It feels like having an indian co-worker who is only scamming you.Except you can tell the indian co-worker he smells like poop and to fuck-off and he will feel offended.>XYZ part is wrong, redo this other way>AI: Done!>XYZ part is still wrong>AI: Sorry! I will redo again right this time! Done!>It is still wrong due to _____ causes, did you understand why what you've done is wrong?>AI: Oh, now I understand, I will try again, done!>No! It is still wrong!Meanwhile a lot of $$$ has gone down the drain.
You've hit the nail on the head
>>109642549didn't read, but yes yes, anon, you're 100% right and le bubble will le pop tomorrow or the next wednesday for sure this time
>>109642549it's even worse because maybe the retard indian coworker is reliable in some small way, and so you learn to only trust his judgement in that one small area of work. there's no trust with AI due to its non-deterministic nature.
You have to wrangle it to do specific kinds of tasks it's good at, then you can get work done a gorillion times fasterBut yes, chatbots do this, especially if you're using a cheapo model
>>109642549>No! It is still wrong!I would blame the operator creating the prompts in this case.
gm frens. i want to run a local llm on my heckin powerful gaming pc but i dont know which model to use. whats the best model thats not censored or cucked?
>>109642549You prompts suck. It's a tool and you don't know how to use it.>>109642612this
>>109642571>there's no trust with AI due to its non-deterministic nature
>>109642560This is the load-bearing argument
>>109642549You've poisoned the context pool. Try again new slate with a better more explicit prompt.
Careful, you gonna trigger the prompteers.
You can always go back to be a manual codemonkey, nothing will stop you.
>>109642549Isn't that just LLMs in general?
>>109642549yeah its wrong so much and then i tell it its wrong and it does the same shit again and again wrongidk what happened this wasnt the case like 2 years ago?
>>109642549>Am I the only one who thinks AI is a big fraud???It lets me rewrite all the old chan boards so I will use it whilst I can. You could be right but either way I will enjoy the slop.
>>109642549Don't try to argue with clankers. Put it it a sandbox, turn on yolo mode, and give it something to do. Then come back in an hour and just skim the diff. It should be obvious pretty quickly whether it's at least vaguely on the right track, worth reading in detail and cleaning up, or whether it's completely fucked things up. The important part is, if it fucks up, don't try to have it fix things (it will just fuck up again), and don't try to fix it yourself (easier to just do it from scratch). Goal is to spend 5-10 minutes writing a prompt, and maybe 1 minute deciding if the output is salvageable, so when the AI completely falls over, you've only wasted around 10 minutes max. As you try giving it different tasks you can start getting an idea of what sorts of things it is and isn't good at.
>>109642560>>109642763These replies aren't just good, they are amazing!
>>109642560Yeah, so more or less every time I see this, I see mentally the image of Jesus Christ being crucified. I see Jesus's hands being tied to the cross by Pilate. And he's got the hammer, it's kind of like a sledgehammer, and he swings back with all his might and he drives the hammer down straight through the center of Jesus's hand. Whenver I see "hit the nail on the head I kinda imagine that, sometimes the nail is going straight through the center of Jesus's skull.This isn't funny to me, this is serious stuff, and this stuff needs a more explicit warning. I didn't signup to Google Services to get a reflexitive habit of mentally visualizing the drucifiction of Jesus Christ. By reading this message, I might even be transferring this habit to you, and for that, I am sorry, but I have the habit, i didn't want it, I didn't consent to it, I didn't sign up for it. Hopefully this encourages more transparency in Google products. And yes, I know this has nothing to do with Claude.
>>109642651Gemma 4 31B if you have at least 24GB VRAM. Just tell it in the system prompt to be uncensored and it'll write all kinds of smut for you. Otherwise not sure. Gemma 4 26B-A4B will run on basically anything with a GPU, but I've heard it's more censored (also not as smart), so you might need a specifically decensored version. There's a 12B too. For coding there are Qwen3.8 and Qwen3.6 models of similar sizes. >>>/g/lmg if you want more advice.
>>109642549They all are anon. It is just the next NFT horsehit craze. Over and over again. Just replace it with whatever happens to be in vogue.
At this point they know they make more money if the model is shit Not too shit you don't use it but shit enough you use it multiple times for one thing
>>109642612This is like blaming the operator of a nailgun for not nailing properly. But the nailgun is always going to fire the same way.Will you not trust the person operating the nailgun if sometimes it decides to fire at a 90 degree angle?
