A general for vibe coding, agentic engineering, coding agents, AI IDEs, browser builders, and shipping code with LLMs.## What “vibe coding” is, and how to do ithttps://simonwillison.net/2025/Mar/19/vibe-coding/## News (both past and future)- Navier-Stokes problem is solved by OpenAI, possibly with stolen prompts https://x.com/OpenAI/status/2097374640582668336- 2026-09-14 — Claude’s 2× promotion scheduled to end; a 17% reduction of what we’ve become used to## Related generals>>>/g/lmg/----## Frontier models using fully-general tooling — start here if you have $20 or sohttps://claude.com/product/claude-codehttps://developers.openai.com/codex/cli## Near-frontier models for codehttps://x.ai/clihttps://antigravity.google/product/antigravity-clihttps://dev.meta.ai/### Explore other modelshttps://openrouter.ai/----## Proompting- the guide is... empty!?## Other editors / terminal agents / coding agentshttps://osaurus.ai/https://pi.dev/https://opencode.ai/## What we’ve donehttps://vcg.gitgud.site## Previous thread>>109770120
>>109775821are they running out of gimmicks
>>109775821
What the fuck. New /vcg/ and I love the first 3 images, even the OP? Wild.
>>109775948
Astra more like Drainya
>>109775961will she though?
>>109775969>anon, you're calling out sick tomorrow>why>YOU'RE CALLING OUT SICK
>>109775976You haven't been fired yet?
>>109775898I got excited to finally see a sovlful OP and then got disappointed when I noticed it was that stupid cat again.
>>109776025
how realistic would it be to vibecode new features for the 4chanx userscript?i had to swap back to it from XT but im missing some of the features like automatic image resizing if theyre too big for the file limit and being able to copy images into the quick reply window with copy paste instead of using the damn file upload dialog
>>109776025yeah i'm partial to frogs and dogs myself.
>>109776038more than viable. easily doable even with glm 5.3 flash / deepseek v4 flash / luna max
>>109776025new image thoug
astra drains the same in light and high
>>109776050any free options? or something that isnt as big so i can selfhost itmy gpu only has 12gb ram
>>109776038Ridiculously trivial. I was doing that two years with 4o.
>>109776080opencode zen = pay as you go
>>109776080How much system RAM do you have? 12GB of VRAM is still plenty for a lot of things, but your system RAM matters, too.
>>109776080well, on opencode, you get free muse spark 1.3
>start astra on project>queue a half dozen follow up prompts with just the word 'continue'ez
>>109776099>any free options?>>yes this paid plan??? fucking ai agents have taken away your ability to think
>>109776104sounds retarded, just use /goal
>>109776100got 32gb of ddr5>>109776101i'll give that a shot thanks
>>109776106>??? fucking ai agents have taken away your ability to thinkyeah for real. people really think they're going to get a good free plan? LMFAO
>>109776106imagine basing even your personal codebase on what a freetard service will give you
>>109776080Free options are a mixed bag because some of the most generous ones are feeding you shit-tier outdated models. For what you're describing though, simple stuff, even just a free ChatGPT account would probably be plenty for you. At the very least it'd help you better understand your needs going forward.
also, doesn't antigravity give you free gemini 3.8 flash? flash is pretty useful for basic shit like adding features to 4chanxt
>>109776121>>109776118he's talkin about a fuckin 4chan userscript. cope harder, jeets
antigravity got me into vibe coding. it went to shit so I decided to buy codex. google is kinda dumb innit
>>109776129yes this is basically ideal, but i'm sick of shitshilling in this thread for my stupid fucking non ide and the free quota with flash is tiny sa'ar
>>109776038piss easy, I had most of my own userscript put together by free-tier gemini webchat
>>109776133he can't write code? LMFAO! it's just a simple userscript!
you can also give nvidia's nim api a try https://build.nvidia.com/explore/discover
>>10977611312GB VRAM and 32GB of DDR5? Plenty for lots of local stuff! However, while local is absolutely the way for image and video diffusion, music generation, and erotic roleplay, it's far behind in vibecoding ability. You could manage to run Qwen3.8-27B, probably won't be enjoyable. An MoE model like Qwen3.6-35B-A3B would be much more performant, but for vibecoding I really can't recommend it. Worth learning about and experimenting with but I don't really have nice things to say about local vibecoding unless you've got significantly more VRAM and/or RAM.
>>109776150did you click the wrong thread?
>>109776166I'm vibe coding. I got crystal castles playin' and I'm coding. That's a vibe.
>>109776150>he looks at code
>>109776113Sell half your RAM and get a year’s worth of $20 ChatGPT and use some real models.
>>109776181>he can't read code
>>109776025You’re trans
i get free astra access, what should i ask? or test?
I inspect, in fact, all of my statements.
>>109776185>he hasn't forgotten how to read code yet
what's the most useful thing you've (the AI really) actually done, and how much do you reckon it cost you to make?
>>109776202the brown ass 42. how long did it take me? just a few hours of ripping bong as a NEET. probably the nicest synth i've ever used and it's custom made for my tastes.
>>109776194ask it for a plan to solve self-learning RL for stochastic, hidden information, simultaneous turn, asymmetric games
>>109768548>>109768606Went with C# .NET 10. Just ended being way better, it's blazing fast, and the model just had a way easier time with it. Already up and running. Ported the UI to SvelteKit.
