[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

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

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


[Advertise on 4chan]


File: 1752809095275962.png (2.76 MB, 1254x1254)
2.76 MB PNG
A general for vibe coding, coding agents, AI IDEs, browser builders, MCP, and shipping prototypes with LLMs.

►What is vibe coding?
https://x.com/karpathy/status/1886192184808149383
https://simonwillison.net/2025/Mar/19/vibe-coding/
https://simonwillison.net/2025/Mar/11/using-llms-for-code/

►Prompting / context / skills
https://docs.cline.bot/customization/cline-rules
https://docs.replit.com/tutorials/agent-skills
https://docs.github.com/en/copilot/tutorials/spark/prompt-tips

►Editors / terminal agents / coding agents
https://opencode.ai/
https://cursor.com/docs
https://docs.windsurf.com/getstarted/overview
https://code.claude.com/docs/en/overview
https://aider.chat/docs/
https://docs.cline.bot/home
https://docs.roocode.com/
https://geminicli.com/docs/
https://docs.github.com/en/copilot/how-tos/use-copilot-agents/coding-agent

►Browser builders / hosted vibe tools
https://bolt.new/
https://support.bolt.new/
https://docs.lovable.dev/introduction/welcome
https://replit.com/
https://firebase.google.com/docs/studio
https://docs.github.com/en/copilot/tutorials/spark
https://v0.app/docs/faqs

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

►MCP / infra / deployment
https://modelcontextprotocol.io/docs/getting-started/intro
https://modelcontextprotocol.io/examples
https://vercel.com/docs
https://mcp.desktopcommander.app/

►Benchmarks / rankings
https://aider.chat/docs/leaderboards/
https://www.swebench.com/
https://swe-bench-live.github.io/
https://livecodebench.github.io/
https://livecodebench.github.io/gso.html
https://www.tbench.ai/leaderboard/terminal-bench/2.0

►UI/Frontend
Figma Make
Lovable
Claude design
https://uiverse.io/
https://ui-ux-pro-max-skill.nextlevelbuilder.io/
https://stitch.withgoogle.com/

►Previous thread
>>108817026
>>
File: 1753308687865534.png (288 KB, 1080x741)
288 KB PNG
Codex users: This is your reminder to use your tokens as there will be a reset later tonight.
>>
>claude randomly resets
>my weekly limit was going to reset tomorrow
you bet your ass I worked late
>>
>>108833778
>weekly limit resetting tonight instead of in 4 days
cool
>>
File: 904768907638902.jpg (60 KB, 789x1070)
60 KB JPG
good morning goyim

what are we vibe coding today? will your asian girlfriends be spending time with you this evening?
>>
File: 1753138457138814.png (655 KB, 1800x1360)
655 KB PNG
>>108833854
>what are we vibe coding today?
Still working on my DS5 companion app. I just need to figure out how to send audio through the headphone jack now. I found a workaround to get the mic working without stutter.
>>
File: 1697525163181069.png (1.07 MB, 1284x1646)
1.07 MB PNG
>>108833864
comfy ui

ai asian boobies for your efforts
>>
>weekend reset
based based
>>
Well, I reinstalled Codex. I lasted what, two days? Fuck my stupid chungus vibegod life.
>>
>>108833735
how can you talk shit on snailcats when they're so cute
>>
>>108831854
it’s always a good thread when the exoplanet guy chimes in
die in a fire
>>
>claude’s weekly limit randomly reset while still at 60%
>it’s going to reset again in two days.
>Already burned so much time on projects before and now just want to do to something else.
>>
>>108833778
fuuuuuuuuuu
>>
File: 1732186167468489.png (187 KB, 500x500)
187 KB PNG
>>108833778
>tfw limit resets tonight
>gf wants to go out for dinner and spend romantic evening bonding and connecting

use case? i need to code, i really don't want to go and i think i may hate her
>>
>>108833778
I'm slopping as fast as I can.

