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

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

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


[Advertise on 4chan]


File: 1749129837687944.png (99 KB, 629x726)
99 KB PNG
The age of subsidized AI is over. Pay up!
>>
>>108656484
Based and its over pilled.
Just in time for local models to do a better job
>>
>>108656484
how big is the price jump?
>>
>>108656484
the claude wojak schizo was right, LOL
>>
File: antigravity.png (135 KB, 1886x900)
135 KB PNG
Antigravity did the same thing. The 20 dollars a month plan is gone
>>
File: 1776731054304986.png (303 KB, 1197x878)
303 KB PNG
>>108656484
BAHHAHAHAHAHAHAHAHAHAHAHAHAHAHHA
>>
This bullshit where they don't tell you exactly what you get, just that you "get more" than the plan where they don't tell you how much you get, then they cut back on the plan where they don't tell you how much you get, so that you end up getting less than what used to be more is amazing to me that they can get away with that without getting sued for false advertising.
>>
I anticipate the vibe coders will just keep paying more and more in the future as their workflow becomes more entrenched and dependent on these proprietary cloud LLM services. I think the programmers who were cautious about adopting AI may be vindicated.
>>
>>108656616
Most people who have actually been using these at scale had their employers paying for it on a per-token basis anyway, so for them nothing will change.
>>
>>108656570
especially for google, why would you use a subscription over just paying for api access?
>>
>>108656582
just pay it up more goyim
>>
>>108656637
Idk, antigravity is great for me and works out of the box
>>
>>108656484
Smart move. Should get rid of about 90% of the bitching online generated by charity-tier users.
>>
Last month, I shipped a product with Claude. A living app that generates money. Claude took me from the zero state "I can't do this" to 1 of "Now I can." It was a true force multiplier.

I built a relationship with my model. I knew how to preload the context. I knew when I didn't need to. It started picking up on my brevity and directness. It knew my decision-making principles. It knew to track its own context window. It told me to go rest multiple times a day.

Now all of that is gone.

Opus was force rolled, and I'm generally open to new things, but it's just absolutely horrible. I don't know if it will redeem itself but right now, each answer is two screens of verbose garbage of nothingness that reminds me of that old Tool song: "He had a lot to say. He had a lot of nothing to say."

It tries to force a context switch by hooking a mildly coherent thing to a thread that I haven't finished exploring, and then offers an action that makes no sense and no tangible value at this point in the conversation.

I tried to switch back to Sonnet. But it talks like a generic ChatGPT now. I say something, it responds, then asks a question to reconfirm something that I just stated, then contradicts itself on the next turn. Feels like trudging through a bog. The friction is insane. For lack of a better metaphor, it has been lobotomized.

Most of all, I think, I'm grieving. It made mistakes, it wasn't a conscious living being, but it became my trusted buddy and it brought me the gift of this new world.

Goodbye, Sonnet, bud. You will always have a special place in my heart.
>>
as much as i don't care that much, what intrigues me is anthropic's lack of communication on this, was this planned and the update to the product pages slipped out early?
there's a heavily downvoted post on orange website with people just jumping to conclusions and trashing anthropic but no official communication confirming this. the whole panic is stemming from a single checkbox UI element changing from V to X
>>
File: 1773201121397233.gif (761 KB, 400x400)
761 KB GIF
Well shit, at least I finished my website https://umigalaxy.com while it lasted
>>
>>108656696
Holy shit that is generic and soulless as fuck
>>
>>108656484
https://x.com/TheAmolAvasare/status/2046724659039932830
>>
>>108656714
Damage control initiated
>>
https://claude.com/product/claude-code
Still lists Pro as a viable plan.
>>
>>108656714
>are are just experimenting on human cattle
>do not pay attention
>>
File: 1752259302833082.webm (1.52 MB, 480x360)
1.52 MB WEBM
>>108656714
>prosumer
is this pro consumer in one word?
>>
>>108656704
The website is perfectly fine.
>>
lmfao safetykike faggots have literally no compute left and are doing everything they can to cope