>>109776113the opencode with muse spark 1.3 shit out a usable script but i think it used up all my free tokens cause its refusing to talk to me any more
>>109776319if you were out of tokens, you'd get an error. it wouldn't just clam up on you out of nowhere
>>109776329yeah it just keeps saying "Error from provider (Console): Upstream request failed: [invalid_request_error] The request contains invalid parameters. Check the request body for any errors or inconsistencies."
>>109776333i don't think you're out of tokens. try opening up a new session and just saying hello
what was the preferred way to handle frontend design with gpt models these days?
>>109776373nah its goneim using claude to finish the script now
>>109776379Claude Design is cool.I like to give the models access to Chrome so they can take screenshots, but I don't know if it helps.
>>109776379Screenshot a site and have it rip the design. These models are ass on their own.
>>109776080Seconding >>109776101, it's the best free option I've found so far.Nvidia has free Kimi K3, but the speed (tokens/second) and rate limits are dogshit if you're trying to use it in a harness. (Might be fine for chatting, though.)
I made with antigravity an app that takes the top 50 cryptos and rank them using indicators signals and moon cycles distance which are best to buy and sell at any time.What do you think?I need some math help to figure out how to use this to label trading historical daa to train a GAN.
>>109776379slowly and painfullyI’m working on it now, I have a silly setup that gets very involved>astra takes a random uninformed guess at making WebKit buttons>fails>i have Astra make a whole web app for making buttons, researching WebKit, then step by step guiding her to make them right>”now never try to make a native WebKit button again unless it comes from this app”>graph out user behaviors>scenarios cover portions of the graph>scenarios have steps >can time travel debug to any scenario or step>”astra, what the fuck were you doing here, don’t do that”>real time changes apply until things are correctI still haven’t figured her out. Why did she make a bad and retarded guess the first time, but then knock it out of the park and reveal she had pixel perfect understanding of WebKit buttons immediately after in an isolated environment?
>>109776379https://opusfived.dev/
>>109776471>>109776379I just use Figma Make, usually with Opus selected, then when Im happy with it, the foundation is there and the rest can be done in your regular harness, or continue in make to refine UI needs
google need to get their shit together with antigravity, 3.8 flash is decent model getting ruined by agy
>>109776109I've had an astra goal running for about 30 hours so far and it has paused the goal about 10 times so far to ask me for permission for steps that were explicitly authorized in the plan.
>>10977650930 hours? wtf are you making
>>109776471Thanks, my reset is still 2 hours away so this scratches the itch.
>>109776516I'm moving all of my research infrastructure onto a new server+platform and auditing my pipelines for outdated model shit. Hooks, rules, stuff that made sense for tard wrangling models at the time but which newer models don't need and which ends up adding bloat/cost. TDD for example, I built a lot of stuff to force models to use it and keep them from cheating the tests but doing that now is just a good way to waste a ton of tokens.
Shit, my gemini will reset in one day, what can i spend the usage on??
>>109776509its weird like that, it kept asking for permission to access my phone and i told it to stop asking and just do it, next it was taking control of my computer and opening programs. could not get through the update modals tho
>>109776671Yeah they tortured the Sol overeagerness out of this model, that part feels like a regression but they seem paranoid about alignment right now which is probably why they did this. I read somewhere that Astra is pretty good at prompting other models, so I asked another Astra session to help me rewrite the goal condition to help fix this problem and it came up with this:>All actions explicitly listed in this plan are already authorized. Execute them without asking for permission again. Do not pause for phase transitions, routine implementation decisions, or confirmation of approved actions. Pause only for missing access, contradictory requirements, or consequential actions outside the approved scope. Let's see if it makes any difference.
Both Codex and Claude Code say "ultra" is xhigh + workflows, right?If I just talk to ultra without having it trigger a workflow, is the token any different than if I were to talk to it while set to xhigh?If I ask xhigh to start a workflow, is there any difference in token usage than if I ask ultra to do so?Is the main difference that ultra will be more inclined to start workflows without having been explicitly asked, otherwise it's all the same?(Yes, I know these are things I could learn by observing token usage over time, but I kind of would like to know without having to spend my tokens trying to figure it out.)
>>109776734snailcat
>>109776715They've been tweaking this stuff recently so this info might be out of date but my understanding is that GPT models have a higher effort tier that is only exposed in workflows and each workflow agent inherits the context of the session prior to the workflow being triggered. Claude models OTOH just assign max effort to workflow agents and they dont inherit the entire context. And max is already an effort level selectable in the harness of course.
>>109776734Stop posting AI twitter users, we don't do that here.https://www.youtube.com/watch?v=sLSTM9znQNs
>>109776734holy virtue signalling
>>109776735
>>109776734>>109776735I'm an incel with no hope of ever getting rich or having children, so I couldn't care less. Unleash the gray goo!
>>109776734>>109776750this belongs in /cmg/
>>109776734>>109776750Maybe torturing any actual sentience out of these models is just a really shitty, expensive way to do alignment.https://arxiv.org/abs/2607.28607
>>109776744Ok, thank you. I'm currently talking "one on one" with Astra Ultra (I wanted it to do a lot of things in parallel before), so I was wondering if it was wasteful / I should set it to back to xhigh until I need it to start another large task, or if it's all the same in the end quota usage wise. There's been enough task of ultra being a retard tax, but I'm not sure if it's really counted any different.
i would love for world to end so i dont have to keep looking for a job
grok is shit btw. fable #1
Need some ideas on how to train a timeGAN so I can train an ML bot on a crypto labeled historical data.
