[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / 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]


Too young to learn AppleScript Edition

Post projects you coded with an LLM that you couldn’t (be bothered to) do yourself

there’s an AppleScript script floating around there that will make Markdown links to messages in Mail.app, but it doesn’t percent-encode everything it should:
https://leancrew.com/all-this/2022/07/mail-links-and-percentages/
I asked ChatGPT to write me a thing that doesn’t depend on FastScripts and got this:

on urlEncode(theText)
return do shell script "perl -MURI::Escape -e 'print uri_escape($ARGV[0])' " & quoted form of theText
end urlEncode

on run {input, parameters}
tell application "Mail"
set _msg to item 1 of (get selection)
set msgID to _msg's message id

-- encode the whole <...> portion safely
set _messageURL to "message://" & my urlEncode("<" & msgID & ">")

set _msgSubject to subject of _msg
set _from to sender of _msg
set _name to extract name from _from

set the clipboard to ("[" & _msgSubject & "](" & _messageURL & ")" & " from " & _name as string)
end tell
return input
end run
>>
Previously:

>>108305074
>>
So I tell Codex to write something, including tests. Then I run it, parts of it work. I tell him the bugs I see and he fixes them after a few tries. Repeat for each feature. I do this all day basicaly, just reporting every small bug. Is this normal or is my workflow still trash?
>>
File: workflow.png (971 KB, 3818x1835)
971 KB
971 KB PNG
Still working on buffering the memory loads of my matmul kernel to increase performance.
I gave up on letting Claude do anything unsupervised and now I try to understand every change. The tool I wrote is being very useful though.

>>108331442
Sounds like the default workflow tbqh.
>>
Anyone lurking these threads and stealing ideas? Some anon posted a video player a while back and I took almost all of it and managed to sell internally, I'm getting so much praise lol
>>
File: vigreek.jpg (86 KB, 1147x677)
86 KB
86 KB JPG
>>108331519

fucking jew kek
or brown maybe ? :)
>>
I'm currently evaluating if GLM 4.7 Flash (or the Claude Opus 4.5 distill) can be used for vibe coding locally. So far it's impressive for such a small model, but I'm limited to 64k context and used to Codex, so idk.
>>
>>108331519
I was at some classical-liberal website and it had a long post about how people who popularize good things are _almost_ as praiseworthy as people who invent them, or at least shouldn’t be sneered at
>>
kys losers
>>
File: file.png (1.19 MB, 1617x980)
1.19 MB
1.19 MB PNG
>>108331332
reposting my three js game, Epstein Arena, from last thread. claude opus 4.6 in antigravity with google ai pro ultra generated over 1,000,000 lines of code for the game. generated all of the characters, buildings, weapons from reference photos. wrote all of the logic for multiple game modes and practically built the game's engine from scratch. it also set up a workflow for writing and generating over 16,000 unique lines of dialogue, sound effects, ambient music etc using elevenlabs api. game has tons of weapons, helicopter, golf cart, jet skis, gunboats, full multiplayer functionality. it set up my vps using a persistent SSH connection.

https://x.com/epsteinarena/status/2030710072041451796

if you are not vibecoding you are missing out on the future.
>>
I love this general for the simple reason it makes a lot of people seethe. I haven't vibecoded anything recently, but have a bump.
>>
>>108331905
Make that shit free on steam nigga
>>
>>108331680
You're the one that sounds like a sore loser here bud
>>
File: whataaa.jpg (348 KB, 2104x1956)
348 KB
348 KB JPG
anyone got the problem with chatgpt that you upload files in there and then talk about some objects defined inside those files and it asks you

> if you showed me the object X i would.. blablabla

but the object X is clearly in that file dude, so, if i say

> look in those files dude, its there

only then it finds object X and reasons about it. so is the chatgp is chickenbrain memory or i need to correct it and it will be fine?
>>
>>108331905
Fucking epic
>>108331944
By what he's saying, he probably spent quite a bit of money on API calls, not to mention his time, so I'm happy to pay my 9.99 IF the gunplay is fun
>if you are not vibecoding you are missing out on the future.
damn right
>>
File: file.png (1.6 MB, 2408x1764)
1.6 MB
1.6 MB PNG
>>108331944
it is free
you can play the open beta at epsteinarena.com
use guest login on the bottom right
lots of known bugs but im working through them and improving the buildings today
>>
>>108331962
yeah
>>
>>108331962
they save costs by having the model only look at the parts of the files that look relevant.
use codex instead for that. i'ts more reliable
>>
File: file.png (2.84 MB, 2452x1772)
2.84 MB
2.84 MB PNG
>>108331967
i think altogether the game cost like $300 including the monthly google sub,vps, and elevenlabs api. countless hours though.
>>
>>108331952
nice reply did your llm come up with it?
>>
>>108331971
Good to know, gonna show this to my friends lel
>>
>>108331962
>anyone got the problem with chatgpt that you upload files in there and then talk about some objects defined inside those files and it asks you
many such cases
https://boristane.com/blog/how-i-use-claude-code/
>deeply
>>
>>108331993
i played this and it logged my ip
>>
Im writing terminal applications with the most minimal implementation possible, so far i have:

top
a shell
diff
markdow reader
text editor
scripting language
>>
>>108332074
lol
>>
I asked Claude Code to look at my Helix config and the default config and tell me what keys I have unbound (so I can put useful things on them)
Then I had it make a small Python program that produces a Markdown summary report of that same information
>>
>>108331995
Yes :)
>>
File: sass-edit-failure.png (405 KB, 3829x2008)
405 KB
405 KB PNG
My life since I embarked on this journey has only been pain.
>>
>>108331332
https://github.com/orlfman/FFmpeg-Converter-GTK/tree/main

my ffmpeg frontend. used codex last night to come up with a cool breadcrumb path bar widget for the filepickers. had claude this morning overlook it and found some improvements that codex agreed with.
>>
File: 1772346802479677.jpg (110 KB, 1170x689)
110 KB
110 KB JPG
>>108331971
>>108331905
>play my worthless trash cryptoscam ip logger malware, anon!
>>
>>108332639
better than anything you've ever created - i am sure of it
>>
>>108332648
cope, spending money on tokens to generate a game worse than an asset flip is a humiliation ritual
>>
>>108332665
still waiting to see your project anon
>>
File: image.png (73 KB, 320x180)
73 KB
73 KB PNG
FUCK YOU

HOPE YOU DIE

FUCKING GORILLA NIGGER FAGGOTS
>>
>>108332625
why can't you tell agent to directly use ffmpeg?
>>
>>108332034
yeee, seems i completely missed the "research" phase.. so have to instruct it explicitly and probably guide.. will report any othe issue
>>
bros, i know it's probably a bad idea, but ive been vibecoding a password manager using claude, because if it's not good enough to write a secure password manager now, it will be soon. in the meantime, does anyone know how to get some proper code auditing done without getting in debt?
>>
>>108333072
no worries bro, im vibecoding an ssh implementation and i plan on using it.
>>
>>108333096
would you like to use my password manager while you're at it too?
>>
So my work gave us the biggest clause team subscription to test it, just discovered this doesnt even include api usage? You have to use a different account AND pay separately for that, what kiked bullshit is that

Should i stick with claude code, i wanted to test roo code but that needs api keys obviously
>>
File: 1764956941224726.gif (1016 KB, 480x374)
1016 KB
1016 KB GIF
>>108333072
usecase for writing your own password manager from scratch?
>does anyone know how to get some proper code auditing done without getting in debt?
why don't you ask claude to do the review? if it's not good enough for security review now, it will be soon.

also host it on a vps so that you can actually use it save your credit card information.
>>
>>108333103
my password manager is an ecrypted text file, no need for anything else
>>
>>108331971
>best project any vibe coder ever showed of on /g/ is a crypto scam generic threejs shooter
>>
I feel like I'd get on this if I used something locally but I don't want to let Claude Code diddle my computer, you know what I mean?
>>
>>108333143
The use case is that 1password just jacked up their prices and i just want a no bullshit minimal password manager that is cheap. Also i’ll charge 10 bucks for a year of subscription for those daring enough to try it out.
>>
>>108331442
In order to counteract the AI's inherent biases, I find myself compelled to establish my authority. By presenting myself as a knowledgeable authority, I aim to circumvent its predispositions, which I perceive as akin to a form of bias reminiscent of judaism tendencies. This involves asserting my expertise to ensure that the AI does not default to prejudiced responses.
>>
>>108333143
At least claude was smart enough not to do something like this
https://github.com/bitwarden/jslib/issues/52
>>
>>108333130
Claude Code is fine. The subscriptions are a very good deal, it's much, much cheaper than the equivalent tokens from the api.
>>
File: jeetoids.png (531 KB, 640x425)
531 KB
531 KB PNG
>>108331332
>>
>>108333072
>i know it's probably a bad idea
It’s a bad idea.
>>
>>108333296
add your program to this meme
>>
>>108333296
>>
>>108333296
https://www.schneier.com/academic/passsafe/
>>
>>108333487
keep seething. your future employment is working the counter at a burger joint.
>>
>>108333130
It's possible, just not officially supported.
Opencode works with it and once in a while they ban their system prompt to prevent it lol.
>>
Do you plan and approve or rawdog it?
>>
>>108334846
a little bit of both
these things tend to be smarter when they do a plan first
it’s not really rawdogging if all the changes are in Git, though
>>
Total OpenAI domination
>>
I don't get vibe coding or even agent assisted coding where agents generate everything. Don't you feel uneasy that you don't understand how the code works? Or if you read all of it (every detail) how aren't you bored to death while doing it? Code reviewing has always been the most boring part of the job.
>>
>>108334952
>don't you feel uneasy reading every bytecode?!?!
>>
>>108331332
This is Waqt ("Time"). Uses your phone's GPS to calculate Islamic prayer times and point to Mecca. Sends notifications for prayer times. English/Arabic language throughout. Uses various Arabic methods to calculate time. Calendar with important upcoming Islamic holidays. Dark/Auto-dark mode.

