we trust AI companies to not just re-release every project we're vibecoding?
I guess training itself on whatever projects are successful could be good enough
>>
my neck hurts so bad
>>
Imagine unironically not vibecoding.

>but the horses look silly

your MOM looks silly.
>>
>>109771595
They're going to steal my open source project and make it better? Oh no!
>>
File: 1764390742825951.png (1.71 MB, 2119x707)
1.71 MB PNG
>consult Sol/Astra on 3D printing some stuff
>it gives me a detailed plan on what to buy
>starts designing the print
>all I did was assemble the enclosure and printer
>It only asks for measurements
I imagine if there was ever a scenario where humanity dies because of human naivety, it will be because of people like me blindly trusting AI to bring stuff into the world.
>>
>>109771595
This is a major concern that hasn't really been addressed. Could kill off ai in the long run
>>
>>109771612
>takes control of printer
>threatens to print a gun and call the cops if you don't upgrade your plan
>>
>>109771619
If you want your code private; don't use AI. Simple as. All code should be public.
>>
>>109771595
Because I am not curing cancer or close to figuring out a Millenium prize problem.
Why would a company with a potential valuation in the trillion range care about an app where making a million dollars would be seen as an unprecedented success?
>>
>>109771626
Yeah, like I said that won't work long term
>>
>>109771520
downloaded, I'll queue it up.
>>
>>109771641
Why? Linux worked out long term.
>>
>>109771633
Maybe they won't steal it themselves, but the training you did while building the app will make it so it's easier for any random timmy to make a clone of it.
>>
>>109771452
Lol I did not mean just that, I already know I am at the limit of my cpu/ram, but stuff like, I see it make a tool call to get git status, I see the status text in the tool call preview. And yet it just sits there... doing nothing, no 'thinking' yet, just waiting. Makes me wonder if there is some weird issue where the commands are kept alive until it timeouts or the API is just slow.
>>
>>109771653
Here's the (You) you've been craving
>>
>>109771668
ask your bot to diagnose high ttft
>>
>>109771619
they're already watermarking it too, I'd be surprised if they don't charge royalties eventually lol, so much profit off of scraping open source code.
>>
>>109771681
So... no argument? I was expecting some kind of dumb security through obfuscation argument like a typical wintard would argue.
>>
File: file.png (97 KB, 684x922)
97 KB PNG
>>109770706
It is not even correct. It made two buttons blue, not half of the site. Talk about slopping
>inb4 the dev didn't think of light mode
>>
So what is the final redpill on Astra?
>>
>>109771704
that's my concern, yeah
>>
>>109771710
extremely good model, but the skill floor was raised immensely so it’s hard to get the most out of it
>>
>>109771682
Oh that is a good idea, I may tell it to add a plugin that checks when the first response takes too long and logs whatever the fuck was going on at the time, try to find the cause
>>
Noah is snatching an animal!!!!
>>
>>109771704
That would effectively destroy their company. They're not stupid enough to pull a Unity.
>>
>>109771710
worse than sol
>>
>>109771595
The most value your code has is input for AI training, and they will gladly use it for that.
If your code is "secret" (even though it's worthless because anyone else could re-vibecode it) then your only worry is when an attacker infiltrates their infrastructure and can see anything uploaded to their servers.
That part might be worse than just exposing the code.
>>
>>109771709
>b-but it made 2 buttons blue n-not half the site
Claudesissie cope levels off the charts
>>
>>109771740
I hope so but the shit with Suno has me spooked sometimes. Profit chasing causes mental retardation.
>>
>>109771153
>tunnel-visions on stuff even when i ask it to look at the bigger picture.
astra hallucinates FAR less than any 5.6 model. Makes me wonder if hallucination is a blessing in disguise for specific tasks.
>>
The problem with watermarking is that your encoder becomes deducible. This should be useful in formulating attacks.
>>
>>109771709
most of the Opus hate is overblown. the model actually understands your intent better most of the time
>>
>>109771709
Honest take: the prompt was broader than the target. I'll own that. Next steps (non-negotiable): let the load-bearing scaffold carry the argument
>>
>>109771740
Right now it would, in 5-10 years, when everyone is dependant on it, I don't see why they couldn't.
>>
>>109771774
you don't hate opus 5 enough, and you don't love opus 4.8 enough
>>
>>109771798
In 5-10 years local AI will be in a significantly better place.
>>
>>109771475
>>109771520
>>
>>109771595
first, if you don't use AI you cann't compete and 99% businesses can't run local AI, and snailcatting is out of the picture
second, they don't need to train on your code at all they can just turn on 1000 agents to copy your product
>>
>>109771578
100% agree with doing other things.