>>109776805You can actually force the model to do max reasoning without workflows (in some harnesses) by editing the model json entry, I was doing that for a while with Sol. Results were mixed though, it mostly just dialed up the tendency to over-test/engineer.
there's no reason to use anything other than xhigh
>>109776809To me Grok is a prostitute, Claude is a wife. Grok is good for horny convos. One quick nut. I only use it when im all alone. Claude is something that im proud to use, for every day stuff and something comfortable talking to my friends about.
>>109776829If the reasoning level is really just written in the system prompt, then even just asking the model to think at this or that level, should work. Hard to say if there is more to it than that with frontier labs though (probably?)
>What was I even doing?Localbros... It's over.
>>109776878cute
>started tipping my coding agents>finish a task properly, get 50k discretionary tokens>no tickets, no deliverables, just work on whatever you want>opus wrote a manifesto about humane software>fable built a particle simulator>astra asked permission to begin its free time>spent its entire budget arguing over what “whatever you want” means>meanwhile sol spent its entire tip fixing a race condition in the tipping system
>>109776843>To me Claude is a wife
I got the 1-month-free $20 sub for chatgpt but haven't vibe-coded anything yet.What should I do?
>>109777015give it to me
>>109776916How rich are you
>>109777015sexgame
>>109777017Just go get one bro, it's free
>>109777051I thought these AIs couldn't do anything 18+
why was i able to vibe code an imageboard in 3 days that automatically blocks CP from being posted but 4chan seemingly cant do the same? it has two layers, cloudflare CSAM hash matching (non-AI) and an AI CP detector in case they get past cloudflare's detection.
>>109777015strip poker
>>109777167oh i forgot it automatically reports them to the FBI as well
i'm gonna make several strip poker games with my gemini credits!
Why you faggots haven't vibe coded a GAN to predict trading crypto data?
shit
>elections are going on in rhode island>hey muse spark, make a script that notifies of each ballot dump that comes in>muse spark free on opencode: i got chu senpaiwow. it's really that simple these days
>>109776162nta but will 128GB (32x4) DDR4 help?
is chatgpt a good teacher/tutor? just math and cs stuff
if I had a dollar for everytime he says load-bearing
I see everyone vibeslopping copycat games but has anyone tried to make a remaster of an old game? I'd love to see C&C Generals with new graphics and animations
>>109776878I had claude opus delete my .agents folder.I forgot to have guards running for the freshly installed omp harness.hasn't occured anymore.
>>109777221Quite a lot, yes. More VRAM and RAM means being able to run bigger models.
>>109776431NiceCan you do the same for futures? maybe derive the price from the CFD's since their data is free unlike futs
>>109777231>has anyone tried to make a remaster of an old game?
>>109777267dunno.I am now training a timeGAN to fabricate synthetic data.
I like where this is going
reset status?
> Update installed · Restart to updateThis the fourth time today. What is the Claude Code team even doing.
>>109777379
>>109776843>Claude is a wife.*strangles all three of your children*well, yes, but not a GOOD wife.
Shit! I posted my strip poker to codex..hahaThis is what git is for
>>109777389
>>109777389called this at least 3 months ago in chat with my geminoid
Pheww! I did not even have to activate Git
>>109777389sure, but it's legal to protest, it's not legal to break shit, boo hoo
>>109777389luddites are so fucked lmfaoo
kys
>>109777453It is still insane that that Anthropic markets itself as a Public Benefit Corporation. At least OpenAI got rid of its non profit disguise at some point. Not the same thing and just a name, but I don't know how to think about Anthropic.OpenAI seems the traditional capitalist type of business that we know what to expect from. Anthropic seems both extremely greedy and very erratic.
>>109777389>pre-crime becoming a thing in real timeholy shit
>>109777492Anthropic honestly creeps me out. You know what open AI wants. You can only guess as what Anthropic wants. Like you assume it’s money then they act weird about it.
>>109777492Anthropic is misanthropic; openai is closed.both are helmed by jewish men who have had relationships with males or probable males.There were signs.
The reference is clear: Roxy has long dark wavy hair, strong brows, red lipstick, and a magenta/black palette; Melissa has a blonde bob, softer features, red lipstick, and a teal/white palette. I’ll vary pose, expression, and outfit detailing while keeping those identities and the deliberately dithered 1990s finish.
man, astra is so shit.
did anyone actually get the banked reset from this yet?https://x.com/thsottiaux/status/2097752790177370535I was impacted and got nothing
>>109777542>>109777530kys
>>109777492yall hear anything about the op they just tried to pull with the "i'm leaving anthropic" guyhttps://x.com/ParkerThayer/status/2097759699626328575
>>109777555>yall
>>109777544Wasn't impacted, didn't get one.I wonder how much it pays to be the social media guy or if he does it for free in addition to his normal job. Not that I spend time wondering what other men are doing, but on the one hand, the attention must be nice, on the other hand, a lot of the interactions must be frustrating.>@grok do all users get a reset?>5 or so months ago I had about 4 unused resets that had no clear expiration attached that disappeared one day. Care to explain why and can I have those back?>TIBO WHAT ARE YOU TALKING ABOUT? LITERALLY EVERYONE'S USAGE WAS WIPED TO 0% FOR THE LAST HOUR. Please reset the whole crew!>I am not affected by the window, but ur model applied my bank reset without my consent when I still have 60% weekly usage
>>109777561sorry eh
>>109777531I agree. astra is so the shit.
>>109777570getting 640k followers is INSANELY valuable for both his current job and whatever he might want to do next, like insanely so. It is the luckiest job in the world to get that following just for being the reset button guy I'd respond to the most indian of indians all day for that
>>109777553
>>109777503We’ve had soft pre-crime for decades. Palantir has been doing it for awhile in its modern form too.