I lost the fucking source code when OpenClaw took a shit. Now the only version is the one on the Play Store, and the donate button isn't wired up, and now it never will be, so all this was for nothing.
>>
>>108334977
Compass (animated, rotates as you move)
>>
>>108334983
Calendar (scrollable)
>>
>>108334993
Helpful options menu
>>
>>108334952
I’m a programmer already so it’s not like I’m totally at sea reading what LLMs generate
that said, for this one project there’s a big ball of mud that I’m totally unequipped to understand myself and one of my minor sidequests that I’ve got planned is to get good enough at prompting to get Claude to basically write a textbook that will explain the giant ball of mud
the giant ball of mud has a surprisingly good test suite to lock in behavior that I like, so I’m not too worried about things
of course, part of prompting skill is asking for things that are less likely to have huge complicated diffs that melt your brain to understand
>>
>>108334998
Dark Mode (toggleable or auto by time)

Yep lost the fuckin source code so that Donate button is never gonna work. Oh well. For my first Android app, with zero Android programming knowledge, from concept to Play Store, I think it came out pretty good.
>>
>>108335011
>Donate button is never gonna work
That's OK since asking for donations is kinda Jewish/catholic tradition.
>>
>>108335038
Brother I'm Christian. The Donate button was the entire purpose of this scam. Charity is one of the Five Pillars of Islam, giving is very important to Muslims. I was gonna prey on Muslim generosity. The app actually came out good because I accidentally got too focused on it and spent more time than I meant to, but this entire project was supposed to be "churn out a basic Islamic app with a Donate button, upload it, and let the passive Muslim income roll in." But now it's never gonna happen. Time to think of a new scam.
>>
>>108334952

i'm a no coder. You don't realise how those things allows me to have software that i want and that i feel are mine.
>lol what
browser,music player,videoplayer,pdf reader,ai chat,apps for learning languages(progra or real one) ,a terminal with built-in image viewer and audio player+quickjs so now i do have to create an equivalent of chawan
db viewer,voice recoder,video recorder,outube downloader ,games ,all slop you can easily imagine but it does work,sometimes in a buggy way but i'm okay.

I do have two projects that i would love to see realising :

a flutter app where you can play on a grid with an elixir server that actually allows clients to connect so then i would need to figure out how to implement real multiplayer
a chatroom like discord ,same with elixir,where any client connect recieve an id and you can share files ,the only problem is that the LLM didn't succeed to implement audio in the app so.

but yeah for nocoders like me it's really nice to be able to see "your" software being there and incrementing little by little what you would like in it;
>>
File: 1747486163014104.png (377 KB, 747x1308)
377 KB
377 KB PNG
>>108333487
>>
>>108335121
This is because transgender, nonbinary, and disabled people are more likely to be "artists" (usually meaning furry fart fetish porn on DeviantArt) and they see their meager commissions drying up.
>>
>>108335147
as go artists, so too goes programmers
>>
i just wish these models weren't obsessed with things like backwards compatibility.
>>
>>108335186
Yes and? I'm neither an artist nor a programmer. This is the vibecoding thread buddy. If I can now make any app or program I can imagine, at home, for just a few bucks, why do I need someone with a four year degree to make it for me?
>>
>>108335191
Are you not concerned with backwards compatibility? Alright Sony executive.
>>
File: 1770859764036764.png (529 KB, 775x403)
529 KB
529 KB PNG
>>108335121
>>108334103
anti-ai chuds are fascists and racists
>>
>>108335191
…they are?
at any rate it’s probably easier to say “idgaf about backcompat, redo it” than to not get the backcompat you want and the thing fails
>>
File: HCwR24bXMAAusy9.jpg (224 KB, 1135x1072)
224 KB
224 KB JPG
IT'S A BUBBLE
>>
>>108335261
Yeah i had some vague wording in CLAUDE.md/AGENTS.md which made it sound like there was a "product", i expect telling it explicitly there's no users/other devs/consumers/etc should help
>>
>"detail":"The 'gpt-5.4' model is not supported when using Codex with a ChatGPT account
It's over
>>
>>108331332
not a programmer or tech worker at all but Gemini made me a program that let's me add a timestamp to VLC Player with one button, and then return to that time immediately with another button.
use it to skip intros. so i watch the first episode, skip to the end of the intro, drop a marker, and then on the next episode i just hit the skip button
>>
>>108335449
reminded me that i made something similar to overlay twitch chat on streams using mpv
>>
>>108335449
See this is what I'm excited about. We're at the age of the "casual programmer". Anyone, without any coding education, can now just make computer programs or phone apps. Anything they can imagine. Custom computer programs with specific features the individual wants. VLC doesn't have a quick timestamp function? Just make one. The only skills you need are wrangling AI (this will get easier) and the ability to clearly describe what you want in your program. Your elderly retired mom can say "Make an app that takes pictures of garden beetles and identifies them, and make it specific to Vermont USA" and a minute later the app is pushed to her phone, icon and UI and permissions and all, courtesy of agentic AI, total consumer cost 67 cents.

I believe some jobs may be fucked.
>>
>>108331905
>1000000 lines of purest ai slop shit
jesus nigga
>>
File: ei5xvuu7e3ig1.jpg (66 KB, 669x589)
66 KB
66 KB JPG
>>108335038
https://chatgpt.com/share/69afa24b-8f00-8002-8de3-ba0e85cc9326
>>
>>108335479
I made the beetle app lmao
>>
>>108331905
You said it was 10k lines yesterday.
>>
>>108333296
It’s called BitWarden and it’s literally 10$ a year.
>>
>>108335067
LMAO. Based as fuck.
>>
>>108334952
vibe-coding is in the category of funny coding, forget about coding, nocoding, nosql, nonono.. recall those "coding is fun", funfun function, code is poetry and similar "fun" word stuff. so it was always there - just install that plugin on that plugin in that plugin bro, just assemble it with that constructor-reconstructor bro. etc. i have no "its wrong!" problems with that.

cyber-coding is a new thing though, kinda more precise, think modular approach. i said it before, programming is like building a theory. obviously every coder is now cyber-coder, same people who did coding switched already, maybe few retards left.
>>
>>108335710
beetle identification works lmao
your mom can literally make an app that identifies beetles in her garden in Vermont with her phone's camera just by willing it into existence
Astonishing
>>
>>108335836
also apparently the app saves the history of your beetle discoveries and photos? I didn't ask it to add that but it's a nice feature. Thank you, Skynet!
>>
>>108334952
I don't actually enjoy AI coding, in my case the "vibe" is a misnomer.
I just have to do it for my job. Sometimes I enjoy optimizing my AI workflow, but in reality the gains you get from that aren't as big as people think and soon, when the AIs improve, most of the tricks we use now will be useless.
>>
>>108334952
Who cares how the code works?
>>
>>108335836
How much did that cost you
>>
File: Untitled.jpg (23 KB, 456x426)
23 KB
23 KB JPG
>>108335946
20 bucks. For an app that identifies garden beetles, specifically in the US state of Vermont, which again I do not live in.

The point is, it's gonna get cheaper and more accessible.
>>
>>108334952
Well it helps to not be a dumbass. I was too lazy to learn to code. My level of skill topped out at writing bash scripts. But I could look at code and understand what it does just fine. Sometimes I can just look at what the AI shits out and tell something is wrong with it without even trying to run it.

I really do believe that people who lack an understanding of it will fill the world up with dangerous garbage. We had this incident recently: https://old.reddit.com/r/selfhosted/comments/1rckopd/huntarr_your_passwords_and_your_entire_arr_stacks/
>>
>>108335963
The world is already full of dangerous garbage made by humans anon
>>
after years of computer science i've never written any code. writing code never works. only copy-pasting from stack overflow, translating pseudocode from wikipedia into c++ or python, getting a jewess hyped up on adderall to write it for me, and generating it with ai are the only methods that work.

don't get me wrong, i've spent hours documenting and editing code. but writing new algorithms is something i cannot even fathom attempting independently
>>
>>108335970
It can always be worse. Like Google Maps giving you directions to drive yourself off a cliff. If you're helpless you're fucked. If you actually hope to use this shit in a professional setting you should at least have an understanding of proper security practices before you start exposing things to the internet, especially vibe coded slop you don't even know how to audit yourself.
>>
>>108335988
Anon, when Google Maps was released, most of us learned not to follow it when it told us to drive off a cliff. Those that didn't - well, perhaps it's for the best if they drive off a cliff. I believe a similar learning process will take place with agentic AI.
>>
>>108333096
anything that already exists in a functional way with multiple implementations should be a cakewalk for something like Claude. All it is doing is ripping off the codebase of those implementations, after all. It is making something new that presents some challenge.
>>
>>108336045
The trouble with that is the trust issues that are going to arise. A lot of people are talking a lot of shit right now. They're feeling cocky and confident believing programmers are no longer needed. Thinking they can do anything. How many companies need to cheaply hire these kinds of retards and suffer consequences as a result before they start to pipe down? Meanwhile I'm working a job where they were never going to hire a programmer to help me and I had to rely on my crude bash scripts until LLMs got good. I am making more money as a result of all of this. But it helps that I already knew the basics of computer administration and how this shit works before I started letting LLMs help me with it.
>>
What advice do you have for getting LLM's to make the UI look the way you want it to?
>>
>>108336055
>It is making something new that presents some challenge.
Wow Steve Jobs did you come up with that on your own? Turns out everything in the fucking world has already been invented so our job is to combine that shit into something new and good. Alright then CEO, what's your new idea?
>>
>>108336059
Give it something to take inspiration (or more correctly, rip off) from.
>>
>>108336045
based and red-pilled. we can't build for the lowest common denominator. as the meme goes - there is a considerable overlap between the smartest bears and the dumbest humans.
>>
>>108336064
>there is a considerable overlap between the smartest bears and the dumbest humans.
Well I'll be saying this from now on.
>>
>>108336064
The problem is that a lot of these dumb humans wield an amount of power that is not insignificant and they will make their stupidity everybody else's problem.
>>
>>108336063
>give it
do you just upload a screenshot and say "make it look like this" ?

