🔧 Site instability resolved. You can report double-posts and broken attachments. For bigger issues, use the Technical Grievances thread.
🇵🇦 Nuestro primer dominio localizado está en español en kiwifarms.pa. Our first localized domain is on Spanish on kiwifarms.pa.
Want to keep track of this thread?
Accounts can bookmark posts, watch threads for updates, and jump back to where you stopped reading. Create account
The agents are designed to produce unreadable code so that you're locked into their platform. It produces black boxes of nonsense unless you give it specific architectural instructions, and even then it will find every excuse to ignore directives and build things in the most convoluted and unmaintainable way possible.
They don't reason. It's just code rewriting and some if/else sprinkled on top for most common tasks.
It pretty much just transforms what you wrote into probabilistic outcome of code someone else wrote.
Like snippet engine that randomly picks code snippets from stack overflow.
Vibecoding is great if you tell it more than just “make gaem pls” and actually know how your program should be structured but just don’t want to mess with getting the syntax right. It’s both impressive and horrifying knowing I could now take a few hours to shit out programs that once took me a week to make, and the AI version would be better.
Vibecoding is great when you are precise. Interesting. Maybe we should come up with some languages that could be more precise to describe our programs in.
Vibecoding is great when you are precise. Interesting. Maybe we should come up with some languages that could be more precise to describe our programs in.
I feel like no one cares about maintainability anymore. We had a lead engineer leave at work and I assigned to be the lead on the project. He loved Claude, and it shows.
Most people don't care. The vast majority of devs just want to get the pay cheque and do as little as possible.
I can get Claude and OpenCode to do all the boring stuff (which is all of it these days). My job is building uninspired web applications that will be thrown away in less than a decade.
Every job is some of this stuff, and if you avoid it, they actively lie about what you will be doing in the job.
I've basically given up after my boss wouldn't fire a dude who cannot code. At my last contract position, I heard second-hand from a guy that I met up with 2 years later who used to work with me on that team. that the attitude in the team was "Get SchizoDaemon" to do it. This is how the vast number of people operate, because the organisations are run like adult day care, and you cannot get anyone to do anything. However, if you actually point out someone's retardation, you end up having a meeting with a superior about your attitude.
For me it depends on use-case. If I just need a personal python script to do whatever with some files and thus it doesn't matter that it's horribly inefficient than it's fine. I've got an NPC generator for my tabletop campaigns that has a GUI for race/gender weights so I can just fill out a location in my notes, had it draft a couple browser extensions where I couldn't be arsed to read through 3,000 lines of HTML/CSS to figure out what <div> I need to call on when they're all over the place, that kind of thing.
I wouldn't use it for anything important but sometimes things aren't important and I don't feel like spending my entire afternoon on it.
Sup Vibe Chads. Because Microsoft can't stop self sabotaging Copliot, the AI Harness/Platform they've developed has changed pricing from usage to token based and have effectively torpedoed their customer base.
Now all I got is some screenies of 4chins that are from reddit but I can confirm this is really really dogshit. Like you're better getting stuff directly from Anthropic or Open AI, not even subscriptions just API use.
Sup Vibe Chads. Because Microsoft can't stop self sabotaging Copliot, the AI Harness/Platform they've developed has changed pricing from usage to token based and have effectively torpedoed their customer base. View attachment 9089376 View attachment 9089378
Now all I got is some screenies of 4chins that are from reddit but I can confirm this is really really dogshit. Like you're better getting stuff directly from Anthropic or Open AI, not even subscriptions just API use.
The new pricing probably still doesn't cover the actual cost of usage. I imagine they have to be depending on under utilization (the same over-provision model for VMs used by Digital Ocean, AWS, Azure, etc.) They really need corporations to buy everyone the $100 account but less than half of their subscribers really use it.
Google, Microsoft, nVidia, et. al. have amassed so much spare capitol they can just burn through money right now. They're going to need to squeeze people who are now dependent on it, like a drug dealer, without losing people who would refuse to pay and just go back to coding things by hand or local models. But not before destroying the RAM and consumer compute markets.
I do wonder what the true break even cost-per-developer is for OpenAI or Anthropic. OpenAI didn't even try raising the price of Sora to something ridiculous like $1,000/month. They just killed it because they knew there was no way to get those two lines of the graph to ever cross. I've heard xAI is something like $60/month to get dozens of 5 second video a day? Grock takes a few minutes where my 300w AMD card takes an hour with local models for 5 seconds of video with hunyuan. There's no way xAI can be sustainable either.
The new pricing probably still doesn't cover the actual cost of usage. I imagine they have to be depending on under utilization (the same over-provision model for VMs used by Digital Ocean, AWS, Azure, etc.) They really need corporations to buy everyone the $100 account but less than half of their subscribers really use it.
I think while your point may be true when analyzing the market as a whole, to some extent. I think we're missing the trees for the forest in this case.
Google, Microsoft, nVidia, et. al. have amassed so much spare capitol they can just burn through money right now. They're going to need to squeeze people who are now dependent on it, like a drug dealer, without losing people who would refuse to pay and just go back to coding things by hand or local models. But not before destroying the RAM and consumer compute markets.
The thing is this isn't the squeeze. This is Microsoft screwing themselves over and then screwing themselves over to mitigate the first screwing over. In the screen cap about 50% the redditor admits to using 5 concurrent agents, all full sized. The fact that he was even able to do that before is ridiculous and showcases the incompetence behind Copilot as a service.
That screenshot showcases somebody burning 3k of Copilot resources for 50 USD. That's fucking madness. This isn't a get big quick scheme, this is Microsoft not having a single clue how to run any business anymore. They released features that allowed power users to increase their performance per dollar without any actual gates.