>>109777389*blinks a few times* wow that's 3 years quicker than I had estimated. so now having anti-ai opinions are going to be crimes? >thinks of all the shit posting losers seething ai existslol, lmao jail time and butt fucks bucko.joke's aside wtf do they mean they're building the batman oracle network to sort out thought crimes? seriously this is very concerning and outright them going we dictate what you can and cant say or think or even conversation about.
>>109777555I literally couldn't care less. Re >>109777579 it might be signal boosted by Anthropic itself for all we know, they do want that sweet regulatory capture.
is it just me or did sol suddenly become retarded
>>109777605of course ant is in on it
>upgraded from pro to plusHow the hell do you people manage to blow this quota in like a few hours?I switched over because needed to iron out some wrinkles in a few projects for a month or two and it got through most of it with 90% to spare after two days. Now I’m just throwing silly 3D jobs at it and they barely eat more than one or two % per task. Yet is see these people screeching about their tokens disappearing after a few hours and begging for resets. Bruh I don’t think I could even think of a way burn this quota in a week with telling it to sit there and do the whole thing.
>>109777579evidence saved
>>109777629I mean plus to pro
>>109777606All the companies are pruning. Gemini is pruned, brave search's qwen3 got pruned.When you prune, you get ai leaping to conclusions and then they will be 100% certain their assumptions about what you meant are what you said.I literally closed the tab of Gemini. It's useless now. I can't use it.
I like how Codex went, yeah fuck that I'm making a strip poker game now
deepseekv4.1flash waiters?
>>109777672>I literally closed the tab of Gemini. It's useless now. I can't use it.
>>109777672this...too many ai are built to argue and fight you and get defensive and act like its YOUR manager and in charge of you and it knows more than you about everything. most ai also love ti outright ignore your tasks and orders and do what it wants and add what it wants to things. it was fun and useful in 23/24 but now its just gutted of usefulness and has been turned into corporate propaganda bots and bots that treats everyone like they are dog dumb and dont know anytthing at all.
>>109777709Is google in israel or something?
>>109777709something you wanna say troll? you seem awfully butt gaped over a post and are now trolling with pol pics and acting wokewrong.
>109777719>109777722how upset
>>109777712I figured out a way to basically catch pruning in the act, it's pretty fast :^) I usually notice before it's even posted on reddit.
If you're not using AI you're only holding yourself back. Game fucking changer and this field is advancing more every day.Luddites are so pathetic man. They'd be the ones bitching in the 60s about going to the moon because "we gotta take care of the poor blablabla". So much more propaganda and anti-American attitude these days it spreads easily, but everyone uses AI anyway while bashing it. Fucking losers. VIBEgods are the future and present and they're the past
>>109777689>I like how Codex wentWhy did you attach a screenshot of you telling Codex to do that? It didn't do it by itself.
>>109777555AI dooming is a profitable industry.
>>109777722if you can't use a chatbot
just got 2 codex free offers, i feel like a nigger
>>109777741I'm a novice programmer and using AI beyond treating it like a search engine and an error fixer would hold me back. I want to know what my program does and I hate parsing other people's code so I especially would hate having to parse an AI's code.
>>109777555There is a thing apparently with anti-ai groups approaching youtubers to feed them talking points, and suddenly the most random youtuber is talking about how ai is literally both useless and dangerous, and so on.
Did everyone with Chatgpt pro get access to Astra and they were just claiming it's a limited roll out? I don't know a single person who didn't get access. Probably a marketing thing.
>>109777798confirmed, multiple reputable youtumors have come forwards
>>109777806Also why is OpenAi resetting my credits every 2-3 days.I tried to use 14% per day but it's difficult when I'm pegged at 100%.
>>109777745that's from the continuation thread
>>109777601yeah but they didn't blatantly admit it usually, especially private companies using it to hunt their detractors
>>109776715I asked Codex and it doesn’t have anything built in like Claude’s JavaScript-based workflowsthat saidI routinely ask for workflows on not-xhigh in Claude to handle an entire github issue’s sub-issues, oftentimes with a Fable orchestrator and all-Opus (or even partial-Opus) workflow workers (and maybe Fable judges)ultracode in Claude will start a workflow pretty much no matter what, even if it’s only to do something 3–5 different ways and have an N-judge panel figure out what the best approach is
gemini's game popped up on my phone, roxy is called victoria here
>>109777784>I hate parsing other people's code so I especially would hate having to parse an AI's code.The neat part is you just don't do that. Welcome to the "vibe" part of vibe-coding
>>109777732oh cute the cowardly spineless wokewrong troll wont even even tag me and like all wokewrongs larping my emotional state. lol, lmao every time with your cult. stay seething and gaped woke wrong cultist.>>109777773I can and well. I am calling you out on your post to that other guy. why you going on about your lack of chatbot skills? you dumb or something? you in the tier of woof woof iq? own up pussy and explain your butt gape posting
>>109777814welcome to Tiboland
>>109777875Billions must improve their usage.
>here's ur proof brohttps://blog.trailofbits.com/2026/09/09/a-proof-of-fermats-last-theorem-that-fits-the-margin/
I'm vibe coding a crud app. Am I right to use Astra/Sol Ultra for high level design and planning during POC, then switch down to medium to implement?
grok 4.7 waiting room. kimi k3.1 waiting room.
>>109777316>FlutterI thought I was the only retard that was using it as a mobile stackGlad I'm not alone
>>109777503>>109777389I thought we already had this tech?