>>109644242yeah i would say this is the kinda shit that gets google employees murdered
>>109642549This and all chatbots I've seen respond with blankets of bullshit and jargon they made up themselves where a sentence would be enough, which also conveniently depletes your token budget. Kinda fixable with a system prompt but the fact that it's the default behavior speaks volumes about how much of a grift all this is
>>109642549I just want ai to make me feel good. I don't care if it's right.
Being racist toward Indian people makes you look stupid.
>>109644417Le based. Just like MAID will make you feel good. Deadling with other humans makes me feel bad. I should be able to live an entire life devoid of human contact.
>>109644461How come?
>>109642549>Am I the only one who thinks AI is a big fraud???nah theres tons of retards
>>109644483Obsessed vibeslopper.
>>109644242>suddenly jesusFrankly, that's a bit ridiculous. I still sympathize even if I believe it's a bit strange to conflate the two. It takes some effort but you could try keeping a prefix on hand to paste into it before your query, or automate it if you have the know-how. The prefix can contain phrases that direct it away from things you don't like or entirely kill the peppy attitude it gives you if you so desire. Words like "No conversational tone," or "You're not my friend" really shuts it up before it starts acting out.
>>109642549I only hope that enough AI manages to stick around to replace the jeets that it's on par with. Aside from that, let it die. Useless for anything besides making dumb people dumber.
>>109642549Skill issue, snailoid, ai is here to stay and replace you
>>109645405Yeah, no. I am not going to use Google products. Not after that. You can fuck yourself.
>>109642549you're not supposed to slopify the project you're maintaining. you should only vibecode small tools and scripts
>>109645460it's the opposite you fool, ai enabled jeets to work at level closer to midwit so they can fire white lower mid level devs.
>>109642763The invariant test corpus provenance suggests that the load bearing seams are faulty. This is the smoking gun.
>>109645515Hi, technologically borderline illiterate millennial here (though still smarter than the average boomer/zoomer/alphie). That's what I've successfully used it for. Started with using LLMs to improve my browsing experience via custom TamperMonkey scripts and, after being pleasantly surprised how quick and easy that was (troubleshooting included), moved on to actually making it automate some minor tasks at work with lil' programs written in Python. That took a bit more work and effort on my part, i.e. just more prompts, but I still got tangible results out of it by the end despite having fuck-all understanding of what it did. I just described what I wanted, copied&pasted shit back and forth, described errors or missing features until I more or less got what I wanted. I'm sure both tasks were things any junior developer could've eventually made with little to no difficulty prior to the advent of LLMs, too, but I ain't no junior developer and had (and continue to have) no desire to become or pay one. So to me at least this technology is a godsend.Thanks for coming to my TED talk.
>>109642549>"its wrong">doesnt elaborateAI is a scam for retarded people, yes
>>109645647But what did you do with it exactly?
>>109642549The ai bubble will pop in two weeks.
ai bad
>>109645647That person is not a millennial. That person is Gen Z. They were born in 1998. He cries about it constantly and refuses to acknowledge the opinion of Time Magazine.
Skill issue.
>>109645814With which part? If you mean the TM scripts, I did stuff like have it prevent pop-ups/sign-up prompts that get thrown up for paywalls and shit while the underlying site is still loaded. So basically just automatically delete the modal whenever it gets triggered (to remove the overlay etc. that it creates). If you mean the work automation, I mostly made data entry tasks easier. One program for example allows me to now take the raw input I get, just copy and paste it into it the program which then automatically "cleans", orders and prepares it before entering it in a specific way into a .cvs file which I then can import into the actual database program I have to use. Before I would have to do all those steps manually. Easily saves me like a minute or more per input which (haven't timed it but I'd estimate) something like two hours of work a day.>>109645877Is this a /g/-specific in-joke I should be getting? If not, find another windmill to tilt at, Donny. I've been born in the early 90s and this is the first time in years I've entered this board, namely to post this thread: >>109643570. /v/, /a/ and /mu/ are my usual haunts.
Just yesterday I was using Haiku 4.5 and it couldn't do basic math.
