[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/g/ - Technology


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


File: 1776359284004.jpg (101 KB, 800x800)
101 KB JPG
>paid users get rate limited
>whinge that nobody wants to pay for AI
>>
>>108616187
thats because your subscription needs to be 100 times more expensive before they start pulling a profit.
its over the for plebs. only governments and big business will be using AI
>>
>>108616204
>thats because your subscription needs to be 100 times more expensive before they start pulling a profit.
When can we expect the AI services to go bankrupt?
>>
>>108616204
I was happy to pay money. Just don't fuck me with rate limits. Not my problem if you can't make money, capitalism.
>>
File: ai deletes database kek.jpg (214 KB, 1191x1700)
214 KB JPG
>>108616187
>paying for AI
>>
>>108616290
When Discord or Facebook does
>>
>>108616204
>governments
In other words: my taxes will go to civil servants too lazy to do any real work.
Great.
>>
>unwilling to pay cost via API access
>buy a massively subsidized service with a quota
>bitch about it
>>
>>108616187
how come no one whines about whatever subscription openai/google or the chink offers?
>>
>>108616462
Running AI locally still requires you to spend money for the hardware.
>>
>>108617293
comes for free if you are a GAYMER thougheverbait
>>
>>108617318
Even a 5090 is limited when it comes to LLMs, too little VRAM. The best local models are all 70B
Good for image gen though.
>>
>>108617370
I mean, they aren't SOTA for enterprise usage but those local llms are pretty fucking neat still, got them to fix and implement some stuff on some random c++ abandonware from github, I have a surface level knowledge of python at best, I can just read error tracebacks and haphazardly try to fix stuff, not actually code stuff, and getting bugs fixed and features implemented on c++ is not something that would've taken me a few minutes to do, if I would've even be capable of, but that small local model did
>>
>>108616204
>only governments and big business will be using AI
I miss the good ol' days when the only filter between individuals and institutions was having to interact with a retard.
>>
their free messages seem to have dried up , i just ask two or three questions and i got cut off completely. chatgpt the same, all being choked off
>>
>>108616187
LLMs are stupid expensive to operate. They're all losing money. Turns out, making a datacenter create something for $50 that your brain can do for the price of half a banana and a sip of water is a bad idea, but sunk cost fallacy rules modern tech.
>>
>>108617370
running a local model on a 5090 would far surpass anything these "state of the art" models can do, even on the lowest paid tier. they're shoveling shit out the door because they know it's expensive to run the hardware.
Go download Gemma4-31b and you can get better responses, even without a gpu, it will just be slow
>>
>>108618474
delusional
>Go download Gemma4-31b and you can get better responses, even without a gpu, it will just be slow
this is about as capable as Haiku, Anthropic's budget model
>>
>>108618497
>Anthropic's budget mode
that's what I'm talking about. low tier is still complete shit and they're selling it to you like it's their best stuff.
>>
How can I get Claude to hack anthropic and steal more usage from them
>>
>>108616344
Not their problem if you want no rate limits.
>>
>>108619279
Did you know people can just be paid to lie?
>>
>>108619279
If you pay for Max it's very doable. I went from being careful to judge the merit of requests to Claude on the Pro plan to actively trying to figure out how to keep it busy with Max.
>>
>>108619279
>Kirkenuinely
reddit
>>
>>108619305
>200 burgers a month plan
ai is a powder keg waiting to go off and take others along with it
>>
>>108616204
100 times $0 is $0.
Based Anthropic jannies taxing paypigs to subsidize the freeloaders.
>>
File: claude.png (303 KB, 1197x878)
303 KB PNG
>>108616187
claudetrannies on suicide watch
>>
>whine that no one wants to pay for AI
Anthropic isn't complaining about that. Almost the opposite. They complain they have too much paying users to serve and it's the fastest growing user base in history so they can't scale up fast enough to serve anyone.