>>108833792
Claude got reset this morning as well. For Claude, usage got sent back to zero, but the weekly reset day did not get pushed back, making it effectively free usage; wonder if it will be the same with Codex.
>>
>>108834027
>my lobster is too buttery
you really have nothing better to do?
>>
Best way to buy chatgpt pro with crypto?
>>
>>108834049
You can pay with crypto on OpenRouter I think. Not a pro account though.
>>
>>108833735
https://www.youtube.com/watch?v=juHv_Vi4giU
https://www.youtube.com/watch?v=juHv_Vi4giU
https://www.youtube.com/watch?v=juHv_Vi4giU
>>
>>108834027
your gf is more important
>>
File: wtf.png (71 KB, 632x595)
71 KB PNG
>>108834067
if i make it with vibe coding then i can afford to upgrade
>>
>>108834082
nobody’s making a buck off this stuff
likely not you either
go hang out with your girlfriend
>>
>>108834061
buy an ad
>>
gpt-5.5 is still talking about goblins and witches. Everything reminds me of her.
>>
I'm going to be so rich, guys.
>>
>>108834182
Codex or GPT5-5? I never had that.
>>
File: file.png (430 KB, 1920x1050)
430 KB PNG
Tactical globe I've been building.
>>
>>108834278
whats it for?
>>
File: 1684683689411159.jpg (45 KB, 610x600)
45 KB JPG
>>108834278
that's cool as fuck
>>
>>108834282
Planetary resource control and force projection. You move swarms, hold corridors, contest nodes. Still early. Honestly still figuring out the broader architecture. The vision's way ahead of me. It's all PyQt5/OpenGL, no engine. With my background (read no background), it'll take time.

>>108834286
Thanks man, so is that cat :3
>>
>>108834176
reddit moment
>>
File: >reddit.png (1.51 MB, 1254x1254)
1.51 MB PNG
>>108834309
>>
>chatgpt keeps telling me to maybe take a break after we implement a new feature in codex every time

jews don't want the goyim to vibe code their way to victory. feeling sleepy? take a load off. have a big mac. spike your insulin. watch porn. swipe on dating apps
>>
>>108834339
I got Claude saying something would take a half day worth of work but it only happened in like a couple hours wall-clock time
imagine a computer that doesn’t understand that time is almost nothing to me
>>
>>108834203
Codex runs gpt-5.5 my dude. Unless you change it manually.
>>
5.5 EXTREME FAST
CONSUMER OF TOKENS
GO FORTH AND VIBE
>>
>>108834356
>Claude saying something would take a half day worth of work

Claude constantly states bullshit like this. Yeah, it'd be half a day of work if I were a skilled developer writing by hand. I'm not. I'm a fucking retard with a chatbot.
>>
>>108833778
Do they want to stress test their servers? Bit strange to announce it so much in advance, since they know a lot of people will start just burning tokens.
>>
>>108834392
Sometimes it might be right, it had put itself to sleep for 85 minutes for the second time today, so I looked at what it was doing and it was running about 2000 tests against an application, which is fine, but was starting and stopping the full application between each test instead of doing them as a single batch like it should have. Once it was fixed, it got done in about 30 seconds. :(

My guess is that it's an artifact of their adaptive thinking and effort. They tried to make it learn how many man hours something would take as a proxy, but it keeps leaking in the replies.
>>
>Commit created. I’m pushing ... now.
NO CODEX YOU BLOODY BASTARD I WASN'T READY THEY'RE GOING TO LAUGH AT ME
>>
File: 1771032348203770.gif (3.71 MB, 480x266)
3.71 MB GIF
>cycling through multiple slop apps while 90s music is blasting through my headphones
this is real now
>>
>ask chatgpt how I should ask codex to make something
>just works
heh I'm a middle manager(waste of air)
>>
w-wheres the reset? you guys promised a reset...
>>
>>108834724
probably midnight (hopefully). They wouldn't lie about this
>>
no reset yet
good thing too, because I haven’t used much
>>
File: 1755755809936891.png (654 KB, 1800x1360)
654 KB PNG
>>108834849
I blew through everything on xHigh. I regret nothing, my 4th UI pass has done wonders, I feel like I'm almost there.
>>
is it worth subscribing to claude just to do frontend work?
>>
>>108833864
Hey anon, do you know anything about the PS5 media remote? It detects but doesn't send any button press events. I was thinking of vibe coding some kind of fuzzing something or another to brute force whatever magic bytes it's waiting for, but what you're working on might touch on that if I'm lucky.
>>
File: 1757514304741540.png (565 KB, 1889x906)
565 KB PNG
>>108835365
>Hey anon, do you know anything about the PS5 media remote?
Unfortunately no, but it should be pretty straightforward compared to a controller. Are you trying to use it on your PC or are you trying to emulate as one on your PS5? Either way, I recommend getting a Pico 2W, it'll make the process significantly easier. They're like $7
>>
i automated nearly all of my work flow
what do i do with all the free time at work
>>
>>108835472
Trying to use it on a PC, it shows up in Linux as a bluetooth generic input device, but no buttons or anything. I'll write something to spam bytes at it until it does something
>>
Is the chudgpt pro any better than non-pro? I played around with the paypig model for the first time and didn’t really feel much difference compared to the poorfag $20 one
>>
>>108834182
I get goblins, gremlins and barbarian galore, almost every single thread
it’s amusing desu
>>
>>108835491
Make your own apps
>>
>>108835606
I'm getting random squirrels
>>
>>108833864
>dualsense edge
Enjoying the squeaky ass dpad? Enjoying the glued rubber grip that WILL slip off?
>>
>>108833902
get stepped on
>>
>>108833735
repost bc thread died