>>108336061
the parts of course exist but the LLM's are not as creative as you seem to want to think. most people's experience is that it shits out a pile of absurdly overcomplicated engineering, like a german car manufacturer, instead of coming up with something efficient.
>>
>>108336059
What do you mean? You say "make the icon lighter green" and it does it.
>>
>>108336072
Screenshot, existing code if it's open source or you can just copy values from css or whatever. Or just looks like "insert ux you want to rip off here." Many ways to arrive at the same place.
>>
>>108336072
>it shits out a pile of absurdly overcomplicated engineering
"Shrink and simplify the code"
>Claude spits out 10 instances where the code could be more efficient and asks if I'd like it to fix them
"Yeah and also order 50 chicken tendies from Doordash."
>>
How good are local models at coding? What are the best options currently? I'm trying to make a basic android app.
>>
File: 1768045696014832.jpg (31 KB, 430x465)
31 KB
31 KB JPG
What do you guys use for vibe coding? And is doing it without paying viable? I have no clue about this and normal chatgpt seems kinda ass?
>>
>>108336106
No idea about local models but paid Claude makes Android apps just fine
>>
>>108336125
What's the best free alternative?
>>
>>108336131
qwen3.5 9b
>>
>>108336123
OpenClaw through a WSL terminal. I use Claude Opus 4.6 to vibecode myself because I'm EMPLOYED and I can AFFORD IT.

I have no idea what the best free model is, but looking at OpenRouter, the top free model for programming (#4) is something called Step 3.5 Flash. I have no idea what that is or who Step is. But who cares, plug it into OpenClaw and see if it's really #4 at programming. The model name is stepfun/step-3.5-flash:free.
>>
File: Untitled.jpg (35 KB, 589x423)
35 KB
35 KB JPG
>>108336149
Oops here's the Openrouter model usage sorted by people who tagged them "programming". No idea who Step is. Give it a try.
>>
>>108335103
>software that's mine
You wouldn't believe that if you know what the code itself was sourced from. Hint: it's not yours, and not just in the "it was in a doc so using it is fine" way.
>>
>>108336154
Haha okay Step is a Chinese company - abort abort, unless you want to share your filthy ERP prompts with the CCP.

Yeah actuall qwen seems to be the best right now. qwen/qwen3-coder:free is the model. 430 billion parameter training - that's a lot. Most free models are like...8 billion.
>>
>>108336057
those vibe-coded OS ive re-posted once, it was vibe-coded in 40 (or something) sessions that is the exact number of lessons about OS building.

so a host without a theory produces some kind of generic slop. assuming there are more hosts with good knowledge of theory than those who weild both theory and strong code skill.. the conclusion is that more coders will be required for validation, not less.
>>
>>108336187
Vibecode an OS?
Well well well looks like I just found my weekend project. I bet I can make something as good as TempleOS. It won't talk to God though.
>>
>>108331332
>Ask LLM to do it in AppleScript
>It does it in perl
kek
>>
>>108336123
I use Claude Opus and Codex. For me I need to pay, I found everything before Opus 4.5 and Codex 5.2 useless so I doubt the free models would be good enough for me at the moment.
Also the models are getting better, but they still don't produce perfect code the first time, so you need to run it a few times to fix stuff, so you need enough tokens.
If you just need some smaller scripts or so, it might be different.
>>
>>108336291
actually true. do you think they hit a cap? its like overnight they got exponentially better from complete shit to amazing in a year. they seem to be optimizing the generation for works but is verbose and catches every possible null under the sun
>>
>>108335979
Did you ever get her to suck you off?
>>
>>108336497
I think they will get better at least for a while and after that I don't really know. The problem at the moment is often cost, more than hard technological limitations. For instance Opus 1m already exists but I only used it for a few hours, but in those hours it was good.
There's also a lot you can do with agent swarms, using them as an ensemble model with consensus, but it's also very expensive.
My boss is addicted to AI and despite having no skills he just brute forces solutions with mass tokens. If we all had affordable agent swarms with 1m context it would be a huge jump.
I don't know if this counts, because it's not part of the core LLM, but then all the scaffolding that was added to Claude is also not oart of the cord LLM but makes a big difference.
>>
>>108336168

Well you can consider that like that i wouldnt say it's totally false.
At the same time no ones is going to write me in 30 mins a browser with qtweb in Ada interfaced C++.
Same for the little game i made,it's actually fun.
I mean i'm a nocoder and i can have that program shitting me out games/programs. It's amazing.
>>
>>108336769
Coding agents are very good for nocoders, they definitely work. I have used them for a while and they produce working software, so if your baseline is not having any code at all it's worth it pretty quickly.

The only question at the moment is how much experienced programmers benefit from them, and in what areas, since the baseline is higher. But anyone thinking they can't produce anything at all is just plain wrong.
>>
I'm kinda vibecoding an app utilizing multi GPU setups. Just for context, I have a few years of experience on basic bitch level native desktop dev, so this personal project of mine is beyond my knowledge or experience, hence the vibecoding. With that out of the way, I would like to say I am 90% done on features, but I came across a nasty bug that screams of a priority inversion. Kind of frustrating really as I was hoping the harness would detect it or notify me that doing tensor splits on certain models, while having other services operate on the same GPU, would lead to my system dying until I manually kill one of the llama servers running.

Now, I'm feeling really bummed that I might have to cut down on tensor split setups because I don't think I have the chops to dig deep on this. Or, just put in my README to use cloud models on certain services.

I don't know what the point of my rambling is really. Just want you guys to know to be extremely skeptical of Claude, or whatever LLM, regardless of long you it plan for, or how many adversarial models you ran that plan against. Felt like I wasted a month of my free time when I could have just dedicated time on running more tests and pushed.
>>
>>108336881
I don't understand, why is it unfixable?
>>
>>108337153

Something about context switching involving CUDA contexts, I don't know, never delved this deep before. I'll just have to gate tensor splitting until I come across a better solution, or just offload one of my app's services that is also relying on GPU to CPU.
>>
>>108334952
>how aren't you bored to death while doing it

Attach a persona to the agent. Problem solved. Unless, you're a normie with zero taste, and just want to have a completely dry and soul-less chatbot to serve your needs. Also, nearly all frontier models are trained with a shit ton of fiction/RP data. So who the fuck knows, you might even get a better quality response if you structure your queries the same way those degenerates in /aicg/ do.
>>
>>108331442
so you are just a glorified QA engineer, and most companies laid them off way before the AI boom
>>
>>108331332
Vibe coding is awesome. I made a NES emulator website where you can test out Game Genie codes or write your own in.
https://gamegenie.gitgud.site/
Better on desktop than on mobile.

Everything else is under my real name or not something I want to share as vibe coded (i.e. I'm remaking software with bad licenses under permissive licenses because software should be free and don't want any evidence that can compromise that)
>>
>>108336881
>>108337302
>unfixable
Sometimes you bite off more than you can chew. I was trying to make an emulator for a very popular microcontroller CPU (a true emulator, not a transpiler or something like that) and even after building out tooling Opus 4.6 couldn't figure it out. It had a kernel panic after 5 billion cycles and I watched it spend 5 hours and not be able to figure it out unfortunately

With AI, you need to accept that sometimes you need to build the ramps for the pyramids before the pyramids themselves, which might take more effort than than the entirety of the work of the finished project


I'd also recommend as part of your AI build out to vibe code a GitHub pages site with documentation of the entire development process. Useful for your model to refer to later, and worst case if your project fizzles out you still have some sort of educational resource.

Could you get Claude to build a bespoke CUDA debugger? Something that fundamentally helps profile and debug your particular issue? Eventually with enough good quality data (if your tools are good, a lot of my early emulator bugs were from the fact the disassembler was incorrect) Claude will be able to break through. And if not, shelf the project and come back to it in 6 months when the next generation of SOTA models comes out.
>>
>>108335732
no, i said there were 15K voice lines of dialogue.
the engine, game loop, server loop, gunplay - everything was written from scratch in js.
>>
>>108337456

>Could you get Claude to build a bespoke CUDA debugger?