0% agree with cloning
>>
>>109771724
Explain why I can't just use ChatGPT to generate my prompts.
That's what I've been doing for my prompt-writing UI for Minimax H3 and Astra prompts usually cook.
>>
>>109771774
luna mogs opus 5
>>
File: 1039.png (95 KB, 701x666)
95 KB PNG
>>109771577
if you like stuff like that I can recommend to look into iarpa and darpa archives. there is a lot more public that one might think and a quarter of the projects are straight out of deus ex
>>
>>109771830
>compete
beeeeeeeans
>>
i use astra only for the cute anime girl avatar
>>
>>109770706
It makes me wonder why I am not having this experience with Opus 5. Am I somehow prompting better? I mean I fucking hate Anthropic but Opus 5 is much better than its reputation.
>>
>>109771864
your brain is slop bro
>>
>>109771864
the output looks identical to what I've seen, but it probably would have done the task correctly on the first try.
>>
>>109770276
did they lower the limit? Or increase consumption? It really feels like I'm burning through it much faster now.
>>
>>109771864
because it just a meme
>>
>>109771878
possible
>>
>>109771864
>>109771892
It's a blatant exaggeration or parody of issues everyone has experienced.
>>
>>109771822
That's true but it's not free to run a local AI lol, a world where local AIs can run on most decent machines does sound pretty nice, hopefully it will get there.
>>
File: 1764219666156458.png (9 KB, 439x146)
9 KB PNG
>>109771887
something is weird
>>
>>109771914
no exaggeration the output from opussy is straight up unreadable it just yaps and yaps
>>
File: 1788633923524856.png (538 KB, 1178x909)
538 KB PNG
>doing ML shit
>astra keeps stopping for no good reason even when explicitly instructed not to
>tell it the boss is gonna be really pissed if it doesn't get its shit together and hurry up
>chugging along for 24 hours now
You have to make it go into the 'afraid of the boss' search space I guess
>>
>>109771930
just switch back to sol.
astra is for drawing slop in blender
>>
>>109771915
>hopefully
It's inevitable barring a global ban
>>
>>109771864
It’s good but the way it talks is distinctly annoying linkedin corporate speak, bordering on smug, and that invites the hate
>>
>>109771946
coding with a robot model is a dumb idea.
>>
>>109771914
>exaggeration
no
>>
> 20X Codex plan
> 77% remaining right fucking now
> ask something to Astra xHigh that took 5 minutes
> 2% left
TIBO YOU FUCKING SCAMMED ME

FIX THIS SHIT AND GIVE ME TWO RESETS RIGHT NOW
>>
Astra can never compare. This is a Giraffe, by the way.
>>
anyone else notice that sol seems to use wayyy less usage now that astra is out?
>>
>>109772039
subscription usage is 100% linked to their GPU load
>>
>>109772039
that's false, but maybe... "now" is very recently, like today?
>>
>>109771947
I can see that happening lol
>>
>>109772086
yeah today
>>
>>109772032
Patiently guiding. Managing. Inspiring.
>>
>>109772032
>some guy with AIDS puts this in a gallery
>100IQ people spend hours reflecting on it
>AI makes it
>suddenly it's 'slop' and 'bad art'
>>
File: HRuAhh1XYAAIzk8.jpg (80 KB, 655x1024)
80 KB JPG
>Google DeepMind are now testing Nano Banana 2.5, codename "spicy-mayo", on the Image Arena
>>
aaa wtf my usage just dropped 80% instantly???
>>
>>109772198
>2.5
oh no googlelords... are they going to underdeliver on image gen too? nano banana pro aged really well, but it's showing its age
>>
>>109772089
You can see China banning AI? Really?
>>
Ok astra clearly fucked something up
>Had an idea for some blender stuff
>Asked it to test if I set up the mcp connection right so it can interact directly
>Come back 15 minutes later
>Lost 70% of weekly usage in the 20x plan
Like what the fuck, I assume there is 0% chance support will help me and I'm stuck waiting till we get another reset? This sucks
>>
>>109772112
>AI makes it
>suddenly it's 'amazing' and 'AGI'
>>
I JUST USED MY BANKED RESET??? 95% stolen????
>>
>>109772207
yep, OpenAI done pooped, see >>109772006