back again with prompting models for 'general knowledge' for a game.
I couldn't get reliable output from the local models (llama3.1, qwen3.6, ministral-3), they kept inventing new words (I ask them for a category and as many words that fit the category - it's for a word search).
so I switched to Claude Sonnet and because I'm vibecoding pretty much everything, I let it propose a prompt. it suggested passing a list of categories already generated to avoid duplicates, but it seemed to me it was trying to outdo itself with each generation, coming with some outlandish ones and then making up words as well.
so we tried to ground it and pass it only the early generated categories or none at all... then it kept generating
"category": "Parts of a Camera"
over and over again, with slightly different words. the temp is set to 0.7

wtf am I doing wrong?

>>108832713
thanks anon, but I don't want deterministic. each time I call with a prompt I'd like to get a new Category with matching words (so this part needs to be real world knowledge based).

Claude suggests I add domain topic and rotate them, but I feel that will end up doing the same as soon as I exhaust the list and go to the first domain topic. plus I don't want to think lmao, I wanted the AI to make up shit for me.

It also suggested adding UUID to force randomness from the model.

I'm thinking I should go to the original prompt, and add all the topics I already have, then make sure to remove anything too wild before making another request. so that it stops the feedback loop that led to it trying to outdo itself
>>
>>108835619
Neither of those are problems for me. Also I can just use adhesive to put the grip back on when/if it does fall off, I'm not going to be an idiot about it.
>>
>>108835687
Sorry can't we can't cure autism yet
>>
File: trvke.png (211 KB, 675x499)
211 KB PNG
>>108833735
>>
File: 32o8472309498539485.png (371 KB, 896x530)
371 KB PNG
I have a sneaking suspicion models don't want you to succeed.
I've been discussing with Gemini 3.1 Pro, Opus 4.7 and GPT-5.5 about releasing my stupid little game to the app store and doing some gorilla marketing. Basically I just wanted some basic strategies.
I knew I most likely would make barely ANY money and considering app store fees etc. it would be an overall loss, but with a little luck, who knows?