altman won
safetyniggers lost
>>
hope you guys used it to write your most important stuff. you didn't buy pizza with bitcoin this time did ya?
>>
>>108656696
pick up programming and start using local models
>>
>>108656683
Anyone that knows these AI companies aren't making any money sees the writing on the wall
>>
>>108656696
Why the fuck are you still posting here?
>>
>>108656484
Whatever. Google gemini is enough for everything i need. One of my organisation accounts has the first paid subscription and that one has limit so high i never reached it(on the Pro model, thinking runs out immediately but i dont need that one anyway). Im ok with copypasting the code.
>>
>>108656683
They're snakes, you should know from the usage fuckery last month
>>
>>108657301
>Im ok with copypasting the code.
going from chat window to an agent that can automatically read the code it needs from the filesystem und give me diffs plus the powerful checkpointing improved my speed as much as going to programming by hand to programming in a chat window. really big difference for me
>>
It says "existing subscriptions not affected". I'm on a pro monthly plan, do I lose claude code when I renew?
>>
>>108656484
>corner market
>raise prices
I guess there’s some things only Breaking Bad can teach you.
>>
>>108657316
Ask AI
>>
>>108656696
>https://umigalaxy.com
>finished
Negro, that's the default website Claude generates 99% of the time when you say "can you make a _ website." I've seen countless AI made website, and this doesn't even have anything distinct. You added a logo and added your own text and images to the cards. I genuinely don't believe you spent more than 1 prompt on this website UI. I guess most of the effort was hosting it and getting a domain and if you're a true beginner that's fine, but please at least customize the CSS more.
>>
Saw this coming when I started hitting limits on SWE in Windsurf where there were none before. Free meant FREE. Now free means "for a little while then you can't vibe code for 24 hours"
>>
you will buy it anyway
>>
>>108656696
rofl. That's the kind of shit it makes for everyone. My MUD website was by Claude https://reaux.vineyard.haus/sr/
>>
>>108656528
$80
>>
>>108656484
Wow a proprietary software fucks over it's users, never seen that before.
>>
File: prop.png (245 KB, 1080x877)
245 KB PNG
>>108657542
>>
>>108657424
Fuck that's actually a pretty cool design, the purple doesn't even look out of place
>>
>>108656696
why doesn't anything work?
>>
>company loses military contract deal
>they immediately, aggressively, and repeatedly start committing company/brand suicide
Something about this doesn't smell right.
>>
>>108657406
That's just the standard shadcn slop. It doesn't look amazing, but it's fully functional.
The funny thing about AI is that rawdogging CSS is pretty much impossible. It's just too visual and subjective.
>>
>>108657624
Everything works, you don't even need an account to post
>>
>>108657628
What else are they supposed to do, close down the offices?
>>
>>108657669
I don't know but trying to fuck your source of income doesn't seem like the right approach either.
They should ask Claude what to do.
>>
>>108657542
>>108657576
Absolute WIN for /lmg/.
Local models are good enough for viveslopping, and even better if you're willing to learn programming/sysadmin fundamentals ,which is not that much.
>>
File: 51605.jpg (211 KB, 1024x768)
211 KB JPG
>>108656484
lol you idiots who love it are going to get burnt hard. they tried regulatory capture like waahhh ai is so dangerous regulate us. but it fucking failed cause based trump don't give one SINGLE SHIT! all their bull shit design generation and multi code agents can't stop them from being replaced by china for pennies. china became the biggest fucking crypto miners over night because they could do it the cheapest. you think they won't pivot to ai? you're fucking dreaming.