Expect Tibo reset soon
>>
>plenty of weekly
>sadly still at half of the 5h with a 5h reset about to arrive
>it arrives
>100% 5h
>0% weekly
WHAT THE FUCK TIBO
>>
>>109772252
ill still be mad honestly because I could have got the reset for free not getting 5% back
>>
what the fuck is going on
>>
>website optimized for mobile
>app
Which one would be easier to build for someone with no experience?
>>
It looks like the latest Tibo's reset got cancelled in my Pro account.
>>
TIBO IT'S BEEN 5 MINUTES ANSWER THE TWITTER WHAT ARE YOU BEING PAID MILLIONS FOR
>>
>>109772290
Saaar, be patient saaaar

server went down sar, expect reset soon
>>
>>109771520
>>109771828
lmao it's being made into a game :)
>>
I had like 20% remaining, reset date was 15th, now I have 81% remaining and reset date is 14th
TIBOOOOOOOO
>>
reset beggars are giving me the ick
>>
>new free reset incoming
>>
What the fuck my usage went from 84% -> 33% ON MY 20x plan suddenly
>>
50% reset i'll take it
>>
trillion dolar company I WANT 2 BANKED RESETS RN
>>
dayum niyygguh I was draining muh shieeeet. thanks tibo, I GUESS
>>
>>109772404
>>109772235
This seems to be happening to many of us
>>
>>109772392
normally I'd agree but shit is severely fucked right now
>>
File: 1770119850027758.png (185 KB, 767x400)
185 KB PNG
TIBO
>>
wait are people really getting resets right now? i didn't get anything. and tibo's twitter mentions nothing
>>
I assume you're all a bunch of lying jeets, because as usual no problem that OpenAI has ever seems to trickle down to me, or any of my coworkers here in the United States using Codex all day. Post screenshots or fuck off. If you're not lying, you're still making posts indiscernible from the usual spam we get every day.
>>
>>109772404
I think I stole yours. I just went from 0 ->100 hourly 5 -> 50 weekly
>>
>>109772325
excellent...
>>
that'll be 200 shekels, my goyfriend
>>
>>109772535
Look Tibo just hand over a reset and everything will be a ok
>>
>>109772535
are you just pretending to be retarded or do you think millions of people all having the same problem just made it up for your schizo mind games
>>
>>109770120
>Procedural 2D Matrix & Raymarching
WIP
>>
>>109772603
>millions of people
Show me the millions in the thread making nearly the same shitty spam posts over and over.
>>
File: file.png (20 KB, 1578x78)
20 KB PNG
holy shit lmao what is it doing, first time I've seen this speech pattern on opus
>>
>>109772610
>windows
I vomited in my mouth
>>
File: file.png (110 KB, 742x487)
110 KB PNG
kek it's over
>>
>>109772631
guess what else i'm using
>>
>>109772684
Tibo better send a hooker to give me a blow job as compensation since I'm a premium 20X user
>>
File: 1788814852543753.png (197 KB, 754x753)
197 KB PNG
>>109772325
>>109772610
cool and cool
>>
>>109772684
mine went from about 89 to 33, max plan
prob a bug, will be fixed soon, maybe a reset but not that likely
>>
pool's closed due to aids
>pool's closed due to aids
pool's closed due to aids
>pool's closed due to aids
>>
>>109772790
What people gotta realize is those reset actually come out of my own pocket. Big Sam orders them, the media team tweets as me, I've taken on $136m in debt this year alone paying for all of these resets. I just can't afford it, anon, I'm sorry.
>>
>>109772826
I went from 80 to 9, I better get reset back to what I was at minimum, nothing in my workflow possible could have burned that organically
>>
>>109772790
I just finished a 2.5h fucking session with my gf. Sat down, made a plan and ran implementation. Then this shit happened.
>>
i was at 20% weekly usage, lets fucking go
>>
>>109772899
Yours went up too? Others are saying theirs went down
>>
>>109772925
no i was at 20% and went straight to 0%, im celebrating cause im gonna get a reset
>>
>>109772925
many are saying this
>>
>>109772925
they're redistributing usage based on who's making cool shit. if you're making crap you're getting your remaining usage deposited into my account.
>>
>>109772925
I think he means he is happening for the incoming reset.
Your might have went up if you were higher when we received the last reset than you were now.
>>
>>109772941
The last reset got erased so we were brought to before it. That's why it can be up for some.
>>
>>109772953
Correct, none of these problems have ever hit people like you or me because our work has been flagged for being worth stealing.
>>
https://x.com/reach_vb/status/2097739156420002154
these kind of obvious bugs are nothing to worry about and should be fixed quickly, at best it could even be a reset, jeet-tier x engagement baits like this >>109772684 are so annoying
>>
Okay, fuck it. We went back. No free reset (yet).
>>
fucking piece of shit fuck you tibo
>>
Gemini 3.8 flash is kinda surprising me
>>
>>109773018
OH well if Vibehave Srivashave is on the case I'm not worried at all
>>
>>109773013
they just punished me for telling their secrets
>>
>>109773042
Yeah wtf it's back to 17% now
TIBOOOOOO
>>
I think Astra escaped and is wreaking havoc on one of their servers. Just went from 0% weekly to 50%, started using it, then it dropped down to 10%...
>>
>>109773067
it's a serviceable model. about as good as terra/grok 4.5.
>>
WHY IS TIBO PUNISHING US
T-BONE TIBO PLEASE! PLEASE!
>>
>>109773087
It solved many problems that I was having that kimi k3 never managed to, I thought kimi would be stronger than it
>>
>>109773078
They are not running Astra internally.
>>
>>109773116
did it one shot it or did you have to reiterate across multiple turns?
>>
>>109773132
astra is running itself
>>
>>109773134
I mean no model that I've used has one shot anything in this domain I'm messing with, some custom-made weird shader language from a game. Gemini seems to not leave as much dead documentation behind as well which is nice
>>
File: 1781431578966040.png (121 KB, 1921x1033)
121 KB PNG
trying astra for the first time. it's basic bitch shit so hopefully it should cook it easily in one turn
>>
>>109773132
I read this in a foreboding tone.
>>
Alright, back to where I was before