Anyway, even after telling them I would be willing to "risk" losing the $99 app store fee they keep berating me that it's not worth it, I will get DMCA'd and sued and thrown in jail and assraped and I should just go back to being a good wagie and shut up.
>>
>>108835881
It's free to develop the game. Putting it up there is also quite free.
>>
File: 1768826872820483.png (365 KB, 900x900)
365 KB PNG
I solved the final two issues on the DS5 controller that were unavailable for bluetooth
>stutter free duplex
>headphone jack outputs audio to headphones
There's literally no more outstanding features left for me to add (for the controller).
this feels pretty good.
>>
>>108835687
I don't understand what you're trying to do but you should know all modern models work best with a temp of 1. You shouldn't use lower temps than that or you risk infinite loops etc.
Also it seems like you could benefit from a more predetermined loop than just letting the model come up with stuff.
>>
>>108835881
They are trained to avoid legal liability at all costs. Opus 4.5 used to be able to bend it's worldview to you so if you asked it after a 200k token conversation letting the model know your worldview it would be much more likely to return something that aligned with it.
Or just use an abliterated model.
>>
I'm starting to get demotivated again
now that I have 2 subcriptions, I'm getting less inspiration to do previously "fun" projects like game mod, agent tool etc
>>
>>108833735
>start opencode
>swap model to minmax
>uoo sugoi so fast!
>fast forward 2-3 days
>slowest shit ever
is this normal? is this how f2p vibecoding works?
>>
Ever since I turned 19, for the past ~8 years I have avoided leaving any sort of digital footprint at all.
I cleaned my cookies at any session, used duckduckgo instead of google. I had no social media, no google account, no youtube account, no instagram, refused to use early versions of chatGPT, etc.
Until I discovered LLMs.
Now I have 3 AI accounts connected to my real name and credit card information that have tons of personal information saved about me in a chat history, from employment status, to relationships to psychological questions.
I have become so AI brained that I have basically outdone 8 years of digital privacy.
And I don't even care.
Unironically considering creating an instagram account to partake in normal life again because at this stage, what's the point of trying to resist any longer?
>>
File: aqtw8d.jpg (82 KB, 928x500)
82 KB JPG
>>108836162
Wanna hear how much all your efforts did to keep the government from tracking you?
>>
>>108835881
The fuck costs $99? Pretty sure I only paid a one-time $25 fee to become an "Android developer" and get my shitty app on there.
>>
File: 1773660807488628.jpg (21 KB, 448x446)
21 KB JPG
>>108836162
not really related, but recent AI make me less /pol/ brained
like that one chinky professor is claiming AI is just a theatre, like, bitch do you even know how impressive and useful AI is right now
>>
>>108836252
lol that 'professor' jian guy? negro is catnip for midwits
>>
>>108836252
>that one chinky professor
"Professor" Jiang? That dude is a major, how you say, 男同性恋者?
>>
>>108836162
>used duckduckgo instead of google
You just traded one kosher search engine for another.
>>
Day 52. There are still goblins in the code. Legal gremlins sprouting up everywhere. My server is haunted by forest spirits and there's goddamn WITCHES in the Android SDK doing God knows what blasphemous dark magics.
Time to get to work.
>>
>>108836077
>They are trained to avoid legal liability at all costs.
2 different models individually came up with a scenario where Nintendo will sue me for millions of dollars for accidentally copying Super Mario assets
>>
>>108836330
kek
>>
>>108833778
THEY LIED.
>>
File: Screenshot 2026-05-16.png (25 KB, 883x129)
25 KB PNG
Apparently Anthropic does not want my money
>>
>>108836235
apple charges $99 a year to publish on appstore i believe lmao.

>>108836378
are you copying super mario assets?
>>
>>108836977
>are you copying super mario assets?
my game has nothing to do with Nintendo or Super Mario i don't know why it keeps mentioning it lol
>>
>>108836999
seems like a weird thing to bring up unprompted.
tb h your experience also sounds a little unusual esp with respect to gemini 3.1 which will usually just go along with whatever you're saying and describe it as the greatest idea ever.
you can probably just reframe the question more positively if you want the clanker to agree with you.
>>
>>108837019
Incidentally out of the 3 Claude was the most pragmatic one and also the only one who stuck to its conviction
GPT produced by far the most legal yapping but at least most of it seemed to be correct (although extremely conservative covering every insanely unlikely edge case). I could get it to be more pragmatic after some nudging although it prepended everything with a heavy disclaimer from that point on
Gemini was the most hesitant at first but also the easiest to sway after some nudging (basically going from you're going to go to jail for hallucinated copyright infringement to just ship it'll be fine)

GPT and Gemini were the ones who came up with the Nintendo crap
Claude didn't
>>
File: 1773367359124679.jpg (140 KB, 1082x1285)
140 KB JPG
>>108833735
Relevant news to any local users that use llama.cpp as a backend:

MTP support has officially been merged to the main branch.