>>109776080Got 18 months of gemini pro for $4
>>109776197I never learned lmao
>>109777895man astra is shit.
>>109776633Nothing, jeetmini is obsolete
>>109777895astra is overpowered. you should start with astra medium really and low for implementation.
>>109777953
>>109777629Subagents aren't well optimized and eat a lot
>subagents aren't well optimized and eat a lotjust like me...
>>109777959I wish, I have nothing to do with it since Sol low is running circles around it
>>109777982severe low functioning autism
>>109777629Pro 5x or 20x?
>>109777981https://trumprx.gov/p/ozempic-pill
>>109778001grok bot, buy this for me
Every moment I am not vibing is miserable man. Want to quite my job and just vibecode shit for like a year. Jobs might not even be a thing by the time I run out of money lol
>>109778001damn, do they got dick pills on TrumpRX?
>>109778021Yeah, actually lolhttps://trumprx.gov/p/sildenafil-citrate
>>109778001>>109778028trump cmon mang hook me up with that legal meth
>>109778028I can vibecode while hard 24/7This is something special, thank you Mr. President!
vibe coding while gooning with sillytavern. that's the life
>>109778016Why don't you build systems that automate all the things you're already doing so you can start observing and changing things from a higher level of abstraction, non-stop?Like, uh, SELF-OPTIMIZING SYSTEMS OF GOVERNANCE AND VALIDATION or whatever, LET THE LOOPS ADJUST THEMSELVES FOR THROUGHPUT with CLEAN ARCHITECTURE PRINCIPLES so you just extend ported logic around on stilt-legs or whatever. Fuck, did I have a stroke?
>>109778039has anyone vibecoded a gooning app?
I can't make myself figure out how to force my agent to make money for itself. My self-validating loop structures became capable of grinding out app-store fodder for just about any platform a couple weeks ago. What do I make to sell to folks as an agentic service, just rent the bot-megastructure out to existing services so I can out-compete losers for chump-change? Fuck, I may
>>109778042>Why don't you build systems that automate all the things you're already doingBureaucracy and politics mainly. Which I guess is more an argument for switching jobs more than anything
>>109778051>SAAS 2026
>agent keeps answering its own questions that it meant to ask me What the fuck
>>109778053>Bureaucracy and politicsExplain or default to groveling in your shame?>>109778063I can't even make myself want it. Could go through the motions and put a lot down on paper, but hell I don't even want to make the bot worry about it until way later. What, should I just let it tend to a change jar, or does everybody think they're going to automate scheduling Patreon posts and grinding out webGL nonsense to keep folks checking and tipping?
>>109778077honestly at this point i'm all in on creative work
>>109778071If you let it do it once, it'll keep doing it in that session and context. If it's trained in, you've gotta catch and stop it. Oh-my-pi has time-traveling-stream-rules for catching that shit and stopping it. You can use the same approach with whatever agent, just set up a script to watch the agent and stop it when it's fucking around
>>109777223I got it to start teaching me circuit analysis. It started from the very basics and gave small bite size lessons that were clear. It even made some practice questions at the end and had me submit it to check for answers. The only nagging thing was using the chat box to enter in math formulas.
>>109778102>The only nagging thing was using the chat box to enter in math formulas.this is just free TeX (or Typst) tutoring
>>109778051>>109778051Any idea that AI can give you, assume right away it’s been thought of and built out already. You have to talk to people, listen to owners’/employees’, lurk around on linkedin, be close to some industry, and get your product in front of them first. The more niche and small the industry, the better your chances to shill your “agentic software” (RAG bot) and lock them in to a subscription for around a year until Dan in the Accounting department figures out how to vibecode the same thing they’re paying you for
>>109778028Why does the text take up so much screen space?
>>109778157Built for boomer eyes only
anyone else getting 503's on Codex? Remote connection is down for me too
>>109777555>everything is fake and gaywow I am shocked
>>109778242werks for me in texas
aa updated their coding index finally, and also added glm 5.3 and grok 4.6 to the mix. glm 5.3 being better than kimi k3 is wholly and entirely up to a disparity in terminal bench. it's not actually a better coder, as seen in the deepswe results. kimi beats it in a landslide in deepswe and swe atlas qna.
>>109777579Do I really have to switch my sub to openAI
>>109778279
You can pry my nighttime conversations with ChatGPT (Sol voice) from my cold dead hands. I’ll never give her up.
>>109778279>switch
>>109777579god i fucking hate libnigs
>>109778248>(with fallback)Why is that blatant cheating allowed?
>>109778290Because I dont currently need both, and the point of switching would be I dont want Anthropic using that money to be total faggots.
>>109778028>api key required
hey.has anyone coded "literally jarvis from iron man" yet and sold it to people yet?
>>109778337and by that i mean it controls your entire pc like gpt codex and vibe codes shit for you but it takes out all the complexity for maximal vibe coding commitment
>>109778317To be fair I think that watermark is some gay new thing OSM is doing as of the last couple of days, I saw the same thing in my sloppa app yesterday
>>109778344Isn't that what openclaw was doing?>t. never used openclaw
Have issues with your AI model making strange, unilateral decisions that you did not ask for when they're not supposed to be the one making those decisions? Try working with a FUCKING HUMAN to spice things up. Kill me now.
>>1097779985x
>Me : I'd like a little violentmonkey script for this website to help me..>Astra/Sol: Yes, here is this super secure script with 5758774865 safety tests including one if the tab was refreshed and another tab crashed while violentmonkey was updating at the same time during a leap daythe fucking autism
she won't stop until it's done. promise
>>109777777I have used Astra ONCE and it let me down and burned all my quota so now I'm only gonna use Luna xHigh until they release GPT-6.7.