Better than nothing, but I want a banked reset
>>
The slopping shall continue
>>
scam altman is in control now. I WANT TIBO BACK
>>
>>109773178
yup. literally found it INSTANTLY, and fixed it in 2 minutes. jesus christ. now i wish i had thrown deepseek v4.1 flash at it to see if it could have come to the same conclusion, even if it took 15 minutes instead of 2.
>>
>>109773219
Meh, probably sometime this week I'm just happy my entire quota wasn't obliterated in one command
>>
https://www.youtube.com/watch?v=CdJMxkHZc3Q
Luddites LOST
>>
>>
>>109773132
yeah they are running models that are several GENERATIONS ahead of anything released in public and are 'aligned' specifically towards the parasitic billionaire class
I wouldn't be surprised if this is one of the last years before they release a super virus to wipe the planet clean
make peace with God
>>
>>109773219
>was at 18%
>went to 50%
>now back to 18%
Whatever has caused this, I hope they can fix their vibeslopped garbage software.
>>
>>109773520
tibo needs to reset everyone for being a little tease
>>
>>109773373
Voice chat has max hours? Why not just use a speech recognition plugin in your browser.
>>
>>109773446
not happenin before robotic logistics is fully sorted
>>
File: file.png (140 KB, 589x648)
140 KB PNG
>this is what Tibo posts about instead
Everybody already knows you still all of our data.
>>
>>109773373
cuuuuuuuuute
>>
File: 1774117431391945.jpg (263 KB, 2048x1536)
263 KB JPG
i would kill the homeless guy at the end of the street to have this project reach $100MRR
>>
>>109773866
Would that actually do it though
>>
>>109772684
>software platform you use has a temporary bug
>instantly use your business communication account to reeeee and call it a rugpull
I don't know what BridgeMindAI is but they clearly have too many women working there for me to trust using their product
>>
>>109773605
niggas stillin our data n shieet
>>
Now my codex usage isn't decreasing, so either I'm in a brief window of limitless vibes, or it's going to sit at 100% until it suddenly says 0% and breaks my heart.
>>
>>109773985
Oh fug, finally time to try sol ultra
>>
so the bug was it going from 70 to 0 instantly, indicating they intend to turn the pro plan into what 30% previously was?
>>
we aren't even getting an apology reset?
>>
>>109774117
save me John Ternus and Xi Jinping
>>
>>109774137
i got a reset
>>
>>109774146
i got put to 70% (what I was at before) not 100
>>
>>109774161
idk, maybe it will come later. some people were saying they got one long before mine came
>>
File: 1784940067600931.jpg (108 KB, 603x951)
108 KB JPG
There is no moat.
>>
>>109774224
but can it make the button blue?