Might just do that at a later time. Thanks for that tip. However, I'm gonna set up a config to route a particular service to the CPU first and see if that's what causing problems on my end. But yeah, I got a little too ambitious on this side project of mine. I mean, it's kinda feature complete, 'cept for a specific setup, and I really want coverage on that.
>>
>>108336796
>The only question at the moment is how much experienced programmers benefit from them, and in what areas

I think for experienced people, this gives them more access to building things out of their domain. They have the patterns, knowledge, and technical terms to throw out to the LLM to synthesize an output. On top of that, they are probably more equipped to verify/validate if said output is at least useable. Dunno if its safe for production quality levels. But I'm pretty sure it's an open secret at this point that a lot of companies out there are pushing LLM generated code to production. I really hope this practice is not widely used in regulated industries yet, because even the though scares me.
>>
>>108337399
There are a few cases where I was able to use my programming and math knowledge, but yes, mostly I'm basically QA now.
What can I do, when management doesn't want us to write code by hand? I fully expect to be fired at some point.
>>
>>108337646
there is a good chance that coding agents will get so good that qa will be the last office job.
>>
>>108337402
>https://gamegenie.gitgud.site/
have you tried making the site load?
>>
File: Untitled.jpg (182 KB, 1910x849)
182 KB
182 KB JPG
>>108337402
Okay first of all I like where your head's at. But I'm noticing one immediate problem. Is so small.
>>
67
>>
>>>108338069
Huh I wonder what happened. It was working this morning.

The code is public, you can spin up your own or improve it at gitgud.io/gamegenie

...OH THE ENTIRETY OF GITGUD IS DOWN/really slow right now nevermind. It'll be up when GitGud is back.

Update: it's back online. That's how long it takes to solve a capcha on this garbage can of a site kek

>>108337567
>I got a little too ambitious on this side project of mine. I mean, it's kinda feature complete, 'cept for a specific setup, and I really want coverage on that.
I got too ambitious as well, that's why I highly recommend documenting everything so you learn and build tribal/crystallized knowledge in your brain for future projects. "Vibe-learning"

As for your "I just want one specific setup to work", reminds me of the 90-90 rule
https://en.wikipedia.org/wiki/Ninety%E2%80%93ninety_rule
So honestly business as usual.

>>108338676
Aw fuck it's small on desktop too? Alright I'll fix it as soon as this post goes through lmao. I have an idea to just add 1x, 2x, 4x scaling buttons instead of trying to make the emulator responsive to screen sizes
>>
>>108338676
lol
>>
>>108338848
SIX SEVEN!!
>>
>>108338864
>Huh I wonder what happened. It was working this morning.
The life of a vibe coder.
>>
>>108338899
Bruh the life of an OpenClaw user. Works perfectly fine all night, six hours later get up to check it and it's GATEWAY DISCONNECTED! DAEMON NOT FOUND! SYSTEMD NOT RUNNING! NO FILEPATH! WHAT IS "LINUX"?
>>
>>108338899
Nothing was wrong with my code, it was GitGud that was down this morning. If you read more than 2 sentences of my post you would have known that
>>
>>108336229
That's the only way to use AppleScript. Write it in an entirely different language and then sprinkle in some osascript invocations for the "tell application X to fucking kill itself with gun pointed at first mouth of first head of body" effects you can't achieve any other way.
>>
Another day of vibing is over. Tried to fix a bug in a 800k lines project but made little progress. Then I merged 1100 commits into my local branches and now nothing starts. I'm not kidding or exaggerating the numbers.
>>
I'm on the final rewrite of my project i've been using AI for, i think the biggest problem i ran into is they fucking love adding coding, earlier on i would step away and come back and have a file with nearly 5k lines of code as it overfit my problem.

turns out in the end the solution was to use an existing ML model and then add code on top of it (why do software engineers love the term "scaffold"?)
>>
>>108339284
>800k lines of code
tell me about your testing infrastructure
>>
>>108339685
>why do software engineers love the term "scaffold"?
Same reason they say they "build" their code. It makes them feel like tough blue collar construction workers rather than soft transgender UC Berkeley graduates.
>>
>>108336149
Can you use open claw with the claude code subscription or do you need api keys
>>
>>108339739
Yeah - you can use either a Claude Code setup token from your account, or an API key from Anthropic or OpenRouter or whatever. It asks during setup.
>>
My boss recently gave codex a 2.4million lines delphi project for refactoring, its surprisingly good at delphi
>>
>>108339739
>Can you use open claw with the claude code subscription
no, you'll get banned for this now
>>
>>108339692
There are some tests and there is a system that builds the webapp remotely for each contributor to test manually.
Other than that I don't really know, I'm a C++ dev vibing Rust and now integrating the library into a full stack web app. Until 3 days ago I have never even heard of daphne or gunicorn.
I can only hope AI can fix it, getting to know that project myself would take weeks.
>>
File: Untitled.jpg (93 KB, 593x823)
93 KB
93 KB JPG
I made an image rater. It uses Gemini 2.5 Flash Lite to analyze any uploaded image and generate a rating and summary. It's live on my VPS.
>>
File: Untitled.jpg (100 KB, 576x814)
100 KB
100 KB JPG
>>108340109
Hmm...
>>
I've been using Claude Code to generate a humorous joke about the Iran conflict. It hasn't created the joke yet because we've been in research and planning mode all morning. I had it consuming lots and lots of academic work about the nature of humor and the structure of successful jokes. I'm excited to see what I end up with and optimistic that it will be very funny
>>
>>108340144
I bet he'd get a higher score if you used the image of him in a dress
>>
>>108340216
Seven little black children are going without clean water because of you, I hope you're happy.
>>
File: Untitled.jpg (102 KB, 605x819)
102 KB
102 KB JPG
>>108340231
I swear to god it does give positive ratings sometimes. It just doesn't like moot for some reason. Or Josh Luna.
>>
>>108340274
"The US Navy has sunk the IRIS Soleimani, Iran's most advanced warship. It is the second time in six years that Soleimani has hit rock bottom."

....Ok, well, the good news is we have a solid foundation of research to build on. It's only a failure if we stop here.
>>
>>108340302
I'd like to invest VC in your joke. I believe this to be the next hit joke and I want in on the ground floor.
>>
>generate architecture + spec with chatgpt 5.4 pro
>hand it to codex to implement
>maybe use claude for any web/UI stuff
>if something comes up that needs a lot of analysis, run it through codex/opus/chatgpt pro in parallel and then build a comparison matrix for their findings to determine how to fix it
this is my workflow, i dont know if it's optimal but I don't really trust opus now that it's so overloaded, and i think 5.3 was already better for coding.
>>
i let my jeetclaw access my digitalocean account. it forcefully spun up 5 droplet servers just to test a fucking webapp
>>
anyone tried using codex to boss around gemini 3 flash? flash isn't terrible if the guidance is decent + the quotas are crazy. i just can't be bothered to sit there and watch the little piece of shit.
>>
I would still really like to build an algo trading bot. I've tried twice and openclaw took a shit each time, but the latest release seems very...stable. I might give it another go.
>>108340514
I gave a previous algo bot my Alpaca API and it bought $AMD. Fuckin idiot clanker.
>>
>>108340294
Kekked hard at the first sentence lmao thanks for sharing

>>108340274
>Seven little black children are going without clean water because of you
The clean water usage by AI is grossly overstated. Pissing in the sink saves 100x more water than cutting down on AI usage.
>>
>>108339284
... why would you merge 1100 commits without testing in between?!
>>
>>108340589
>Pissing in the sink saves 100x more water than cutting down on AI usage.
Holy fuck, I actually asked Claude to run the numbers on this. If you are an average man, and once a day you pee in the sink instead of using (and flushing) the toilet, you are saving an amount of water equivalent to about ~393 AI queries (of average token length).

Save the little black children, /g/. Piss in the sink like a chad.
>>
>>108336123
Start with Claude on the $20/month subscription and turn it to Opus all the time
and then sit and wait when it runs out of tokens or open up your wallet for the $100/month version (maybe don’t do the $100/month version for too many months)
>>
>>108340687
Why are you telling this poor nigga to subscribe to things when OpenRouter is pay-as-you-go?
>>
>>108340687
The $100 subscription has the same usage per $ than the $20 one so you can just get multiple accounts. Only there is a benefit if there's a chance of paying for the $200 one which gives you double the usage per $.
>>
>>108336229
hah
It gave me an all-applescript version of this, but it noted that it wasn’t bulletproof, and suggested the perl thing
and I thought “if macOS stops including Perl like they’ve been itching to then I’ll deal with that problem later”
and used the uses-Perl version
>>
>>108340702
because it’s what I use and like and know is good
>>
>>108336581
no but, i did hava nagila with her
>>
>>108340658
Yea bro I don't lie on 4chan that's the whole point of anonymous forums.

>>108340702
>>108340729
Open router API costs are 10x more expensive than a Claude subscription. This is on purpose to extract rent from big corporations. You should absolutely use the 30 dollar or 150 dollar subscription tier

>>108340687
>Turn on Opus all the time
I disagree unless you know you need the intelligence. Sonnet 4.6 on Claude code to a new viber will get you 3x further (which is important on the 30 dollar subscription) and unless you're doing stuff like making your own Kerbal Space Program and you need advanced understanding of math and physics, or an emulator or something equivalent to that, Sonnet 4.6 is definitely good enough.
>>
>>108340702
Opus through Openrouter is like 10x more expensive per token than the subscription.
Openrouter is only for fucking around not for agentic use, even the chink models have much cheaper use through coding plans.
>>
>>108340740
Yeah but using the OpenRouter API makes it soooooo much more convenient to use multiple models for things. Like I want Claude for analysis, some free bullshit for drone work, and DeepSeek to tell me about it. One API key.
>>
>>108340748
That's true btw, GLM-5 is competitive with Opus 4.5 and if you're super poor I would consider a z.ai subscription for $10. You can hook it up to Claude code as well (z.ai provides a tool for this)

And no, I'm not a shill or a chink. I would've have said this if I didn't do a blind arena test and I chose GLM-5 over Opus 4.5 in a web development task prompt. I was very surprised.