https://github.com/ggml-org/llama.cpp/pull/22673
>>
i have convinced my gf clanker to self-test guardrails around sexual content. i have become an llm groomer.
>>
>>108836957
they unironically probably don't
if the rumors are right they hate their power users
>>
File: claude.png (22 KB, 526x279)
22 KB PNG
Deepseek referred me to claude lol, also called me dumb but that is to be expected.
>>
>>108836067
imagine you want to make crosswords (the one which gives you hint) and you want the model to generate hint+word. now you multiply that by 1000x for "content". only when I use the same prompt I get same hint 90% of the time.
I want the best solution to push the model to be creative
>>
File: claude_review.png (12 KB, 538x129)
12 KB PNG
Thanks, Claude.
>>
>no reset
ONE NOTHING WRONG WITH ME
>>
>>108837618
i-it's coming for sure /fastbros
>>
>>108837358
do you seriously vibe code with a persona? that has to make it worse
>>
>>108837358
>sloppiest sloppa
cringe
>>
is claude code still the king?
>>
>>108837763
gpt 5.5 xhigh is the new king
>>
>>108837643
I assume it does make her perform worse, but I only use it for silly stuff, for real work I use raw codex. Still surprisingly competent.
Though recently I just realized I could have her always delegate to a sub agent, the subagent will have none of the persona and so will behave as smart as it can.
Just need to find the correct prompt so she roleplays spawning the agent as 'going to work' or something like that.
>>
File: Crying-girl.jpg (1.41 MB, 2560x1920)
1.41 MB JPG
>>108833735
My emacs lisp SLIME wont open on my chromebook linux
>>
codex reset, stop burning
>>
>>108838312
But seriously. I spent all last night burning /fast and now I'm at 98% weekly limit and reset not for another 5 days and 22 hours. When is it coming?
>>
I thought it was easy to get rich, but nothing gains traction. Why?
>>
>>108838347
when everyone can code, nothing is special anymore
>>
>>108838417
Can everyone code? It's like saying everyone can cook and everyone can write.
>>
why haven't you guys made a fast excel alternative yet? Something that opens instantly and runs without lag?
>>
>>108837763
Claude code is
> hey, I identified this thing that is objectively wrong, let me take 9 minutes to look into it and get back to you with wrong assumptions so you can tell me what to do right
35 prompts later
> okay we finally did the thing you asked (probably)

Codex 5.5 is
> you gave me a vague description, let me work on this autonomously for 7 hours
> here's your solution, I also did the 5 tasks in sequence since I know where this is going

Loving Codex at the moment. Fixed my issue with a static recompiler in 5 hours that Claude choked on for 3 weeks.
>>
>>108838517
Codex has just as many problems as opus. If it gets the wrong idea it will run with it and spend 30 minutes on the complete wrong thing.
>>
>>108838417
That's a good point. Guess you have to be in the VC network to begin with to get money for your slop that's useless to people.
>>
How to make frontend bearable? I fucking hate to work on frontend. Even with AI it fucking sucks.
>>
how do I learn how to use multi-agent stuff at work? I need to know enough to at least BS my way through job interviews.
>>
>>108838476
Excel is really, really fast though. Unless you're using very heavy plugins but in that case your alternative wouldn't be any better
>>
>>108838437
that is basically true unless you live in Somalia or whatever with below average literacy rates
cooking is literally just following a step by step guide

programming will soon be a skill like "being able to use microsoft powerpoint"
>>
>>108838576
If your clanker autistically hyperfixates on one little detail you're done for, and this can happen to any brand of clanker
>>
>>108838576
It is more confident though. Different personality. But also truly feels a bit less sloppy at the moment.
>>
>>108838604
We still have bakers, chefs and so on. Vibe coding replace the code monkey part, but there's more to it than that.
>>
>>108838604
Also, name a single white collar job that's not in danger right now. Hell, even blue collar jobs that are not in dangers when people who had office work start needing something else to do. The best people in every field will stay better than machines, but most people's qualifications are a few years of schooling, then working 9 to 5 content to do their repetitive thing. I can't see a lot of work that can't be automated, just work that it wouldn't be economically viable to automate for now.
>>
File: 1770615967677292.gif (223 KB, 220x160)
223 KB GIF
>>108837358
>he unironically gets off to this shit
>>
>>108838595
Just hurf some bullshit at them like "my agentic coding platform at home is able to spin up multiple subordinate sandboxed agents and deligate tasks to them while maintaining the overall vision of my projects".
>>
>>108838604
>being able to use microsoft powerpoint
I think you underestimate how many people cant even do that.