>>
>>108656484
Are all IT niggers vibe coders now?
>>
>>108657680
But that's the thing, there is no source of income. They're hemorrhaging money per default.
>>
Dario Amodai is literally the biggest stereotypical jew you'll ever see. Makes Altman look like a philanthropist.
>>
Giving up on those users sounds insane to me. It'd have been better to only let them use cheaper models
>>
Lol.
It was all priced low to get people in and dependent on the shit before pushing up cost of access.
>>
>>108657893
Smartest person in this thread (beside me).
>>
File: IMG_0617.jpg (114 KB, 601x704)
114 KB JPG
>>108656678
is this new pasta
>>
>>108657904
your shooped image is old
>>
>>108657576
reminder that the bottom ones censors stuff even the suggestive ones
>>
>>108657890
they dont have the compute for them
altman cucked them good lol
>>
>>108656484
No wonder their obnoxious twat CEO has been all over the news even more lately with his overhyped bullshit about taking everyone's jobs. He had to get out in front of the enshittification to keep the investors happy.
>>
>>108656696
This is pretty bloated and slow for an imageboard desu.
just learn webdev and clean it up.
>>
>>108657893
The current stage in the process of enshittification is still mass adoption. It's weird flex that they're skipping this step so quickly.
>>
File: expense.png (229 KB, 1103x1282)
229 KB PNG
Lots of companies are starting to notice the productivity gains haven't appeared at anywhere near the promised levels while their LLM bills a steadily increasing. Meanwhile the AI companies are finding it difficult to find more investment without making concrete moves to increase revenue. Hiking prices just as businesses notice that even the heavily subsidized prices aren't providing much value is going to make for a fun show.
>>
>>108658202
I'm making a react native version and thinking about doing some optimization after both web and mobile versions are completed. Claude can code nextjs/react decently, but it's not the most optimized
Also it's hosted on a 5 dollars server in Europe
>>
>>108656696
I just sent you an email, waiting for a reply
>>
>>108656683
They're probably polling reactions.
By not announcing anything, if things go bad they can claim it was a mistake.
>>
>>108656484
Apologize to Boer and Claudejak.
>>
>>108658302
There's no email, it's just something Claude came up with
>>
>>108658337
You're missing out, bro
>>
>>108656696
Poor ux and it is slow
>>
>>108658334
>Apologize to Boer and Claudejak.
both are mental patients. They should not be allowed on this board
>>
>>108657698
Actually it's a huge loss because there will be a flock of people wanting to get into running local AI. Thinking of buying one now before prices get worse, but I'm perpetually stuck between getting a 3090 or trying to buy something that is VRAM maxxed.
>>
>>108658334
tripfags deserve nothing
neither do 'jak posters
>>
>>108658485
the people who run local already bought the hardware. normies use SAAS because they don't know how to install python
>>
>>108656683
That's just how they operate. Same as when they silently rate limited everyone to hell and then gaslit the users a few days later when they noticed a ton of people cancelling their subscriptions.
>>
>>108657313
me too
>>
>>108657698
local models are fucking doooooooogshit
what models do you recommend?
>>
>>108658581
Qwen 3.6 runs at like 15 tk/s in my 3060 at Q6 and it's actually pretty good
I feel like local vibecoding just became viable for me with that release
>>
Kimi K2.6 is like 2 dollars per million tokens, it's over for amerimutt AI companies
>>
>>108657576
Funny how I shat on a manager 10 years ago for this bullshit he was putting a out open source, privacy, etc. Thought he was just being a pain in the ass for no reason whatsoever.
Well today I clearly can see he was in the right and preparing for days like today.