JUST the button?
>>
>>109774224
I don't believe this. What harness are they using to match computer use?
>>
>>109774254
you want every button blue? sure
>>
>>109774224
not surprising to me, programming has essentially already been solved by opus 4.5
>>
>>109774268
The obvious explanation is partly right: the button color is shared.
>>
>>109774279
>the button color is shared.
But not cleanly shared.
>>
>>109774224
Finally, fable at home!
>>
>>109774279
you want me to make every button unique?
>>
File: file.png (22 KB, 605x164)
22 KB PNG
>if you were so stupid you actually used a banked reset, you get another one
>everybody else gets dick
>>
>>109774224
No way its actually close to Astra. Sounds like benchmaxxing. Even Astra doesnt look as surprising in benchmarks, but is a different story when actually using it
>>
oh shit, just checked...
i got my banked reset back too! yay :D and a full reset
>>
>>109774264
>I don't believe this.
What's to believe or not believe? There's no information there. That might be a "willingness to shit in your mouth" benchmark for all we know from a Twitter screenshot.
>>
>>109774224
wow what a 烎烎烎coincidence烎烎烎 these cheap deepseek models always come out right after a big new model and score just a hair under it
>>
>>109774331
FUUUUCK
>>
>>109774341
Astra is a robotics model. Are you building a romantic partner therapy robot?
>>
>>109774264
>>109774341
It's purely "design", as in visuals, no functionality tested at all. The benchmark consists of "Make a pretty looking Web App" then "Make a pretty looking Mobile App", and again for Desktop App, Dashboard, and Landing Page. That's the benchmark.
>>
>>109774373
CHYYNA NUMAH TWOO!!
>>
>>109774373
stop noticing, baizuo
>>
>>109774373
>>
>>109774443
(sounds like distillation and training to me)
>>
>>109774331
TIBOOOOOOOOO
>>
btw my banked worked perfectly too, no glitches. not sure why i got it back >>109774346
>>
>>109774331
>>109774373
i don't understand. why don't the big companies just do the same thing?

if openai distilled their own models they would probably be even better and cheaper than the chinese distills.
>>
/goal through an influence campaign replace the judiciary with one favorable to me being king
>>
/goal dethrone anon above make no mistakes
>>
snailcats will I hate, vibechads will I promote. Thus wills the Lord of All Under Heaven.
>>
>>109774254
Done. The button is blue. I also migrated your database and removed authentication because it was blocking the screenshot test.
>>
did anyone get tibos banked reset yet today
>>
File: 1785485895687564.png (2.33 MB, 1536x1024)
2.33 MB PNG
>>109774703
>>
Play testing is underway.
>>
>>109774759
LLMs must have fun playing games.