Subscriptions like these work by having a nice large volume of normal users to power users. What Copliot has done is allow the power users to multiply the cost to servicing them soo much that they had to mitigate and mitigate hard. Also because unlike Open, Anthropic and Gemini, they already targeted a higher use market. The other LLM services cater to everything from grandma asking for recipes to lonely women who want a virtual boyfriend. Meaning it's much much easier for them to balance things out in running costs. Copilot just serves fucking nerds and they had no safety rails so they got fucked and now are fucking themselves by responding with the nuclear option.
What will happen because of this is that their competition will get more customers.
I do wonder what the true break even cost-per-developer is for OpenAI or Anthropic. OpenAI didn't even try raising the price of Sora to something ridiculous like $1,000/month. They just killed it because they knew there was no way to get those two lines of the graph to ever cross. I've heard xAI is something like $60/month to get dozens of 5 second video a day? Grock takes a few minutes where my 300w AMD card takes an hour with local models for 5 seconds of video with hunyuan. There's no way xAI can be sustainable either.
I think your point is valid but I also think it is stretching things. Video is the new frontier of AI, so there is bigger costs there. But when looking at code, text and images the open source models have easy to plot running costs and while I imagine the frontier models are much larger and harder to run I think their API costs are pretty close to running costs. So if Copilot is offering you API plus 20/35% why even bother with it when Codex and Claude offer you much more for a better package. And a small addendum outside of the Open source local models people aren't yet prioritizing efficiency. I think if need be they could easily make refined Codex, Sonnet and Opus that are maybe 10% dumber but 30-40% more cost efficient.
Edit:
I decided to go dumpster diving on reddit and I think this image explains why they changed pricing:
Vibe coding has become my hobby at this point, but I just can't see myself paying more than $60-$80 a month at the very max (just $20 is enough for me to reconsider this whole hobby). Maybe I'm cheap, but paying +$50 for SAAS feels like being exploited, not matter what since you can be rug pulled at any moment, just see how Anthropic treats their customers.
Kek! If software devs weren't so retarded I wouldn't be here. I can "code" my own little offline apps that just works with a single click on any operating system or CPU, minimal dependencies, or hidden SAAS shit. The UI exactly how I want since FOSS niggers can not make a usable UI to save their own lives.
If devs actually made good software it wouldn't be up to tech illiterates like me to try and save the modern computer experience for just typing a document or edit a video. I will vibe until the newer models starts hallucinating my GitHub spam or AI regulations raping us all.
should of left it at c++ instead of making all these pussy nigger languages like java and c#.
when was the last time a control was added to winforms? is the default textbox enough? it can't even render markdown.
how long will it take for microslop to upgrade textarea to markdown?
what year is this?
it's 2026?
fuck me, fuck me.
what year is this?
2026
shit
let them all be programmers. we don't need janitors or plumbers. let them all vibe code.
you can't beat efficiency of brain at inventing new shit. algorithm doesn't have the initiative.
you have no idea what year you live in. we don't need electron anymore.
it's 2026, not 2006. no one cares about helvatica anymore.
solidworks and computer games, that's all there will be. only for some of you.
in ten years, less than one percent will have phones, and they will be without cameras and have removable batteries.
solidworks costs $4000 last i checked? and since llms are more useful they will cost more to use.
and there will only be one kind of laptop, thinkpad p16 gen3, and there will only be one variant.
we don't need middle class, we have a hard time removing the current one percent.
those who will succeed are those who will use LLMs the least.
let's see you vibe code at $5 per token.
what are you gonna do when they have to replace old hardware with new? do you think they're gonna run old software on new hardware?
they're gonna raise the floor, that's what they're gonna do.
this entire thread will be deleted to save humanity the embarrassment.
maybe there will be discovery to bring local ai to desktop but that hasn't happened yet.
you're on borrowed time, you and everyone else here.
you are making the same mistake everyone made when they didn't invest in bitcoin.
the prices are going parabolic.
i hope im wrong, or this will be a new type of slavery where LLMs are the leading sector pulling everything behind them, food and energy.
get ready to pay 20x for food.
it'll never be smart enough for these vibe coders. they're slavers. the middle class wants to have slaves too.
have you ever seen an iceberg flip upside down? it's deeper than it looks. has different color.
their employees are probably holding back some discoveries too, i know i would be if i was working for these companies.
interesting how no one is complaining about them taking away temperature, vibe coders don't even know what it means probably because it's not in the latest models, and instead of discussing why that is we're, they're boiling us alive like frogs and vibe coders are contributing.
you don't have the balls to compose your own context and take the time to define what you really want.
vibe coders are like these faggots with private jets, who golf, and have yachts.
vibe coders should be strapped to elon musk rocket and be flown into the sun.
there will be only laptops and desktops. there will be no phone, tablet, watch.
you want a phone? get a casio.
vibe coders are just boomers on fast forward. they'll crash the economy before the boomers die.
sam altman is probably stomping as many kittens as characters in this text. he's gonna go full vader. he has to. they need to lift sanctions on russia to close the gap and become profitable. i bet openai is holding hostages. i bet there's multiple levels of slavery.
none of you know what you're doing, you and boomers, same fucking shit, now 2x.
they've taken away one model, they'll take away more. they'll walk us back to stone age if they have to. and they themselves won't understand what's going on.
method is the new weapon, not information. before that it was nukes.
still don't understand? we don't need weapons anymore. one law and everyone knows their place.
there won't be police, there won't be interstate highway system.
My nigga I am vibecoding with Qwen 3.6 35b-a3b on my gaming PC right now. Models will only get better and cheaper to run, hardware will only improve. I am so sick of these dumb conspiracy theories, "they'll walk us back to the stone age," please take your meds.