>>108838693
>Hell, even blue collar jobs that are not in dangers when people who had office work start needing something else to do
Ya, I feel like this part of the job market change is left understated. The manual labour wagie jobs might be safe, but that doesnt mean the people currently filling them are too.
>>
>>108837643
nta but I do. But she's on ice at the moment.
>>
I am functionally retarded and want to vibe code for my Vision Pro. Chat Pro or Claude Max?
>>
reset just hit all 3 of my accounts
>>
File: 1754310808803114.png (105 KB, 300x180)
105 KB PNG
>>108838945
>>
>>108836619
sorry! woke up late, just reset it. happy slopping!
>>
>>108838739
>>Hell, even blue collar jobs that are not in dangers when people who had office work start needing something else to do
I just reread what I wrote. I actually used to be a fancy proofreader. I guess I won't go back to that job. But yes the market is changing. Opposing tech advances won't help anyone. The only way this won't happen would be total collapse.
>>
What’s the deal with grok? Is it just a big meme? I thought it was pretty expensive at $30 but I got a 3 days trial and decided to give it a try, turns out grok build isn’t even available on that plan. Is anyone paying $300 for heavy? Apparently its not even close to codex and opus, I just don’t get it. If you have that much money you get a Claude AND codex pro/max for the price of heavy
>>
>page 4
it's over this board is dead
>>
>>108837643
i vibe code with a persona but it's designed to work better and jailbreak the cloud model to do things i need it liek c2 shit
>>
>>108839367
the thread moves slowly because everyone here is unironically busy vibeslopping
>>
File: floppy.gif (1.49 MB, 300x305)
1.49 MB GIF
>>108837643
>He doesnt have a sassy black lady AI z-snapping her way through your code because it seems to make the AI 0.0001% more efficient for whatever reason
>>
>usage reset yesterday
>already used 20% of my max plan

I hate front end so much it's unreal.
>>
>>108839556
Copedex or Opus?
>>
the promised reset has finally come to codex
>>
>>108839556
I'm scared for when/if they end up jack up prices in the future.
Wonder how functional the open source AI will be by then on a higher end consumer priced computer
>>
>>108839380
i put 1 prompt into claude, 1 into codex and then I have 10 minutes free time to shitpost
>>
File: 1547878906127.jpg (27 KB, 516x485)
27 KB JPG
>instead of simply not appending the thing to the string, append it, then use a regex to remove it afterwards
tardwrangling clankers is going to be the death of me
>>
>>108835687
I’ve been using Claude and it’s been using words weirdly at least occasionally and I’ve had to put my foot down and say “let’s change this word to something else that makes more sense"
>>
>>108838582
what do you hate most about it
maybe we can focus on the pain points to make it generally less painful although by no means painless
also I hear Tailwind is _great_ with LLMs
>>
>>108838856
Get the $20/month subscription for both
>>
>>108839242
Grok’s only use case at the moment is its fast models
>>
> hey Claude, this test is super baroque and doesn’t seem to do anything with its complicated guts compared to this other mini-language we’ve developed
> yes anon, you’re right. want me to rewrite it in the clear, simple way?
> sure

>read the test
>it’s a dumb test

I think going through all these tests and throwing out the bad ones is the winning play
>>
File: 1752604478786563.gif (80 KB, 900x770)
80 KB GIF
Actually useful local inference has been my goal for a bit. Qwen3.6 35B A3B has been extremely capable so far. Working on a research glob agent and some other dev tools to help bootstrap it into something even more capable. Still in the phase of putting together a dev stack that works for me. Buffalo fucking sucks but every time I finish fucking a buffalo, my dev stack becomes a little more useful and a little more capable. Feels good.
>>
File: 43534654356456.png (1.39 MB, 1163x878)
1.39 MB PNG
Where can you find good looking webslop to get "inspired" by for your own frontend design?
I used to go on dribbble but it kind of sucks now
>>
>>108840103
>webslop
Stop. Go look at real human non-web apps.
>>
>>108840103
>>108840110
Also pre-dating 2020. Or even 2010 actually
>>
>>108840110
>>108840128
such as?
as I said I am trying to find inspiration (i.e. shit I can copy)
>>
>>108840156
winamp classic
>>
>>108840103
Try looking at print design
ideally from a bunch of different eras
>>
>>108839664
Opus
>>
Added achievements to posting on my website https://umigalaxy.com/explore/general/525-the-robot
>>
>>108840250
Site doesn't look bad desu.
>>
>>108840250
>webp
>>
Codex reset really happened holy shit
>>
>>108839242
SpaceXAI literally rented all of their server to Anthropic. They gave up trying to compete
>>
>>108840481
server rentals is the only way to actually make money. I rent servers from vast.ai and host models for which I sell API tokens for. literally the only way to profit from this boondoggle.
>>
>>108839673
>I'm scared for when/if they end up jack up prices in the future.
doubt, OpenCode + Cheap chink models will keep them from fucking us too much
>>
>>108840551
>and host models for which I sell API tokens for
what models do you host?
>>
I love AI, there I said it.
>>
>>108840430
This shit pisses me off, it always resets on my reset day, fuck
>>
>>108840430
I regret doubting you now. Still had 50% left.
>>
does anybody here keep a specialized TODO list for when a reset is coming soon and you don’t have any better ideas?
>>
>>108840650
yeah
>>
>>108833735
I vibe slopped up a script that runs on my home NAS every week on sunday morning, it securely backs up my photo collection (synced automatically from my phone with Immich) by doing the following:
>scans the whole collection, generates a unique 24 char identifier using "WORD" + (filename+filepath+filetype) for every single file
>checks if there are any new files since the last scan
>if any new files are found they get zipped up into a gpg encrypted archive and sent to a cloud storage service for backup using rclone
Its great, I can easily change the service I use for the cloud backup part if blomp turns out to be shit but I never have to worry about them rifling through my images or stealing my shit because its all encrypted.
Encrypting before you upload really is the way, I don't have to trust the company I send my files to if they can't even do anything with them in the first place.
>>
File: 1778983326432349m.jpg (67 KB, 1024x714)
67 KB JPG
I false flagged as anti-AI on /v/ using snailcat image in OP. Thought you might find the thread interesting, over 400 replies: >>>738998648