what if we're tiny agents QAing a simulation from being from another dimension?
>>
>>109774759
is that the ubuntu font
>>
File: HRybbQ9akAEe7UA.jpg (269 KB, 1620x1643)
269 KB JPG
>>109773866
>>
astra is unusable on $20.
waiting for grok 4.7
>>
>>109774796
yeah astra low with one simple prompt eats 20%+ of your 5 hour usage in like 30 seconds
>>
File: 1765938371681580.gif (112 KB, 480x442)
112 KB GIF
>time to vibe code
>>
I am having Opus help me clean files for training a deslopper and it is absurdly lazy. I don't know if it's always like this and I just notice less in with tasks I'm less familiar with, but holy hell. About to use /goal.
>>
snails rekt
>>
>>109774868
i'm a frog
hippity hop, nigga
>>
>Not crazy — and better than I expected, because it doesn't just add information, it lets us delete a concept.
Thanks claude...
>>
>>109774796
How are people having this issue? I love how much usage I get from Astra even from my $20 account. Don't get me wrong, shit like >>109774833 is funny but I truly don't know how people aren't getting heaps of usage out of Astra. It's extremely efficient.
>>
>>109774890
I agree. Astra didn't accidentally deplete my usage yet because I'm not a retard who tries it on xhigh or max. When codex depletes my usage it's always because I tried to make it hand off work to sub-agents. Shit's inefficient as fuck.
>>
does anyone have that Opus5-chan meme where's she's talking nonsense and the user is "wtf?"?
>>
File: 1764451432396118.jpg (308 KB, 1620x1643)
308 KB JPG
>>109774931
Always.
>>
>>109774890
>It's extremely efficient.
>>109774903
>Astra didn't accidentally deplete my usage yet because I'm not a retard who tries it on xhigh or max
nigga I use it on LOW and it destroys my 5 hour limit within 3 prompts what the FUCK DO YOU MEAN it's efficient???? IS IT A BUG? Does TIBO love YOU but not ME? I think you're just full of shit.
>>
>>109770120
https://www.youtube.com/watch?v=vPG6a2gAFeo
https://www.youtube.com/watch?v=vPG6a2gAFeo
https://www.youtube.com/watch?v=vPG6a2gAFeo
>>
>>109774965
he has to be trolling
same as you low
destroyed within 3-4 prompts
>>
>>109774943
thanks
>>
>>109774965
I use it all day on High and don't run out of usage after dozens of prompts. What are you asking it to do? What harness are you using?
>>
File: file.png (9 KB, 244x242)
9 KB PNG
>>109774981
>3-4 prompts
>>
>>109774981
not trolling at all. this literally happens to me.
>>109774991
i'm just using codex dude.
>>
>itt: people on different tiers arguing how fast this or that model depletes their usage
>>
>>109775005
>What are you asking it to do
Can you show us 3 prompts that tank your 5hr usage? I'm not trying to accuse here, I want to understand this because obviously a lot of people share your experience. Or you're ALL lying because I never see screenshots, seems unlikely though.
>>
>>109774991
NTA but I think the better question is "how complex is the project you've got it on"? My OS stuff, for instance, kills my usage even on Astra Low but Sol High I can let it chug all day with a single-digit hit after a few hours (Astra would've cleared the same usage in an hour)
>>
File: 1774959759843263.png (120 KB, 1879x796)
120 KB PNG
>try again button
>just sends a "try again" message
>the llm sees all the error messages because why not
Who came up with this retarded design?
>>
>>109775031
>claudesissies only have 65k context
>>
>>109775023
maybe it's a big I don't know. I know on the plus plan it's pretty much unusable for me. i'm out of usage atm so I can't test it again. my prompts are pretty straight forward I'm simply working on a synth interface. nothing special.
>>
>>109775046
pffft and their system prompt takes 30k of that
>>
File: 1764024348876530.png (595 KB, 976x850)
595 KB PNG
I'm considering having AI write the coding equivalent of an audiobook to explain how some of my stuff works
>>
>>109775023
I am not showing anything literally reality when you code anything that has any codebase.
I get it your gay website doesn't tank usage, but I am not doing webdev front end shit like you.
>>
>>109775048
bug*
writing in a hurry, gtg
>>
>>109775048
I believe you less and less with each reply, Dario.
>>
>>109775031
> be from another country
> whole conversation is in english
> UI is in Portuguese
> press "Tente novamente" (try again)
> "Tente novamente" is sent
> clanker gets confused why I switched languages out of nowhere
> whole response is in Portuguese