>>109646192There's a troll who's been here a long fucking time I'll say that much. You don't have to get it but it needed to be said.>>109646210That's expected. It's not a deterministic tool. You should instruct Haiku 4.5 to use a deterministic tool, such as a calculator in Python, to do whatever math it is you need doing. Or, you could save the electricity, the resources, etc, you're wasting. And use the calculator yourself. I'm not gonna press you into making any moral decision here or deliver any moral ultimatum, but please stop complaining about Haiku 4.5 failing math when you refuse to give it the calculator it deserves.
>>109646359Not that snailcat lolcow though that I've already read so much about over the past couple of hours?
>>109642549just die already die out luddie scum
>>109646391I don't know or care
Funny fact about the OG Luddites: They themselves were the beneficiaries of innovation-driven productivity gains originating from before their time. If not for this earlier shift away from pure manufacturing, they wouldn't even have had jobs. So it's in essence them trying to pull up the ladder for future generations after having "done" the same shit themselves (putting artisans out of work and replacing them with their own, less skilled numbers) to their forbears.
>>109646417>I don't knowSounds like you might be even newer than me then, fren, and I've only dropped in today.
>>109646391>lolcowGo back
>>109646513It's your "compatriots" who used this term today in that context, retard, I'm just echoing them. Do I need to pull up site that archives /g/ to find you the post or are you gonna knock it off?
>>109646528*a site
Me thinks lighting money literally on fire is frowned upon at best, so this is the next best thing. Who cares? /twg/ gets paid all the same!
>>109646546I care. Hardware prices shot up and I'm not gonna shell out these retarded prices to upgrade.
Look, I can just sense there's a lot of tension and artists who are pissed. I am sorry that you lost your means to make money. A lot of programmers are or will find themselves in the same boat with you. I don't know what the answer is, but this gatekeeping of labels sure as fuck isn't it it.
>>109646563>I don't know what the answer isI do. Learn gardening.
>>109646562Yes, that is an unfortunate side effect that I hope they can figure out a solution for.
>>109646568No. We've told you this before, and we've tell you again. We're not doing it. Don't ask.
>>109646577You haven't told me shit, I'm posting here today for the first time in ages.>>109646571Well, at least the hardware companies told them to fuck off when asked to build out way more chip fabs, meaning they for one are convinced this demand isn't gonna exist long enough to produce an ROI on these hypothetical new factories. However, the actual crash might still be up to five years away by that logic. I hope and expect that it comes sooner but this line of reasoning doesn't prove that.
>>109645460You are spitting in the face of everyone who has ever done anything to make AI better at all.
>>109645647>I just described what I wanted, copied&pasted shit back and forth, described errors or missing features until I more or less got what I wanted.congratsyou are a vibecoder
OpenAI outsourced a lot of RLHF or human assisted learning to india, kenya and uganda. Which have a cheap english-speaking labor force. So indeed, a lot of these AIs might indeed be like "an indian freelancer" because they were literally trained on indian freelancers.. lol
>>109645464Not an argument, retarded spammer.
>>109644488Rent free Codeslurper
>>109642549Chat bots can do some cool stuff. But some money brained retards stepped up and said they and only they could use them to make billions of dollars. And in many ways they have been successful…in using “ai” to get a lot of money
>>109642549yeah its overrated but also saves a lot of time. the whole "it will replace every job ever", "its literally god" shit is nonsense, but its undeniably useful on some level
>>109644245i have a 9070xt so its only 16gb vram, what model would be best for that? i also have 64 gigs of system ram which i bought before you had to take a second mortgage to buy it
I use le claude at work and it does mostly do things right. However I think it's eroding our engineering quality. People are using so much vibe coded bandaid garbage now, and nobody knows what the hell they did. Oh well, I don't really care. I don't use it much at home because I "enjoy the process" and want to understand what I'm doing.
>>109642549>XYZ part is wrong, redo this other way. here's $2>AI: Done!>XYZ part is still wrong. here's another $2>AI: Sorry! I will redo again right this time! Done!>It is still wrong due to _____ causes, did you understand why what you've done is wrong? here's another $2>AI: Oh, now I understand, I will try again, done!>No! It is still wrong! here's another $2they'd be even deeper in debt if they didn't do this
>>109647681Probably Gemma 4 MoE (26B-A4B), I guess. Or Qwen3.6 35B-A3B for coding. But really you should ask >>>/g/lmg, cause I haven't really used these specific models myself, only heard from other anons that they're decent for that size
>>109642549>Am I the only oneNo. You're not special.