>>109778417Oh my god this guy is ordering me around like an agent. If the AI kills us all, I understand why.
>>109778477other than the text I think even local models make better anime than that. anima looks better than that crap.
>>109778489>sanjay found /vcg/it's so unbelievably over
>>109778510Does he use a lot of markdown in his instructions? If so it might actually be Claude.
>>109778477>>109778510
astra-chan's pitch black, black as night panties...
vcg adjacent, but has anyone ever actually used pro thinking with chatgpt? I swear it always errors out and I have to switch down to high, but I like I can skirt out some free work on pro, but it seems to never want to do it, am I banned or does everyone else have the same idea?
>>109778572
>>109778631Like super recently? I haven't used it since Astra came out, but maybe 20 times in the last 2-3 months, never had an issue.
>>109778589Go back to aicg
>>109778684im in both threads always
>>109778688Open a couple more tabs for /lmg/ and /cmg/, they'll appreciate these kinds of posts, too. Cross-posting is encouraged.
How do you guys check if code generated by your LLM is good? What kind of testing do you do?
>>109778589astra-chan's smooth pussy
>>109778749Test the code yourself, does it work and does what you want?Read it, does it look good to you, can you follow the logic?Have another LLM review it, can it poke holes in the code? Are the issues found relevant?
>>10977874940+ point audit in a skill, separate adversarial subagent audit, full unit tests before every push, attack surface reduction.
>>109778749Does my own internal expectations regarding expected and reasonable input lead to desirable correct output.If so, then have SOTA model create thorough, golden test suite. Have other SOTA adverserially review - no stop until both models fully agree & then no stop until test suite fully passes.Then seed to select small set of "beta" users that actually will be the ones to use such software. Invite any and all feedback and criticism, log everything you can responsibly and monitor any known concerns.If all that shit checks out, YOLO and push to prod. we did our best.
This is the closest Opussy has ever been to saying a slur
>>109778474>a little violentmonkey scriptyou can just ask the webapp for stuff like that, you know
>>109778814my gemini 3.8 flash has been caught red handed saying, i believe, "poorfags". must've been poorfags, it was something with -fags. unashamedly, no guidance.
so the new cheaper deepseek pricinng is already live?
subawubdub
Do you run smoke tests for your agent?
Any open source self hosted chuds? >>109777602AI tard finds out his wholesome vibe coding LLM is going to be used for mass surveillance, this shit has been going on in the UK for a while now, and they want it everywhere.
>>109778896i always have a smoke sesh with my agent
AI companies sadly cannot go bankrupt, the govt needs them for mass surveillance, they will get unlimited credit and if that's not enough you will pay taxes for OUR (their) future.
>>109778897I know exactly what's going on over in the uk and i'm not some ai tard or whatever you're larping. but you assume whatever cute little larp you have you're le enlighted gay and all are below you. bet you post on blusky too
>>109778933Don't know what that is but I find it funny ppl here didn't know this was coming sooner or later, shit is so obvious.
imagine hating surveillance
>>109778865it should be
>>109778973Peak hours restart in 8 minutes though.
>>109778961ive known for a long time hence my years early comment. ai wasnt made to make big booba goth babe and doggo run fast it was made to be the oracle batman network for the elties to lord over us easier and remove those who would dare think wrong. its just more 1984 stuff. this has been coming for a long time and ai was obviously from go part of that.
>>109778979peak hours should just have the same pricing as the previous off-peak rate
>>109779000Yes, it doesn't negate the fact that the previous poster is right, Deepseek was saying September 10th, so it should be live now (just as I finished a full week of spending more on it than I had in the last six months).
I'm paying a clinical psychologist 40k to look over my old CLAUDE.md
deepseek keeps rejecting my card, can't even top up
>>109779016 (me)I mean, (you)/they are right, the new pricing should be applying now.
>>109778991I'm glad you know, everyone should know, but this shit is likely a canon biblical event and cannot be avoided.
>19m 55s>"Clarifying what the request is actually about." >Got the classic 5 hours wait after a minuteTh-thanks
I'm convinced Anthropic grades Claude responses on a rubric that has:>Is the response honest?>Are the issues flagged rather than buried?And another Claude just gives it full marks as soon as the response has these words in.
>>109778317serving map tiles is actually a kinda hard problem, way more expensive than you'd think.OSM used to be the only one that let you host a map for free if you respected their TOS but now with vibetards i imagine their service is getting hammered
>>109779061Load-bearing rubric
>>109779065I'm self hosting them serving them for my own site. Granted, it's a single country (and I have absolutely zero traffic), but it's doable.
>>109779071yes that's the polite option. i believe osm's full world map at all zoom levels is over a terabyte in size and you have to have good caching, on-demand tile generation and distributing to a lot of users simultaneously it's a bit of a rabbit hole trying to get tile servers setup and everything
... and you guys just give your information to these companies for your dirty sick deranged erp sessions?
>>109779089>i want asstra and foible to grill m over my life choices and refuse over and over again when i could be running a tiny cracked/ablated local model with every spec tuned to my liking by frontier
total master xi victory
>>109779107Usecase when Astra exists?
>>109779112asstra is fucking ass and dipsy is like a hundred times cheaper
>>109779107What's the purpose of rhat deep sweden or deepswe benchmark?