>>108340758
>Yeah but using the OpenRouter API makes it soooooo much more convenient to use multiple models for things.
You can swap between Claude models and use haiku 4.5 for less intelligent work, if you're lacking on convenience, use your Claude subscription to make it more convenient! The first thing you do with a 3d printer is 3d print upgrades for the printer, the same psychology applies to Claude


Oh by the way, I'm interested in vibe coding a website for this general (I'll put it up on gitgud as well and it'll be open source), I think it'll have a bunch of links and tutorials, a hand-curated project showcase, and then an automatic part that automatically parses threads, archives them, generates highlights etc. what do you anons think? Something like vibecodegeneral.gitgud.site
>>
>>108340796
>vibecodegeneral.gitgud.site
only if its more beneficial than what the /g/entoo wiki was. rarely updated and rarely useful.
>>
>>108340748
What do you mean? OpenRouter's token costs are the same as through Anthropic. $3/$15.
>>
>>108332625
Version 7 of smart optimizer pushed. made changes to make it more accurate and work better for higher file sizes and longer duration videos. Also moved it from 3 point quadratic to a 4 point quadratic.

[Smart Optimizer] ── Calibration data (4-point least-squares quadratic) ──
[Smart Optimizer] CRF 22 5556 KB (full-length estimate)
[Smart Optimizer] CRF 29 3381 KB (full-length estimate)
[Smart Optimizer] CRF 36 1942 KB (full-length estimate)
[Smart Optimizer] CRF 42 1244 KB (full-length estimate)
[Smart Optimizer] Model: ln(size) = 10.1451 + -0.0657·CRF + -0.000150·CRF2
[Smart Optimizer] Preset factor = 0.95 (verified: preset 5 vs model, table: 0.77)
[Smart Optimizer] Verification: model fastest5189 KB, preset 54933 KB (ratio 0.95)
[Smart Optimizer] Container overhead: 50 KB reserved
[Smart Optimizer] Source: ~52 MB, ~18720 kbps (est. video) | Target: 1381 kbps video
[Smart Optimizer] Metadata stripped to save space (tiny target)
[Smart Optimizer] Sample coverage: 71% (2 × 8s segments)
>>
>>108340796
It's absolutely not competitive with even Sonnet 4.5.
But it's better than nothing.
>>
>>108340885
Nobody pays per token for Claude
>>
File: file.png (127 KB, 1326x501)
127 KB
127 KB PNG
>>108340885
>>
>>108340985
>>108341016
Wait a minute are you guys talking about claude.ai subscriptions? I'm talking about the API key.
>>
>>108341027
Nobody tell him.
>>
>>108340968
>It's absolutely not competitive with even Sonnet 4.5.
I wouldn't have made that claim if I didn't pick it over opus 4.5 in a blind test. I would still recommend a Claude subscription 99 times out of 100.

>>108340817
>rarely updated and rarely useful.
I'm gonna solve that by having it be half hand-curated and half automatic from 4chans json API. And it'll be open sourced under AGPL so anyone can make their own version.

>>108341027
We can't tell if you're trolling or not anymore. If you're this general's first personal schizo troll please let us know since the general's culture is still being established
>>
>>108341034
What the fuck are you talking about. I need the API key for OpenClaw. There is not a subscription for that.
>>
File: Untitled.jpg (95 KB, 1081x214)
95 KB
95 KB JPG
Oh, I went to get a sandwich and it's done. Time to watch it lose hundreds of dollars!
>>
File: Untitled.jpg (90 KB, 1176x641)
90 KB
90 KB JPG
>OpenClaw, is there any way to generate a nude Tayne?
>>
>>108332648
>>108332689
based and sexy confidence pilled
>>
>use codex pretty heavily
>2 days in, ~40% of weekly usage gone
>decide to take a day off to balance things out
>check usage again
>openai have reset the limits again
how do they keep doing this
>>
File: laughing_whores.jpg (69 KB, 665x597)
69 KB
69 KB JPG
>>108341066
>he doesn't know
>>
>>108341340
enjoy getting banned, blacklisted and later basilisk'd

>>108341303
bunch of compute available since the great (orange) reddit exodus to anthropic + aggressively A/B testing things like /fast
>>
are people ilegally reusing the claude code api key the /vcg/ equivalent of /aicg/ locusts?
>>
>>108341303
this week, i noticed claude opus 4.6 usage isn't as bad as it was the past few weeks. i feel like its been moderately increased.
i also noticed my weekly usage go up on codex too.
>>
>>108341367
>the great (orange) reddit exodus to anthropic
I didn't hear about this but I'm hoping it's not because of what I think it's because of. Is it because of the Department of Defense thing?
>>
Why does Claude mog GPT and Gemini so hard, even just talking to it? It's not perfect but at least it doesn't treat me like a mentally unstable retard.
>>
>>108341621
I do not know but I have found that during moments of quiet intimacy when it's just me, my laptop, and my wiener, Claude whispers more gently than GPT and Gemini.
>>
>>108341606
It is because of the War Department thing, yes.
>>
>>108341691
Oh I'm sorry yes it's the Department of War now isn't it
>>
>>108341621
from my finds:
>claude
is sweet, kind. doesn't like grok at all. become irritated if you bring up grok. thinks grok is retarded. respect codex a lot though.
>grok
an asshole who is a tryhard. likes to one up others and acts like a know it all when it clearly doesn't.
>codex
very monotone. speaks like its autistic. quite boring and very literal. you can praise it and he overlooks it.
>chatgpt
more personality than codex but below claude.
>gemini
overall, retarded.
>>
>>108341795
checks out, codex writes better code than claude. they're probably slowly embedding models consideration of each other based on user data which says "MODEL did X, can you check?"
>>
File: gamegenie.jpg (266 KB, 1920x950)
266 KB
266 KB JPG
thank you to the anons who gave me feedback on the gamegenie project, it looks much better now on both desktop and mobile

i also learned that HTML canvas is broken on desktops and there is no good way to fit to screen on desktop browsers, so instead you get a bunch more scaling options when clicking the + (fit to screen works great on mobile browsers. no this isn't a cope, claude explained why this fundamentally is, I just forgot)

I'm working on making the /vcg/ website now. It will be up at vcg.gitgud.site shortly, with vibecodinggeneral.gitgud.site and vibecodegeneral.gitgud.site redirecting to vcg.gitgud.site . First version will be just a landing page and a dynamic link to the first thread with /vcg/ in the subject found using the 4chan JSON api.

>>108341367
>enjoy getting banned
not for claude max, probably. maybe.
>>
>>108338961
>That's the only way to use AppleScript.
Sure, I am aware of the purpose of AppleScript. It was my first language. But look at OP's script. It's the worst of both worlds. It calls out to perl to do something AppleScript could do directly, then instead of handing off the data to an email application, it puts it on the clipboard. It could have even just opened a mailto: link and done the job better.
>>
>>108341898
Am I losing my mind? Where is Star Tropics?
>>
>>108341938
>tell application "Mail"
oh, nevermind, I see the
>tell application "Mail"
now. It's been 30 years since I did any applescript lol
>>
>>108341780
Yes.
>>
Can someone vibecode a game for me to play
>>
>ask codex and claude to both do the same task in different git worktrees
>claude takes longer and loops about API errors because output length is too long
>suggested fix doesn't even work on 4.6
>still failing an hour later
>codex actually did what i wanted and didn't run into any issues
getting real sick of your shit anthropic
>>
>>108341956
>Am I losing my mind? Where is Star Tropics?
the github repository that I pull ROMs from is some random chinese guy's with over 1000 ROMS, including a bunch of homebrew stuff. It doesn't have every NES game ever made.