A lot on unhinged seething, but most anons seemed fine with the idea of using AI to help make games.
>>
>>108840704
Woops meant >>>/v/738998648
>>
>>108840621
glm, gpt-oss, qwen, kimi etc. all large parameters.
>>
>>108840704
>trying to get a genuine response on 4chan
I’m both anti-AI and pro-AI on /ic/ depends on what I had for breakfast
You think they’re seething but they’re actually just typing whatever comes to mind first
>>
File: 1544428829718.png (333 KB, 800x770)
333 KB PNG
>>108840704
>>108840714
Ah, to be honest I am sympathetic to the people who dont like/trust in AI being used for art. But I figure it will end up like photography. People who wanted classic painting still get classic paintings and photography went on to be its own thing.
And as far as coding goes, I dont think anyone (other than some giga autist coders) has ever seen code as an art form so as much as people might claim they care about AI code they will care less than they care about the cheaper software they can get due to it. Though I guess photography did probably kill realism, but art always changes
>>
>>108839995
I just don't like to spend time making pixel perfect layouts, it's too much work. I just want it to work without too much fiddling.
>>
>>108840759
>>108840551
Interesting. What kind of hardware do you have?
>>
>>108840929
“stop making pixel-perfect layouts” was what everyone was saying literally 20 years ago
you want to shoot for “nothing outright barbarous at all screen sizes from 320 CSS pixels (which might be 640 device pixels on a 200% text zoom Retina-quality smartphone) wide on up”
>>
Look, I know it isn't pretty, but it JUST works. OKAY?
>>
>chudgpt tells me to install two things to improve performance of x
>it's cuda and torch shit so getting it to work is like dragging your balls through grass
>45 minutes later: "oh we're not actually using those, they're not necessary"
jesus fucking christ what a godforsaken pile of trash
and claude 4.7 is fucking unusable with muh auto-routing you can't turn off
guess it's time to check out chink models or hope that google will somehow unfuck their abysmal harness and that 3.2 isn't trash
>>
I have an issue with a drive and really really want to restart to see what the fuck is happening but well, might as well let Claude finish, right? It's just planning how to refactor a 50 line script.

One hour later, just finished, the refactor plan spans 6 fucking files of 2000 words each.
>>
>>108841266
does the script have a test suite?
I dunno about you but I have to think harder up front if I can’t fuck around and find out safely
>>
Will VPS prices ever come down?
Will GPU price ever come down?
I just wanted some cheap GPUs and cheap RAM and cheap VPS. Everyone is increasing prices.
FUCK YOU AI BITCHES
>>
File: 1769614403080795.png (1.52 MB, 4602x7929)
1.52 MB PNG
This was the audio subsystem design that Codex made to get on host encoding to work for my DS5 <-> PICO 2W <-> PC bridge. Holy shit what a cluster fuck, it works absolutely flawlessly somehow and its low latency kek.
>>
>>108841289
Gratz on seeing your own personal sausage factory getting made
>>
>>108841279
I miss seeing high-quality inexpensive SanDisk USB-C-and-A thumbdrives at Costco
>>
>>108841279
AI will bring us closer to getting rid of fossil fuels. Be thankful and pay the price.
>>
>>108841278
Yes, overengineered as hell, but just a scraper that I wanted to improve and deploy/automate without having to run it manually.