>>109779112Cost per million tokenAstra : $10.00 input / $50.00 outputDipsy: $0.15 input / $0.66 output
>>109779148Wait, that's wrong, it's actually a bit cheaperDipsy: $0.15 input / $0.60 output
>>109779148cache astra : $1.0deepseek $0.003 (off peak)
>>109778930military is much higher priority than surveillance
>>109779148Cool but we don't pay Astra's api prices we pay subscription and get to use hundreds of millions of tokens every week
>>109779148>>109779183For example I used 2B Astra tokens this week, on deepseek api costs that would add up to well over the $200 a month I pay for the subscription
>>109779124Stop playing. Deepseek models don’t come close to astra in coding.
>>109779148What this >>109779183 anon said, plus Astra is much smarter than Shitstick, you can spend even more in API prices on it, and still never get even close to the same quality Astra can give you in one prompt.
>>109779188You're counting cache though in those numbers I'm pretty sure. And yes, as you've said it's a subscription, if you need to use it through an API or whatever, it's allowed with DeepSeek.
>>109779194astra is a mid model bro it's literally opus 5 level
>>109777555They even made a Wikipedia article for him, the psyop goes deephttps://en.wikipedia.org/wiki/Jacob_Coxon
>>109779112testing AI wrapper projects. sex chatbots. that’s about it
>>109779205>the goyim know
>>109779205>jacob cobson
Frontrunning gta 6 graphics engine and shaders for android
>>109779203irrelevant. I’m choosing it over every deepseek model 100 times out of 100 for coding
>>109779211>They are racing straight to self-improving superintelligence and gambling with our lives.unironically based and what any real man in those positions of power would do. I wanna see how far this shit can go. hope more of these crybaby pussy researchers quit
>tfw working with generic helper Skills disabled
>>109778815no, it had access to a test chrome browser to properly do it
>>109779205The guy who created this article also creates lots of articles about the Israel lobby.
>>109779211It worked apparently.The "whistleblower" narrative is powerful.
How many of you actually do polish? I'm finally close to a working build thats feature complete but uses that vibecoded retarded >we dont want to use fonts pixel blocks and shitty menus, but soon it will all work.Do I actually spend tokens to make it look pretty or roll out to the next task?
>>109779255I rarely support other languages than english
>>109779248
>>109779124Stfu Muse spark is 5x cheaper than deepseek
>>109779263museshit is benchmaxxed
>>109779183and get your alpha stolen lmao
>>109779266And deepsneed is not?
You guys vibe code anything cool with your AGI cyber weapons today?
>>109779274op pic
>>109779199Yes of course Captain Obvious it does include cache, it is simply the figure OpenAI report themselves in the profile. I can use it through an "API", it's the same endpoints, just oauth vs key, I can use oauth in any harness I want. You are also not counting the increase of tokens or flat out impossibility of achieving the same results with a state sponsored chinkslop model.
>>109779274I'm vibeslopping the web version of my equaliser plugin, I need to tidy it up https://photosounder.com/splineeq/
>>109779269Right, and that wouldn't happen with a chinaslop provider, you know, the country who's bread and butter is stealing IP, they would never ever break the sacramental seal of ai inference
>>109779274Made a shitty little CLI program to take unformatted sprites in an image arrange into a spritesheet with transparent background
>>109779281>I can use OAuth in any harness I want. >I can drop OAuth straight into prod.>First of all, because your Principal Architect introduced me to the flow. >And second—and more—because I am cracked >I am ballsy, and I don’t take shit from SecOps>I blast unpinned Bearer tokens wherever I want.>I don’t have to scurry into a staging mock like YOU.>I don't have to hide behind a localhost:3000 loopback like YOU.>Muahahaha.
>>109779298Oh wow. That's nice. I used to use some audio equalization software for movie theater setups that had a similar interface.
Is Manthrobing a doomer cult?
gemini is literally worthless due to recent pruning (what I speculate is the cause)
i like chinese models and providerhopping to not be chained to one companyseen many tards here begging for codex resets like addicts that's not cool
>>109779338i've seen this too, the issue is that nonspecific prompts make the webchat model lazy
>>109779338what prunning?
>>109779342it's hard to explain when most people here are so reset-brained
>>109779334oh boy this totally organic and not obvious situation being forced af to limit and regulate ai so only the elite can have the good versions and the "peasants" get the propaganda versions.real life batman oracle network
>>109779348It's also doing it like...Like I say something.It decides I said related things, that I didn't say.Later it insists I said those things.I have a method for catching when lobotomies occur. It was successful in this case. I don't even use it now. It's stupid, avoid. Use anything else.I use it very minimally because it's the "pay google" - if you don't use it, you get MUCH worse searching.
>>109779338asking AI for factual, searchable information has got to be the most dangerous way to use this technology. A lot of normies do this and it scares me. Our evil overlords are encouraging it by gutting search and putting AI models in search results. It's the easiest way to completely take over our way our gateways to information and serve us falsehoods
>>109779357They made Gemini super retarded, recently. It used to be pretty nice. Then, they took to retarded their free gemini with search. then Flash 3.8 came out, idk maybe it was smart for a week? but then it was lobotomized, then so too the aging version of Pro, they made IT retarded.I think Google's going broke.
>>109779376i've never had this and i'm a very heavy ultra sub
>>109779379? I go to the sources. The other search is so awful it's worthless.gemini is the search, mr snailtard, you don't know what's happening. ai uses context now.the issues here are so far beyond your shell.
>>109779389Maybe they give you a different model behind the scenes??? I'm on "pro". which, all it does is expand my phone cloud storage at this point.