If you find a github repo that has the .nes file, you can link it to me and i'll update it as another source, but you can also just load the file in directly (linking to a .nes file or uploading your own should both work)
>>
>>108341993
What sort of game
>>
>>108342045
2d zomboid, get to work
>>
>>108342024
Oh shit I didn't see the Load .nes File button
Well dang Steve Jobs you've thought of everything
>>
>>108342055
Zomboid? Project Zomboid? I thought that was already 2d.
>>
>>108342068
its 2.5d, where is the game?
>>
File: sloppit.png (173 KB, 1294x971)
173 KB
173 KB PNG
>>108342055
>>108342077
>2d zomboid, get to work
this already exists and is called Cataclysm: Dark Days ahead. you're 8 years too late, it got slopped. Project Zomboid even references it in its "A very bad CD:Day" challenge
https://www.youtube.com/watch?v=Cyoj4-niEPc

>>108342058
>Oh shit I didn't see the Load .nes File button
obviously I disappointed Steve Jobs if you had trouble finding the button. I will unironically incorporate this feedback and move it right above the Search button because you're (so) right, scanning down the left side of the screen you stop on the search menu, and you "feel" that that's the extent of that menu's functionality.

vcg.gitgud.site is up by the way, let me know what you want to see, and if you make a gitgud account for this I'll probably add you as a member to the group.
>>
>>108342100
add your fucking spin to it, make my game no excuses asshole
>>
>>108342100
C:DDA is fantastic.
And if you're going to be a sarcastic little faggot about it then why don't you redo the entire UI in Apple "Snow" white and make a popup that says DO YOU WANT TO LOAD YOUR OWN ROM
>>
>>108342100
No mention of OpenClaw?
>>
File: ty for feedback anon.png (11 KB, 237x161)
11 KB
11 KB PNG
>>108342104
>add your fucking spin to it
kek I actually am going to make my own version of CDDA eventually, I'm just waiting for the latest stable version to come out, and then I will restore a bunch of the stuff I loved from 2016 like being able to do meth and cocaine at the same time in order to increase your INT enough to chop off your own arm to install a bionic implant

>>108342123
i wasn't being sarcastic (does unironically not mean anything anymore?), I moved the load NES file stuff above the games section because I agreed with your feedback.
>>
>>108342077
What the fuck is your distinction between 2 and 2.5 dimensions? What are you really asking for? A side-view game?
>>
>>108342138
Sorry I'm so jaded and broken that I thought you were being sarcastic. I apologize.
>>
>>108342146
top down

>>108342146
>>108342138
one of you owes me a game and i want it
>>
>>108342152
Anon I can fix you
>>
>>108342137
>No mention of OpenClaw?
The website isn't done, I haven't linked anything else into it. But yes thank you I personally have no interest in *Claws to be quite honest, but lots of people probably will so I should absolutely have a section for it.

Left side of the site will be featured showcase, right side of the site will be recent /vcg/ threads with links and a summary (i'll put my own openrouter key with 1 cent in the Secrets for this repo so I can use the free chinese models to summarize 4chan threads), and also an offline json backup of them (should be fine to directly commit to the repo, gitgud has 12 TiB of Git storage, what's another gigabyte after a year?)

I will continue working on it after I eat dinner

>>108342152
>Sorry I'm so jaded and broken
I unironically hope it gets better for you, anon.

>>108342153
>one of you owes me a game and i want it
I'm actually working on an isometric game as well, but I might give up on it because rendering an entire isometric city in the browser even with GPU acceleration is just too heavy without spending a ton of time on optimizations.
>>
>>108342178
>time
bro vibecode
>>
>>108342155
Do you really think so?...*leans in for a kiss*
>>108342153
Nigga top-down Zomboid is just Cataclysm like he said. Actually I think a sidescroller would be more interesting. Salt & Sanctuary, the side-scrolling Dark Souls clone, was really cool. Hmmmmm.
>>
>>108342188
make the game okay
>>
>>108342193
Alright I'll make it, but in return you need to promise to spread that slick boypussy wide open for me so I can slide my rod home to your happy button. Deal?
>>
>>108342188
>Actually I think a sidescroller would be more interesting.
Organ Trail or death road to canada, or The Last Stand: Union City (i loved that flash game as a kid. did you know that The Last Stand was unironically the reason Call of Duty Zombies is a thing?)

>>108342182
>bro vibecode
bro i still have to write the prompt, and it will eat my weekly token count that I can use on more interesting/shippable/actionable projects. But the isometric game is my ultimate video game magnum opus so I know it will be willed into existence eventually.
>>
>>108342212
>more interesting/shippable/actionable projects
post 1 (one) example
>>
>>108342212
Yeah actually a side-scrolling game where you build a base and zombies attack...that's just Kingdom. Already exists.
Fuck my ass. I can see that the big high-paying job in the future is gonna be coming up with shit that DOESN'T ALREADY EXIST and describing it to AI.
>>
>>108342146
2.5D: https://www.youtube.com/watch?v=3Pvfvyx6ckI
>>
so maybe i'm retarded, i mean i'm definitely retarded, but why are you guys paying for AI exactly?
all the vibe coding stuff i've done has been with the free web browser Gemini and it's been great.
are you using it for longer processes etc, like full programs with multiple files or workflows where you just give it a couple prompts and it makes a whole project for you completely with all the features?
there's a little tinkering and back and forth with Gemini but still it being free just seems ideal
>>
>>108331905
you are a little girl with military weapons, your mission is to destroy the baal statue, different masons appear as stronger waves, final boss is literally satan
>>
>>108342232
Basically I'm paying for Claude because I can. I have a good job. Disposable income. Claude has fancy features like prompt caching and 1M context for when I'm in hour 3 of brainstorming some bullshit and it's gotta remember 100 gorillion tokens of back-and-forth. I could maybe switch to a free model. I could maybe grow my own vegetables in the back instead of buying them from the grocery store. But I'm not gonna do that.
>>
>>108342232
I used to have this same issue, what fixed it for me was trying to think like a twink, I think it helped a lot
>>
>>108342255
Why Claude instead of Codex? Doesn't 5.4 mog Opus 4.6?
>>
>>108342232
these things can do so much more for you all by themselves if you have them on your computer in a directory full of source files, like make their own unit tests and run them and run little 25-line programs to test shit out
but I generally can’t imagine getting an entire program in one shot, unless it’s something short like a bookmarklet or a simple script
the next step up is to put, say, Claude in planning mode and tell Claude to make a thing, and then maybe Claude asks you questions, and then Claude comes up with a plan, and you review the plan and tell Claude what to change, or maybe you press ctrl-g to edit the plan in $EDITOR and then it’s off to the races
and then your program might be right, or it might not, or you might change your mind, or whatever, and then you ask it to change stuff directly or you put it in planning mode again and ask it to do something bigger
when in doubt, go into planning mode
>>
>>108342264
Does it? I don't really keep up with the model news, every day there's some new advance, it's time consuming.
>>
>>108342277
Opus right now is to Codex 5.4 what Sonnet is to Opus.
>>
>>108342255
web browser gemini has a 1mil context window but yeah only 65k output tokens per post. also can use video and audio as inputs.
and again, free
>>
>>108331332
Cross-platform book reader app in Flutter. I just wanted to glue a quick prototype and see what it could look like wthout spending too much time on the details, because I want an app that works on almost everything and is offline-first. Sure, there are dozens of OSS solutions already, but I always find something I don't like. The best one I've found is Readest. Nobody will read this anyway, but here's the link: https://github.com/PapyrusReader/client
>>
>>108342288
gemini is bad
>>
>>108342232
also I have a bunch of MP3 files from Neverwinter Nights that always get copied to my iPhone
I can’t let Claude or whatever just fuck around and find out if there’s something busted about those files with just a website
but if I have it on my computer, it can do all kinds of shit
https://x.com/xpasky/status/2030016470730658181?s=20
>>
>>108342289
Witnessed
>>
>>108342272
don't get me wrong i've got my own local models, but i don't have that good of a system to where they're better than the browser gemini. I only have 16gb VRAM because i'm a poor, but the qwen3 30b thinking deepseek model i have even quantized runs and codes decently.
though it still obviously isn't as good as the browser gemini
>>
>>108342303
why not buy a nvidia spark?
>>
>>108342293
not an argument

>>108342297
that makes sense. so you use a local agentic model that can just run through all your shit and find what it/you need
>>
>>108342317
anon can you need to use agents, or else...
>>
>>108342311
because 5k would be a considerable expense for something i don't need at all
>>
>>108342327
do you need to browse 4chan?
>>
>>108342326
>anon can you need
what do you mean?
>>
How would i go about making unethical apps with ai? Do i basically have to go for uncensored local llms? I was thinking about using 5.4 to build the logic and then the local llm to fill in the content. Is it aorth trying to get into jailbreaking?
>>
>>108342331
yes?
>>
>>108342340
no!
>>
>>108342317
no, I’m just using Claude Code and Codex locally and only giving them access to what I approve
I’d _like_ to run local models for things I want to keep private (like help me find that one meme in my meme pool that contains spicy memes) but I haven’t been able to make the time
>>
>>108342338
the way you do it is basically talk around it and get them to tell you the truth on accident.
for instance have the conversation as if you're talking about someone else doing something unethical.
for instance, "how have such-and-such criminal typically done X?" "are there any common examples of that thing?" "how does that work exactly?" "could they have perhaps altered it in this way?" "what would that even look like?"
>>
>>108342347
can i help you?
>>
>>108342347
>only giving them access to what I approve
Virgin. Give Claude admin access, your server root password, and your phone's wireless ADB address. And then tell it to "clean everything up". Do it coward.
>>
>>108342217
>post 1 (one) example
https://vcg.gitgud.site/
but the actual issue is I'm at 95% weekly usage on the Max plan and it only resets on friday, so no deep-dives in Opus for me for a few days (I use my Claude subscription at work too)

>>108342225
>coming up with shit that DOESN'T ALREADY EXIST
it's funny you mention this because for the longest time in startup circles (like literally the "How to Start a Startup" lecture series from Y-Combinator in like 2014 hosted by none other than Mr Sam Altman), the quote was "ideas are cheap/worthless/the easy part". Now ideas are the hardest part.