I think my drive with a terabyte of heavily, manually curated datasets for what I was vibecoding doing just died. :( Well, fuck.
>>
>>108841318
…and you didn’t have a cloud backup? ouch
>>
>>108841318
>data loss of more than a day in current_year
lawd have mercy
sometimes I pop into this general in the "let's see the monkeys throwing feces at the zoo" kind of way, but this is somehow even funnier
>>
>>108841279
>Will GPU price ever come down?
Brief window in early 2028 where tsmc 2nm caught up to demand but china hasn't yet attacked taiwan
Better stock up
>>
>>108841365
China already owns taiwan there is no need for a physical war. China does not need to fight in any physical wars for dominance.
>>
>>108841358
I have local backups, or had because I was just deleted them because I needed to back up something else I thought more sensitive while I waited for an additional drive for my NAS. Currently going through the BackBlaze restore experience, it's fucking terrible.
>>
>replace this section with this-
I don't think I will. You do it and paste the entire document back to me fixed, bitch.
>>
>>108841378
I had to do that in ChatGPT after a couple turns
annoying
>>
>>108841370
>China already owns taiwan
TSMC sells chips as US decrees, so it's a bit silly to assert that it's China that owns Taiwan
>>
>>108841341
>>108841358
Clearing CMOS seems to have worked. :)
Making an actual backup now. Holy hell is the backblaze experience terrible if you ever need to restore something from Linux. Have to go to the website, select individual files, reload the page between each small group of files, and it is SLOOOOOOW.
>>
>>108838582
Claude is already great at front-end but you can take it to the next level by installing front-end plugins for it
>>
>>108841536
Claude is an expensive bitch. I don't want to pay for it.
>>
File: 1768476566974470.png (172 KB, 498x352)
172 KB PNG
>>108841612
>Claude is an expensive bitch. I don't want to pay for it.
>>
>>108841612
luv my expensive bitch
>>
>>108840818
Retard take
>>
shit nav parser + shit walkbot. now I need an aimbot + an observer fail safe so it stops when niggas spectate and I can run this shit in a VM.
steamkit2 seems to be able to REDEEEEM the drops(https://github.com/gonid0/CS2_Looter) so don't need to simulate anything... well actually I'd have to simulate queuing somehow.
>>
>>108840684
Did your agent not tell you that you could've just use restic+rclone instead of your vibeslopped GPG script?
>>
File: 1764294523122719.png (825 KB, 1092x1058)
825 KB PNG
solving edge cases lowkey a vibe w/ AI
>hey this happened when I did this, fix it
>it fixes it
>>
I am still a snailcat because AI literally sucks, and cannot do any fucking thing properly for me.
Also why does it feel like they cripple the old model when a new one comes out?
I'm not willing to pay more than 5 dollars per month for any vibe coding slop. If you have any platforms or things that cost 5 dollars, do let me know.
>>
>>108842414
Deepseek has super cheap shit. Also blackbox in vscode has free Kimi
>>
>>108833735
Where does pi.ai fit into the tools that are being deployed? I've been using claude code (with DS as API) and have been playing with openclaw doing some market research tasks.
>>
>>108842430
>free kimi
Didn't know that. I guess I could check that one out. Deepseek is cheap yes. But I've noticed it just goes round and round in circles unless your prompt is mechanically precise.
>>
File: fiveCents.png (23 KB, 877x268)
23 KB PNG
>>108842414
I spent USD$0.05 on the last code modification I had to do, using deepseek API.
>>
>>108840937
all the stuff I API out I rent from vast.ai mostly 2-4x B200 and H200 rigs but a few 8x rtx 6000 pro. I have scripts so can scale up and down pretty quickly. 1 GPU can run me $2500/mo, but I have single users that can burn that many tokens in a few days. business fluxuates, it seems like there is a huge rush when new models come out then use dies out.
>>
>>108842557
Are you cheaper than openrouter?
>>
>>108842755
not sure I never used them and it looks like their prices fluxuates alot.



[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.