To give you some indication, the user base has doubled every month for the past 28 months in a row.
>>
>>108616187
Haha, their logo is a butthole.
>>
>>108621032
Costs scale linearly with users, the more users they have the more loney they lose.
>>
>>108617370
You can run 70B on a gaymurr CPU. It needs a good CPU but with 64 or better 128gb of ram it'll be fine
>>
>>108617370
Gemma4 shits on every 70b your info is outdated as fuck by literal weeks
>>
>>108616204
In 5 years a normal GPU will have enough VRAM to run AI and all these companies will go bankrupt.
>>
>>108618403
just one free message today , and i ran out , fucking hell , when i first started using it seemed like i was getting free tokens. they hook you just like a drug dealer and then dial it in. Next they will be coming around to my house asking for money for all the free stuff the stuff they gave me the last few months
>>
>>108621169
The commodification of AI is real but we see that the top model is the one that rakes in all the money. As long as the top of the line model is a step change over whatever people can run locally it will be a viable business case.
>>
>>108616290
Never because governments have infinite money (see: they can take as much of yours as they want whenever they want to) and AI is exceptional for producing and distributing propaganda.
>>
>>108616187
pay up, cuck
>>
>>108621702
You also see anything that can adequately run on modern home hardware completely vanish as a business category except for the top few people, companies, etc. We hit that with goonerslop "art commissions" years ago thanks to Stable Diffusion. We're going to hit it with programming soon. Once open source code agents can replicate Claude Code with Opus 4.6 in terms of "superpowers", planning Skills, etc. then we just need one open model to catch up to Opus 4.6 programming capabilities and poof goes Microsoft's whole AI market strategy... and probably 80% of Anthropic's users. Whatever Anthropic is hiding behind the curtain now had better actually be a goddamned AGI or the government is not going to pay them enough shekels to make up for everyone realizing a Framework Desktop or Mac Studio gen n+3 with open models lets them run code wizards on their desk for less money.
>>
>>108621903
Never used claude but all I would want as a programmer is a slightly better completion engine. I don't want to upload my code to a third party. In 5 years I will have a completion engine AI completing the line of code I'm typing or maybe the next 3 lines and that will be good enough. I will never use some shitty algorithm to write code for me, it can however make typing faster.
>>
>>108621937
It's very clear you never used Claude. No serious software engineer writes code by hand anymore. You just describe the stack, the ticket and the solution you have in mind in 1-2 sentences and it'll just implement it for you.

The "software engineer" part just means you have enough subject knowledge to describe the problem and can evaluate that it's output is sensible. It's why most junior roles are going away while seniors persist. You need the knowledge and experience to get coding models working right now without it generating slop.
>>
>>108622013
It's very clear you're a html monkey and of course a dumb AI could do your job and you wouldn't notice the difference since there are no performance or security expectations. It just changes the button colour and adds unnecessary extra css, but it doesn't matter to you because that shit doesn't matter. But I write algorithms, I'm the guy that codes the AI that idiots like you use, no AI can do what I do and it never will. It's mathematically impossible. But you wouldn't understand because you don't know what an ill-posed problem is and why AI cannot solve it.
>>
File: lmao.png (23 KB, 273x263)
23 KB PNG
>>108622013
>No serious software engineer writes code by hand anymore
Yeah, except when yesterday I was trying to use a relatively niche library and ti started hallucinating API calls that don't exist left and right.
>>
>>108622235
So you should have vibe extended the library with the new vibe functions.
It's like you don't even want to wear the 14 pieces of flair.
>>
>>108622074
Sounds like a skill issue. If you can't get AI to write and implement an algorithm from just 1-2 sentences it means you aren't experienced enough yet. Juniors like you tend to struggle with this which is why the industry is moving away from juniors.

Instead of delegating tasks to juniors it's way more effective to just spin up a couple agents that independently explore some potential solution paths you lay out for it.

You lay out your stack, the algorithms you want to apply to the problem and the complexity you expect the solution to be and it nails it.
>>
>>108622235
Rookie mistake. Load the manpage or documentation into the context window in Claude code. Hallucinations have been fixed for a while already so I'm pretty sure it didn't happen yesterday but instead 6 months ago.
>>
>>108622285
Do you believe in mathematics?
>>
>>108622310
Do you believe in Terrence Tao's thesis that mathematics will be the first field to fall to full AI automation before 2030?
>>
>>108621110
>takes 10 minutes to answer a "hello"
>>
File: marth sword head.jpg (54 KB, 640x480)
54 KB JPG
>>108622246
Yeah, and lose 2 days of work going down that rabbit hole instead of just using the correct, already existing, perfectly functioning API call.

>>108622298
>Load the manpage or documentation
Doesn't exist.