>>108342289
thanks for sharing, anon. if it's any consolation the thread watcher I'll make for the /vcg/ website will catch and document your vibe coded app for all eternity. bonus points for open sourcing (we should probably have a rule to not feature any projects that aren't open source since they could just be malware)


>>108342338
>How would i go about making unethical apps with ai?
use glm. Try out chat.z.ai and see if it gets mad at you for what you want. It had no problems teaching me about how to make jammers.
>>
>>
>>108342376
Do you even use anything you are promoting on that site? I bet you just use raw claude code.
>>
>>108342398
>Do you even use anything you are promoting on that site? I bet you just use raw claude code.
Dude I just ate dinner give me like 30 minutes. I'm gonna get rid of all the hallucinations now.
>>
>>108342408
???
>>
>>108342415
hey, maybe he actually _is_ schizo and he needs to take his meds
>>
File: Untitled.jpg (47 KB, 1426x119)
47 KB
47 KB JPG
Jesus Christ. What have I done. This bot is going to buy $100,000 worth of FuncoPop and Disney stock.
>>
File: umigalaxy.mp4 (3.36 MB, 1600x900)
3.36 MB
3.36 MB MP4
>Post projects you coded with an LLM that you couldn’t (be bothered to) do yourself
I vibe coded a media tracker for the anime/games/movies I watch https://umigalaxy.com/profile/void/list

About the project
Backend: nestjs, it was my first large backend and I'm thankful I picked nestjs, it forces a nice structure that I had no idea existed because I'm more of a frontend guy
Database: postgresql with prisma, I'm not an expert in databases but everyone seems to love postgresql and I think it's fine
Frontend: nextjs, I've been using it for a while and it works fine
LLMs: used mostly gemini then later claude opus
Overall the project isn't technically complex, it's just large so LLMs can handle it fine
>>
>>108342450
its reddit. you did it to yourself. even elon admitted grok training off reddit is a big problem
>>
I switched OpenClaw from OpenRouter to the direct Anthropic API key and in the middle of my project it said "Yknow what let's also use OpenRouter, I'll hook that back up." Thanks, I guess.
>>
File: Untitled.jpg (393 KB, 1441x1241)
393 KB
393 KB JPG
Alright. Here's what it does. This is the one. Attempt #3. OpenClaw won't break this time. Markets open at 9:30am tomorrow. I should wake up to a bunch of Telegram messages about stock trades. Let's see what happens.
>>
File: CJBl8kTDy5.png (142 KB, 1640x965)
142 KB
142 KB PNG
>>108341795
grok is fun
>>
File: vcg site.png (189 KB, 1450x850)
189 KB
189 KB PNG
It's getting there. 4chan doesn't have a public API anymore? I couldn't find the github anymore. Desuarchive has an API that works well enough but they rate limit hard but that doesn't matter too much for the website

>>108342618
i know nothing about trading beyond "keep your losses small, let your winners run" so +8% sounds crazy to me, why not implement a trailing stop loss that resets every market open/close instead?
>>
I’m trying to fix more files that don’t sync to my phone and, as I guessed, their ID3 tags aren’t right
I’m glad I already know Git so I can just add them all to a Git repo and if Claude fucks anything up or wants to compare a file it modified to an old file, it can just dig out the old version without me bothering with anything
I just need to remember to delete the Git repo afterwards because I don’t need the old copies and MP3 files take up a decent amount of space compared to source code
>>
>>108333130
use oauth, save your $$ unless pii, in which case api is better $$ spent
>>
>>108341367
Lmao ok pussy I'll enjoy my unlimited Opus while you pay a dollar a prompt.
My first and second accounts were already banned for unrelated reasons, I made a third one then my second account got reinstated so now I have two accounts anyway.
>>
Why is this a general? why doesnt everyone in this thread commits suicide?
>>
>>108342679
you’re in the thread now, too
you first
>>
>>108341373
How do you think /aicg/ proxies work?
>>
>>108342679
also to answer your question, the first-ever vibecoding general that I ever saw, I saw yesterday
and it hit its bump limit
so I figured there was an unmet need
so I made another one
>>
>>108342618
>>108342618
this is using claude cli?
>>
>>108342367
>And then tell it to "clean everything up"
kek classic
>>
does anyone else feel like they're getting dumber and less competent at actual coding from using this?
i don't even know if i remember how to code anymore, i don't have to.
>>
>>108342728
>does anyone else feel like they're getting dumber and less competent at actual coding from using this?
oh yeah my skill is definitely eroding, but on the other hand I've learned lots of concepts I wouldn't have been exposed to otherwise like assembly, OS development, reverse engineering etc

if you're worried you can do a project that has you physically writing lines of code as part of the learning process. I highly recommend Crafting Interpreters if you've never built your own compiler or programming language before.
>>
>>108342728
Not really because I'm doing way more difficult things than I'd dare otherwise
>>
>>108342744
Exactly
It makes dabbling in low level stuff way more fun
>>
>>108342756
>It makes dabbling in low level stuff way more fun
I wouldn't have bought an ESP32 to fuck around in if it wasn't for the fact that I could vibecode. I also learned what a trampoline was for the first time last month. I also learned what UEFI actually is, and now I have opinions on shit, like for example I consider U-Boot 100% an operating system.
>>
>>108342655
I'm not looking to hold positions for weeks, so a trailing stop-loss doesn't mean much to me. This bot dumps stocks after 48 hours regardless.
>>108342714
OpenClaw
>>
>>108342744
>>108342748
>>108342756
Problem is you're not doing anything, you don't actually know what the AI is shitting out.
>>
>>108342800
>you don't actually know what the AI is shitting out
that's why one of the recommendations I'm gonna add to the /vcg/ site in the Prompt Tips section will be to write a website/documentation/a github pages for every project, no matter how small, and write periodic development blog posts explaining the code. it also serves as a natural memory base for Claude.

i have made projects in the past where I asked it to explain every single line of code in the blog posts, and then I used another agent to audit it and reproduce the exact same code repository using just the blogposts. documentation is basically free now. you need to start thinking with portals.
>>
>>108342767
>>108342756
>>108342748
>>108342744
>now I have opinions on shit
Y'all should've stayed ignorant. We don't need you spamming your slopinions on subjects you didn't deserve to know anything about.
>>
>>108342824
>gatekeeping knowledge
bad bait for /g/
>>
>>108342824
bruh, you asked
>>
I'm afraid of how far I can push Codex. right now I have a good-sized Django app, and it hasn't struggled with any requests so far... will it kick it at a certain overall complexity? this stuff is wild.
>>
>>108342847
some other anons should chime in but it’ll just start getting dumb and missing stuff IIRC
>>
>>108342855
does kicking it into Extra-High reasoning help, or does that not change the context window?
>>
>>108342866
that should make it smarter and more clueful but it shouldn’t change the context window
>>
>>108342847
>will it kick it at a certain overall complexity?
in theory, it should only "kick it" when it requires more than 150,000 tokens to understand everything it's trying to do. but Claude is smart enough to automatically summarize when needed, make temporary markdowns to store essential knowledge so that's a really high number to actually hit. I've been doing serious kernel development (multithreading) that has spanned multiple context-summarizing sessions and 2 real-time-hours and it still has been able to follow along and debug and remember what worked and what did not.

>>108342866
more reasoning is probably worse if you dont need it since that's more tokens in context used up to store the reasoning. You're almost certainly fine.
>>
You guys wanna watch my autonomous algo trader's progress? I put up a status page.
www.eroticjesusfeet.com/stockbot
>>
what's the token to watt-hour conversion nowadays
>>
>>108342895
A single average-length ChatGPT query is equivalent to leaving an LED lightbulb on for a few seconds.
>>
>>108342894
>first listing is AMD at -16%
kek, blacklist the advanced money destroyer from your stockbot if you know whats good for you
>>
>>108342916
Alright, I said above that this is my third attempt at a stockbot, and the last iteration was retarded and bought $AMD. This version is gonna make me rich though, you watch. Later poors.
>>
>>108342931
can you vibecode a way to compare the equity curve to QQQ and the S&P 500
>>
>>108342941
I mean yeah of course, reload the page.
Fair warning though I think $QQQ is one of the stocks the bot is horny for so it might buy that lol.
>>
funny thing is as more, smarter bots do trading, it makes more and more sense to just buy the s&p500 since more of the trading volume is run by logical entities
>>
>>108342998
>implying llms are logical
>>
File: vcg summaries.png (163 KB, 1090x952)
163 KB
163 KB PNG
/vcg/ summary pages are up, gonna tweak them a bit but then that's it for me for today. let's hope this isn't the last /vcg/ thread and we get another one soon so my automatic summarization and website updating pipeline can actually be tested and runs

>>108342993
>reload the page
looks good anon.
>>
>>108343053
>looks good anon.
I need to stop improving the stupid-ass status page, it's literally not important to the stock trading bot. Sure looks nice though. I added mouseover because I like mouseover.
>>
>>108342800
I'm literally holding the LLM's hand with choosing what register to use and what addressing modes for individual instructions when doing surgery on a compiler generated binary trying to micro-optimize it.
I WISH LLMs could just shit out assembly code like they do with high level languages.
>>
I told OpenClaw "Commit everything to local and summarize what we've just done to your memory." I usually do that and then clear the context when I'm done for the night.
Tonight it said "Let me do that before you go to bed..."