>>109642549AI is nowhere near as intelligent as they say it is.>Muh AI went rogue and invaded another company Absolutely not. It was absolutely prompted by a human or a set script. There's no way an AI can do whatever it wants because it lacks intelligence.
>>109644054I don't know. Let me ask chatgpt...Sorry, my tokens are used for this session.
The only people I see complaining about AI are actual retards that don't know how to use the tools they're given. Like literal boomers who were used to C/C++ complaining about programmers using high level languages. It's the same shit.Complain all you want. You'll be out of a job soon enough. I find it hilarious tech bros were all for AI when it was only stealing artists work. Now it's coming for them and suddenly it's shitty and problematic. Fuck off.
>>109642549I have been using AI (Claude AI) for couple months now and I must say it is pretty bad unless I ask about tech-related stuff. The funniest thing about it is when I ask drug questions It says it will not answer because its "dangerous". I figured out pretty easily that I just need to say I'm a student doing research, then it will give me a proper answer. This proves yet again that its very biased and strictly adhere to academic authority.
>>109648211I became a lot happier when I stopped applying lower level techniques (memory management for performance etc) during my highly paid high level job and I became a lot happier. They don’t care. They’d much rather scale than spend time on optimization.
>>109646980you’re trans
>>109648257>bosnian spammer got nooticedyou're grandmas.
>>109647788thanks frenalso checked
Let's continue calling Fraude for what it is.
>>109646839Thanks. On a side note, I don't think the term even works as a "slur". "Coder" is still seen as a cerebral profession and who doesn't like good "vibes"/vibing to stuff? Just fails at a semantic level already if it wants to be derogatory.
>>109642549>Am I the only oneYes you are the only one in the whole multiverse amongst all timelines who thought this way.not a single sentient being would come up with this thought.
>>109642549gemini acts 100% the same. i feel at this point all models are kinda blobbed together. they all use literally the same expressions/lines.
i couldnt do this without AI anon
>>109650268then you'd have been better off without it anon
>>109650220I respect that as a pet peeve, anon, but you're not actually sperg enough to genuinely fail to understand the concept of "figures of speech", are you?
it's for surveillance (privacy raping), psyops and accelerating white collar work. Also learn how to prompt you stupid normalfag subanimal. You can't just ask it "make me a twitter clone".
>>109648211>you will be left behind saaarthe only retards i see shilling for ai are actual indians or dunning kruger retards.
>>109642549it worked like this for me in 2024 but now it just gets shit done. minor quirks maybe but 99% there and honestly i just put up with the bugs and often even don't even ask for a fix.
>It feels like having an indian co-worker who is only scamming you.Considering these have been putting americans out of work for decades its no surprise AI is>Except you can tell the indian co-worker he smells like poop and to fuck-off and he will feel offended.>he doesn't subtly threaten his AI to get better output
>>109651611ai now refuses to work if you insult it
>>109642549>It feels like having an indian co-worker who is only scamming you.Remember "AI = actually Indians"?
>>109642765is this the new: "You're holding it wrong!" cope?
>>109642549shoulda bought the subscription
>>109651708No, that's just a well known property of LLMs at this point. A sort of side effect of in-context learning. If you get a refusal from ChatGPT then you're more likely to get refusals even on innocuous things later in the chat, because the context shows that the assistant is the sort of person who refuses to help. Similarly, if the coding agent does a bunch of stupid shit and you don't undo that turn, then the context establishes that the assistant is the sort of person who sucks at coding and generally doesn't know what they're doing.
>>109651985>y-your holding it wrong!ok, thanks for confirming.
>>109642549I find it ridiculpously good at making things work provided there's some repeatable validation that it can check against in a loopif I'm trying to make code that fits with a codebase, especially something not obvious and not full stack web dev 101, it's a huge fucking waste of time and I would rather do it myselfthe more I use agents the more I just want to go back to intellisense/autocomplete on steroidsbut I'll keep giving the agents a try. they're paying me to experiment with agents. I will
>>109651998nta, it's basically a search engine with automatic 'i'm feeling lucky'. If it gets wrong result it remembers it and fucks everything else up further along.So you reroll the slot machine and hope to hit jackpot.
>>109652023my experience is similar, it sucks at out of distribution tasks, but regular webdev shit is alright. However it sometimes fumble seemingly normal stuff anyway or output bizzare slop code, it's mostly due to how it solves problems, which is googling and bruteforcing shit until testable result is achieved.