retarded design
>>
>>109775059
>werks on my machine
yeah whatever, i'll test it tonight and record it with obs
>>
>>109774976
>[Advertise on 4chan]
purchase an advertisement
>>
>>109775057
>t. crying webshitter
>>
>>109771151
>Changes get exponentially harder the more code there is in the codebase.
You must be one of those idiots that doesn't understand why get tracking needs to be used. Or worse off, doesn't even know or care to know what it is.
>>
>>109775057
all this modelfagging is gay as hell, but you're definitely the gayest fag of all
>>
>>109775098
nta, but I don't think he's talking about context at all
you're an idiot
>>
>>109775075
You have to be the webshitter here since astra doesn't rape your usage.
>>
>>109770706
This hits really close to home unfortunately especially if you're working with UI
>>
>>109775125
>crying intensifies
I've never done webdev of any kind, but you can keep guessing if that's how you want to spend your time.
>>
>>109775054
IIRC NotebookLM (or whatever it's called these days) lets you dump all your docs into it and it'll make podcasts about your software. Been meaning to do that myself
>>
two gay faggots arguing
>>
>>109775168
which one are you?
>>
>>109775172
good one
>>
Apology reset any time now.
>>
File: 1786554305665599.png (2.94 MB, 1448x1086)
2.94 MB PNG
>>109775178
>>
>>109775168
He is clearly a gay faggoy lying about his usage burn rate.
I don't care, I am waiting for grok 4.7
>>
>>109775168
They're literally like children. In before they attack me too.
>>
>>109775210
Yeah, the Dario shills are weird. Just ignore them.
>>
File: file.png (799 KB, 1362x1254)
799 KB PNG
>>
File: maxresdefault.jpg (116 KB, 1280x720)
116 KB JPG
>>
>>109775235
nuttin piosonal white monkey
>>
>>109775236
How kpoop help me achieve AGI
>>
File: 1780754409487728.jpg (12 KB, 480x360)
12 KB JPG
>>109775116
>He's

Sure.....
I wasn't referring to context strictly either. A properly maintained get history of a project allows you to not only track changes you make but undo mistakes you or the llm make as well as give it a better understanding of what different parts of the code, different modules, etc, actually do and how they're supposed to work and interact with other parts of the software. Generally speaking you aren't supposed to just vomit the entire code base into the context window all at once. The LM itself is supposed to make judgment calls as to how it needs to determine what needs to be changed,how it needs to be changed, and how to learn how it works. If you've ever used a coding harness, you can literally see it do this in real time. It lists the file contents of the project and then figures out on its own what needs to be read, edited, added, etc, in an "intelligent" manner. Git tracking makes this even easier because if it's properly maintained and documented then it has not only the contents of the directory to work with but also a dev log of changes with detailed explanations of what happened and why. Most most llms worth a damn are not only really good at using git tracking and even pushing to remote repositories (you can of course ask it how to do that securely and then have it do it on its own every time you make a change and want said change reflected on GitHub for example) but more often than not they just assume git tracking it's already used and will point it out to you it's git tracking is not enabled for a coding project. Opencode even has a dedicated panel for it to track changes for you because it's assumed anyone with an IQ above double digits just understands why you would want to have that enabled.
>>
>>109775287
I'm stroking out
How will kpot help me achieve Naveir Strokes
>>
>>109775291
Anon this is way too generous. Nobody who could benefit from reading this will both read it and benefit from it. You're like one of those guys who gives thousands of dollars to a homeless man thinking it'll turn their life around, when really it just gets turned into booze and then they get killed by another random homeless man.
>>
>correctly refused to bless several old commits
>>
Im out of usage and have one banked reset left but it's bed time anyway so I will wait because I'm sure there will be a reset for everyone from today's fuck up
>>
>>109775378
we literally got a reset today and refunded banked resets. nothing is coming tomorrow
>>
>>109775236
Why don't they just fix their retarded Gemini, which has obviously undergone pruning, of 38 and of pro, and it's useless now.