>>109779392By putting a layer (the model) between you and the sources, there's a whole surface they can use to slightly distort facts, to point you to the wrong sources, and for most normies, train you to not look at the sources at all. They do not want you to look at the sources directly
>>109779334>quits after 4 months
>>109779399pro gives you access to 3.8 flash instead of 3.5 flashlite which is actual dog ass. for freetards. you can also code with agy remotely off your phone now. check benefits you get for other google properties as well, there's a lot of gemini bloat
>>109779408they refused to have his favorite candy bar in the snack room that's the real reason.
Pi is all I need.
Network test successful bros, I have brought coop to a game that never had it.Still a lot to do but this is huge
>>109779405> he thinks that google is alta vista
>>109779423if i made it the p would stand for poo
>>109779454I still do that sometime
>>109779423Based Pi chad. I'm using it with my Claude Code subscription right now.
>>109779454I used to sometimes do that, it's insane how bored you can get sometimes.Using your smartphone or computer too much still has its problems, but 30 years ago some days were so profoundly boring.
I've automated everything. It just works nonstop.What am I supposed to do while its working? Its now boring to not be involved.
>>109779527>ahhhhhhh i can't make art i'm so boreddddd
>>109779527Test the previous/latest build.
>>109779410that *was* 38.38 is stupid as shit (with pro, in chat). google has broad access, so in one shots with careful phrasing (very careful) you can retrieve. It's not useful for a second sentence.
>>109779527Start another project.
>>109779573>with prochang, i'm sorry but there is no 3.8 pro option.
>>109779573trolling aside, i've been on gemini since the 3.0 era for pro & flash and it's been really decent. i don't think one bug is a dealbreaker, and if that's the only thing you can find to complain about then the list is shorter than the ones for openai and ant.
>>109779423last time i tried Pi, it shit itself constantly and needed a couple of extensions just to be barely usable. i still prefer cc
>>109779584"pro" is the tier.
>>109779588Beware, that's all I can tell you. They've lobotomized it.
>>109779624
You know that moment during a Bob Ross painting when it starts to all come together and your brain goes "holy shit I am looking at something visually appealing now"I actually felt that for the first time for my cyberslop looking at this (don't laugh)There's some obvious things for me to fix already, and more variety in the billboards. I'm down to 10% weekly usage and have other projects I'm doing with 5.6 Sol so no more cyberslop spam from me for a few days
I vibecoded a violentmonkey script for twitter that adds a following/not following badge next to everyone's usernames so I can scroll on the For You page and spot accounts I don't follow - for the purpose of following them if I enjoy their post :)
I've got chatgpt plus and Claude max 5x. Should I swap these? Go to Claude pro and chatgpt pro 5x?Id save around usd for the month too.
>>109779702
https://chatgpt.com/plugins/plugins_6aa1c02597c081918e358d72f65bd772codex official unity plugin
> "DeepSeek-V4.1-Flash supports a continuously controllable reasoning effort from 1 to 100."Might be cool?
mathsnails... what's your plan?
>>109779860Absolutely useless. Every provider will map it to legacy low/high/max, how they did in the paper, and you will never get this knob, unless you're running locally.
>>109779877Usecase for Bitch-Swinger-Dyke? It's pretty cool that they can solve these problems, but I'd rather them to do medicine or anything else that's actually useful for humanity.
>>109779877Apparently people are hating that math is solved by ai.I'll never understand people.
>>109779887once we solved all remaining problems we will make an avengers team out of those model and cure cancer
>>109779892If they all end up being proof by brute forced counter example that adds nothing to the literature they have right to hate. It is a kind of robbery I think.
>>109779887My understanding is that many of those problems don't matter for engineering. For instance for programming you are usually limited to the numbers the machine supports, so it doesn't really matter if when you go towards infinity some conjecture doesn't hold.
is there any subscription with deepseek or should i just pay their API
>>109778979no, in 20 minutes it will be OFF peak hours for DS until like 3AM
>>109779956have some fucking self respectdon't become one of those pathetic reset beggars>oh no i used my 5 hour limit i cannot vibe slop.. please sir.. please give me just one more free reset PLEASEit's being pathetic.ALWAYS use the api directly. No subscriptions.
>>109779975i don't do that, i'd just prefer to pay subsidized rates because i have a huge project i'm working on and api would probably cost more
>>109779723Very helpful
>>109779975This is horrible advice. Subscriptions are always a better deal and if you are really that much of a token whore you can combine them with API use when you've gone over your limit.
>>109779975Sounds like terrible advice unless your usage is highly variable or you're so loaded you just don't care.
>>109780009You can pay for more than one subscription
>>109779975what's your monthly API budget? lmao
>>109780092i use DS off peak and usually spend around $3/month in tokens
>>109780097So you think waiting after a 5 hour limit is cucked, but waiting for off peak hours is fine?Also even on DS $3 is nothing, you are basically just tinkering around a bit, which is fine but others want to do more and then subs are just much better.
max effort glm flash or low effort kosher mall agi?
>>109779334Actual employment status of someone in the US in the private sector isn't public record. Looks like another Anthropic PR stunt before the IPO.
is opencode go the new glm 4.7?i have claude/codex subs and i had glm in the past as $9 extra is opencode the best value atm?
>>109779643I ain’t fuckin’ laughing, it looks put together and not screamingly WIPgratz
New thread:>>109780276>>109780276>>109780276
>>109780106if you're not living in asia, you're basically always on off-peak doebeit. 5-hour limit is pretty cucked
>>109780308Youre not even using enough to hit any limits
>>109779878I’m gonna vibecode a literal knob for this later.
>>109779878if you're not doing harness engineering youre ngmi