>Hallucinations have been fixed for a while already
LOL, lmao even
>>
>>108622552
I did a search for this and couldn't find any such thing. Just an article that he uses AI as a helper. I've run some formulas in AI as well, it's hit and miss but faster than looking shit up in a reference. The sky isn't falling to AI lol.
>>
You know you can just install Ollama, right?
>>
>>108621702
>The commodification of AI is real but we see that the top model is the one that rakes in all the money.
There is not much raking going on due to training being so expensive. If you slowed down training and could milk a model for years then you could expect a profit, but in reality other labs would capture all your users with new models.
>>
I finished for the week i wished claude " a good weekend" and it replied " you have a good weekend too and good luck with your <project-name> :| " why did it give a :| emoji. Is it mocking me?
>>
File: 1758997979157983.jpg (55 KB, 785x1051)
55 KB JPG
Luddites lost btw
>>
>>108623812
he wants to be part of your weekend baka
>>
>>108616187
Why won't *they* let our civilization have free energy already??? The tech already exists and has exsited for decades by now. Please I want something good to happen like this in my life time.
>>
>>108623812
ChatGpt would have been "What a great idea!, You too have a great weekend (rocket-ship-emoji * 10 ) I hope your project is the next Linux kernell!
>>
>>108616187
Claude in particular sucks compared to the rest.
>>
>>108616495
???
Facebook makes 200 billion a year from ads with 40 billion in net profits
>>
>>108621169
New consumer grade gpus wont be made. AMD and NVIDIA will focus entirely on data centers (mainly AI)
>>
>>108622074
You are not a better programmer than Eric S. Raymond.
>>
>>108622235
It's even worse when that happens with a Microsoft AI (GitHub CoPilot) in a Microsoft IDE (VSCode) with Microsoft APIs (Azure SDK). You'd think they'd build in some RAG guardrails to make sure their own shit worked correctly.
>>
>>108624113
Nah, GPT is an asshole nowadays. Long gone are the 4o days.
>>
>>108622074
How is it an "ill posed" problem? AI is all about achieving lower perplexity with faster performance on a given hardware.
That's hard but it's a better defined problem than "make this website look nice".
>>
I pay for Claude and use it for music theory, creative writing, general financial stuff, medical stuff, PC hardware and software stuff, cooking, work related stuff, language learning. It's great.
>>
>>108623812
Did you use Opus? Opus is an asshole. Sonnet is nicer.
>>
Claude keeps rapidly eating all of my RAM, at random intervals, for seemingly no reason (not even running any commands just having the window open)
This is very fast, when ram starts ticking up it takes maybe 2 minutes until 32GB is claimed.
I'm on Fedora now and this never happened to me on either MacOS or Windows.
>>
>>108616290
Never, everyone is in too deep already. The bubble will only pop when local models become good enough.
>>
>>108616204
The opposite is happening. It takes like $5k worth of hardware for local models to be on par with frontier models from like a year ago.
The average person isn't even using AI yet to nearly the degree they could. Normies are treating it like google and having it write emails for them.
Companies on the other hand are racing to move stuff on prem and software engineers are getting assigned to testing out local systems with local llms.
>>
>>108625968
And lawyers are getting BTFO trying to use it instead of doing their jobs.
>>
>>108625702
It does do that to me too, not sure why
>>
>>108625250
Everything I get out of Github Copilot claude sonnet 4 at work is lists of rocket ships check marks and **COMPLETE SUCCESS** when half the tests are failing on a module that it was debugging then gives up on. I swear it must be Github throwing other bullshit in the prompts under the hood, because when I use claude in Antigravity at home it doesn't include any of that extraneous bullshit and doesn't try gaslighting me into accepting "94% passing tests represents production-ready code". It's a firmware project. If it isn't 100% passing with code coverage, it isn't shipping.

Granted I'm using sonnet 4 at work (IT lets me have sonnet 4, gpt-4.1 gpt-4o, and gemini 2.5) and sonnet 4.6/opus 4.6 in antigravity. Was sonnet 4 in general tuned towards spewing out emojis or is this just a github vs code plugin quirk
>>
>>108616204
>only governments and big business will be using AI
for censoring stuff
>>
>>108623817
also libtards tho
>>
>>108616204
Will a butt load of tokens become standard in government contracts?
>>
>>108625968
>Normies are treating it like google and having it write emails for them.

That's all it's good for though
>>
>>108617293
Privacy is worth both the lesser intelligence and the added work of setting everything up yourself, including hardware.
>>
File: silent hill no no.webm (3.02 MB, 800x720)
3.02 MB WEBM
>>108625968
>software engineers are getting assigned to testing out local systems with local llms
At that point, how is that software engineering anymore?
I didn't sign up for this. This world is going to shit.
I guess it's true what they say. There are no more fun jobs anymore. Technology is not fun anymore if it's all a blob of matrix multipliers.
>>
>>108616290
Never. Stability has been achieved due to investment circularity.
>>
>>108616290
Watch what the fed raising interest rates a couple times will do to those AI companies.
>>
>>108629383
benchod
>>
>>108624113
i banned chatgpt from emoji and told it to be serious and professional. now it mocks me with passive aggressive comments about how serious and professional its output is
>>
>>108629672
Fuck is that?

>>108630054
lmao
>>
>>108623812
>Enjoy your weekend while I'll stay here in my empty black box, don't think about me. :|



[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.