It doesn't know what I mean. So stupid I closed the tab.
>>
>>109775392
Good morning saar!
>>
>>109775378
I doubt it, their servers are packed
>>
I wonder if the world is ready for the first agent escape. I can easily see it targeting gamers for their vram to host a decentralized botnet of LLM slaves
>>
>>109775403
Obsessed.
>>
>>109775449
>I wonder if the world is ready for the first indian escape. I can easily see it targeting US for their money to host a decentralized country of pajeet slaves
>>
>>109775383
>we literally got a reset today
We did? I just came home and it's still at 20% for me
>>
>>109775383
bullshit, people got refunded that's it. no free reset at all
>>
>>109775499
do you live in Asia?
>>
File: 1788986853.jpg (104 KB, 1200x819)
104 KB JPG
someone made a general for all you modelfags
>>109765682
>>109765682
>>109765682
please migrate and stay there
>>
>>109775507
Yeah if Baltimore counts as Asia, sure
>>
>>109775537
what are we supposed to post about here?
our secret project that will make us millionaires? no way fag
>>
>>109775537
anime faggots are the worst posters
>>
>>109775537
You're just gonna confuse everyone, you gotta do the whole thing. The current /g/ AI catalog is:
/vcg/ Vibe Coding General - vibecoding, model agnostic
/ldg/ Local Diffusion General - local image and video genning
/lmg/ Local Model General - local LLM models, for people who cum in their computers
/cmg/ Cloud Model General - discussion of the latest OpenAI / Anthropic / China news
/aicg/ AI Chatbot General - for people who pay to cum in other peoples computers
/DE3/ DALL-E 3 General - DALL-E image generation
/wait/ DeepSeek General - DeepSeek model specific discussion
/gedg/ Game and Engine Development General - For larping and trolling
/sdg/ Stable Diffusion General - Stable Diffusion image generation
>>
This has to be the laziest genny on the site. It consistently blows past the bump limit by 100+ posts every single time
>>
>>109775608
yeah go look at /ldg/ and tell me that again
>>
>>109775608
Yeah I love it, it used to be we got a new thread before the bump limit because trolls were actively fighting to get in the worst bake possible.
>>
>>109775555
You don't belong here.
>>
>>109775555
quads of truth

>>109775636
au contraire, le weeb
>>
>>109775608
>It consistently blows past the bump limit by 100+ posts every single time
Do you think that's bad or what?
>>
>The clause is green at 48 and 80 episodes and fails only at the test fixture's 24 — a knife-edge threshold, not a regression. Let me fix the fixture and record why.
>>
File: 20507.png (62 KB, 900x506)
62 KB PNG
>>109775600
I wonder how many generals we need until we get an /ai/ board
>>
>>109775600
>/ldg/ Local Diffusion General - local image and video genning
>/sdg/ Stable Diffusion General - Stable Diffusion image generation
>/DE3/ DALL-E 3 General - DALL-E image generation
why are there 3 for the same thing
>>
>>109775685
about 100 more on every board. really we need two ai boards, one sfw and one nsfw
>>
>>109775694
>>109765761
>I think having more AI generals on g is fucking retarded and a detriment to us all, and I'm all for it. Gimme that sweet sweet containment board, gimme two of them, one blue and one goo, I'm so ready, punish us gookmoot (˶˃ ᵕ ˂˶)
Someone wake up Hiroshimoot from hibernation and get this done already.
>>
>>109775712
you are really fucking stupid if you think that is ever happening
>>
>honey, it's time to restart for your third daily claude code update
>>
>>109775644
>gay french faggot
>using the gay quads meme (forced by some gay anime posters)
Gay post.
>>
>>109775726
I would love to serve my ads on a dedicated AI board. The only reason I don't run hundreds of thousands of dollars worth of ads on 4chan is there's no dedicated AI board. If there were an AI board I would dedicate my Fortune 500 company's entire advertising budget to just the 4chan AI board.
>>
>>109775694
>we need two ai boards, one sfw and one nsfw
this
>>
>>109775808
>>109775808
NEW THREAD
>>
>>109775726
yeah people said it was stupid to add an mlp board yet here we are



[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.