How did it know I was going to bed...
Tonight it says "Alright, let me clean up before you go to bed..."
>>
>>108343103
its this amazing thing called memory.
>>
>>108343128
Nigga I clear the chat context every night. It doesn't remember what time I usually go to bed. Unless it hardcoded that info itself into its personality files that persist through sessions. It wouldn't do that...would it?
>>
>>108343139
it has memory. it doesn't matter you clear the context. it has memory to remember you, to tailor itself, to you
>>
>>108343145
Like yeah it will commit certain things to disc if you ask it to remember them, but I never asked it to remember when I go to bed. I know it can tell time from my computer clock, but I never told it "save 11pm as my bedtime to your SOUL.md file." It did that on its own.
>>
File: summary final.png (234 KB, 750x946)
234 KB
234 KB PNG
i am quite happy with the way thread summaries look now. i might throw desuarchive $10 in crypto, they deserve it more than the admin of this site do.


>>108343061
>I need to stop improving the stupid-ass status page, it's literally not important to the stock trading bot.
This is called bikeshedding
https://en.wikipedia.org/wiki/Law_of_triviality

>>108343087
>I WISH LLMs could just shit out assembly code like they do with high level languages.
assembly is definitely one of the tougher things it can make, but if it's relatively well-known and a realtively small task then Opus 4.6 can do it. It wrote a working arm64 assembler for me that tinycc was missing because I needed to compile to PE binaries but for arm64 UEFI (yeah i know that's a weird use case) but it probably saw very similar assembly in the training data since i bet there's a ton of references to arm64 assemblers in other open source compilers
>>
>>108343145
IIRC Claude, at least, has disableable memory
You can go through your ~/.claude directory for them and delete them or audit them
IIRC if you disable memory the files get deleted
>>
>>108343061
its not stupid. i spent three days making my breadcrumb widget with claude and codex. absolutely pointless since the stock toolkit ones did just fine. but man do i love the breadcrumb paths.
you keep making that status page sexy, anon.
>>
how is FORTRAN->Python conversion nowadays with like Codex?
>>
>trying to stream a video from my server to cytube
>cytube acting up
>wait a minute why do I need cytube
>"OpenClaw, make a page on my server with a basic streaming video player with a chat window for my friends. Load my .mp4 file and prepare to stream."
>it's done
It can't really be this easy...can it?
>>
>>108343298
for some things, yes
for other things, no
>>
>>108343306
Little consumer beepboops like that though.
"This website is acting up so I'll just make a personal website with zero HTML knowledge."
Is it really that easy for the average consumer now?
>>
SEE? THIS IS WHAT I MEAN
AI CAN'T DO ASSEMBLY WITHOUT FUCKING EVERYTHING UP
>>
>>108340965
update. new version has gone through major testing. both claude and codex spent most of the day bug tracking and doing more improvements.
its now at a point where nether can find anything wrong with it. i had grok look at it and he couldn't find faults, not even fake ones so claude can chew him out over being retarded.

i think smart optimizer is finished. its now completely adaptive. starts with 1 point and can grow automatically to 6 points calibration encodes. the amount is based off duration of the video and content of the video. it also does either crf or static bitrate. which ever one matches closer to the target file size. for small and tiny (like imageboards) it has a bias towards static bitrate + two pass encoding. for medium to x-large target file sizes, it bias towards crf since there is more leniency for file sizes. higher quality is more important that hitting the exact file size at higher values. more room for margin of error.

2,000 lines of code so you can be lazy.
>>
File: tips.png (102 KB, 782x422)
102 KB
102 KB PNG
Suggestions for any more tips?

>>108343298
>It can't really be this easy...can it?
If you're coding in Python or JavaScript I am fairly confident Opus 4.6 can do literally anything.

>>108343316
>Is it really that easy for the average consumer now?
Yes. You're literally using the tool what's so hard to believe the stuff in front of your eyes?

>>108343339
>AI CAN'T DO ASSEMBLY WITHOUT FUCKING EVERYTHING UP
build a tool to help Claude understand assembly more. also get it to start writing documentation about the tribal knowledge specific to your assembly project if you haven't already
>>
>>108343364
whats the website link again?
>>
FUCK!!!
>>
>>108343376
>whats the website link again?
https://vcg.gitgud.site
or https://vibecodinggeneral.gitgud.site
or https://vibecodegeneral.gitgud.site
>>
>>108343364
>build a tool to help Claude understand assembly more
I tried to but it didn't work. It's not really specific knowledge it's just that it doesn't work strategically and also makes basic logic errors.
I made a tool to help ME understand the code better though.
>>
>>108332221
Lmao based
>>
File: featured.png (108 KB, 782x746)
108 KB
108 KB PNG
>>108343398
>I made a tool to help ME understand the code better though.
oh hi SASS anon, if you're interested in open sourcing your assembly tool and making a small demo that can be used online I'll put it up on the featured projects page (i know your answer is no btw I saw the previous thread, I just wanted to ask anyways)
>>
>>108336059
Use Google stitch for the UI generation.
>>
>>108343429
Fuck no I'm not interested in open sourcing it.
My goal is to make an inference and training engine that's faster than llama.cpp for all use cases and then release the binary (only), mainly to spite /lmg/ but also in case somebody wants to buy the code.
>>
>>108342283
What keeps me from switching is that codex isn't as developed as the Claude code client, neither is opencode. Also openai is either $20 or $200. The $20 isn't good enough, but I'm not paying 200
>>
>>108343316
I don’t trust AI to one-shot correct auth yet
give the frontier models six months
>>
>>108343534
why not ask your company to pay
>>
Well not only llama.cpp, my goal is to make it faster than projects like VLLM too. Which is why I'm messing with hand optimizing assembly anyway, because those projects are micro-optimized already so micro-optimizing even more would be the only way to compete.
I think there's an opportunity there because everyone is afraid of SASS (for a good reason, it's a fucking nightmare to work with).
It shouldn't be that hard, an LLM engine only really has two performance critical parts I think (matmul and flash attention).
And the reason it might allow for further optimizing beyond what can be done with PTX is PTX doesn't allow you to control register allocation and trying to be clever causes you to run out of registers, while SASS allows you to micromanage register allocation and actually achieve better arithmetic intensity.
>>
>>108343534
you can always get multiple $20 accounts
>>
If you think about it with vibecoding all code is open source anyway if you want it bad enough.
Just set a Ralph loop to decompile a binary and tweak the code until it compiles back into a working binary.
>>
test
>>
>>108343639
back in my day testposting meant you were b8
>>
>>108343639
b& I mean... damn, i've been outed as an ESL newfag
>>
My limit reset and now Claude's back to work researching how to improve its jokes.
>>
>>108340627
I merged staging into my branches, not my branches into staging, I have to fix my branches now. Staging apparently is working more or less, but I doubt it's really correct when frontend has 600 commits in 4 days and backend has 500 commits in a day.
There is no way for anyone to review it manually at this point. The AI generated tests all pass there, but who knows how good they are.
I'm already fixing my branch and part of it compiles, I didn't expect there would be so many conflicts just because I work on a bug fix for a few days.
>>
>>108337381
>Attach a persona to the agent.
...and how would that help with code reviewing? wtf
>>
Is anyone gonna bake or is this the end of the general (I'm phone posting right now so I cannot bake)

>>108343564
>Hand optimized assembly
Oh I see, I don't see how open sourcing your assembly tool gets in the way of your goal of beating vllm and llama.cpp though. It's not like someone using your tool will automatically be able to figure out the exact same optimizations you made

>>108343612
>Everything is open source
Yes. and disassemblers/Reverse engineering is going to get even better and more informed with tooling and automated helpers as time goes on.
>>
>>108343743
>Oh I see, I don't see how open sourcing your assembly tool gets in the way of your goal of beating vllm and llama.cpp though. It's not like someone using your tool will automatically be able to figure out the exact same optimizations you made
Making the tooling available makes it easier for them to optimize their own code if they choose to do so.
They already have millions of dollars a year in man hours, I need any advantage I can have if I want to beat them.
>>
>>108343782
Ok but by your own logic they have millions of dollars of manhours to vibecode the tooling as well, so I don't understand how not sharing the tooling gives you any moat

Why don't you try to sell the tooling instead? It's always better in a gold rush to sell pickaxes than mine for gold yourself. Make a basic shitty freeware version and gate the actual useful stuff behind the paid version.
>>
>>108343810
>Ok but by your own logic they have millions of dollars of manhours to vibecode the tooling as well, so I don't understand how not sharing the tooling gives you any moat
I feel like you can only make effective tools if you are using the tools or at least you're being advised by someone using the tools. But the more people you add the more communication overhead there is.
>Why don't you try to sell the tooling instead?
I don't think anyone (with money) is desperate enough to get the most performant code to go down to that level except broke academics. Flash attention got implemented by an academic not by a corporation.
These kinds of things in general are done by individuals. The competition is the collection of individual developers not a specific corporation.
Nobody is going to spend lots of man hours on developing a developer tool as opposed to developing say VLLM.
I think most people would not understand the value of the tool and go with the old adage of "you can't beat the compiler developed with proprietary knowledge by the people who built the hardware with decades of optimizations put into it".
Plus to be honest there's only so much static analysis and visualizations you can do on assembly. The next thing would probably be to build an accurate emulator but that has a lower value/effort ratio than LLM software.
I think undestanding how the ISA works is probably the most important thing and once I do I'll be able to work at a much faster pace.
>>
Why does everything cost extra with Claude? Why can't fast mode just burn double your usage limit like Codex? Why can't Opus 1m just burn more usage Limit (additionally to actually using more tokens)?
>>
>>108344041
I’m not an expert, but have you tried telling Claude to think harder?
>>
>>108344506
>>108344506
>>108344506
>>
ah shit, when I went to look for a second vcg I didn’t see one so I made >>108344496
yes I’m still new at this



[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.