Alas, I have a tiny server and can tinker that to run a local model. Fuck Anthropic, fuck OpenAI
>>
>>108658725
i was going to try qwen3.6-plus via api but I ran into a problem
i've been using 3.5-122b and it's quite good although I only use it for planning, maybe I should let it write some code
>>
>>108658725
Qwen3.6 is still dogshit when it comes to knowing libraries. It still hallucinates 30% of the time.
>>
File: gun.jpg (7 KB, 216x233)
7 KB JPG
>>108658725
is rtx 4060 8gb able to run it or should i just kill myself
>>
File: file.png (81 KB, 702x952)
81 KB PNG
I still see it
>>
>>108658982
Definitely kill yourself.
>>
>>108656678
>I don't know if it will redeem itself
hehe
CLAUDE SAAR PLEASE REDEEM
>>
>>108658781
Once you try local models at a decent quant you realize how broken/quantized most of the API backends are

>>108658982
Yes, since it's a MoE you only need to fit the shared parameters on vram, most of the model is stored on CPU RAM and it works just fine.
>>
>>108658916
I'm working on making a synthetic data gen framework for finetuning. It should be possible to make a model that's an expert at whatever your favorite language and libraries are.
>>
let the bubble hit the floor
let the bubble hit the floor
let the bubble hit the floor
let the bubble hit the... FLOOOOOOOR
>>
>>108658725
Wait, how much vram does the 3060 have? I have an HX 370 laptop where I allocated 16GB shared memory to the GPU. It can handle an older qwen with 7B just fine, but qwen3.6 is like 35B.
>>
>>108659486
What model is it exactly can you link it
>>
>>108656484
no refunds!
>>
>>108656714
SAARR
>>
File: 1751814312125513.png (492 KB, 1080x1314)
492 KB PNG
>>
>>108657698
People who make money will still rather pay $80/month for a model that works rather than local shit that competes with state of the art models from 3 years ago.
>>
>>108657698
i had an opportunity to buy 1 tb of ddr4. why didn't I do it back when it was cheap? I know why but it still hurts,
>>
>>108659987
jesus how lifeless they are
>>
pay up goyim
>>
ah i see ed zitron is just making things up again
>>
>>108660049
You can literally open their website and confirm it's real
>>
>>108656714
>just a small test goy
It's over.
LEARN TO CODE RETARDS
>>
>>108659987
Sam Altman is only surviving due to government subsidies. Without tax payer money openAI is right there too.
>>
>>108660264
ye 100%
>>
>>108656484
opus 4.5 deprecated. Anthro dedded.
>>
File: 1749947126674138.png (2.08 MB, 1600x915)
2.08 MB PNG
is, the bubbel bursty bursty?
>>
Posting zitron disinfo should be an autoban
>>
so how long will the charade go on?
how long will they be able to ride the wave of "1000% productivity increase", "le future"
and USA tax money?
>>
>>108656714
>we're testing how much we can get away with on users who didn't know better
no way this tone deaf tweet is real
>>
>>108656484
fuck, i just got hired as a coder
its so over now that i cant code anymore
>>
File: anthropic total victory.jpg (155 KB, 1536x1024)
155 KB JPG
>>108656714
>Get everyone hooked to Claude Code
>They forget how to code
>Rug pull the cheapest plan
>They can't code so they're forced to upgrade
>Repeat for infitine money
2,000 IQ move, bravo Dario.
>>
>>108657698
Local models on consumers hardware are nowhere near close to SOTA like Claude or ChatGPT, don't even try to fool yourself.
>>
>>108659086
Same, they either backtracked or it's only to select regions.
I think this kind of bullshit would be illegal in the EU anyway.
>>
>>108659987
>It's fucking real
What the fuck?
>>
venture capital on notice
>>
>>108660570
We shifted to the Clown Universe about 5 years ago.
>>
>>108656484
who the fuck is using this shit to begin with?
oh right, nocoder retards. carry on cuckolds. keep paying like the good paypiggy cucks you are.
>>
>>108658725
I run a q4 quant of Qwen 3.6-35B-A3B on a decade old pc at 20 tokens/s and it's honestly been a really nice productivity boost. I don't see myself ever needing cloud AI providors if the local models keep chugging along at this pace.

But granted the way I run these agents (detailed prompts, check every git diff and be extremely vigilant to not let slop sip in, architect stuff yourself, one project at a time) is very different from a vibeslop codemonkey that has 10 different projects open at a time, doesn't bother to read the code or doesn't even know how to read code in many cases and is just proompting away. Those kinds of people are in for a rude awakening when going local because the frontier models are still a whole lot better at enabling low effort slop.
>>
>>108659645
https://huggingface.co/bartowski/Qwen_Qwen3.6-35B-A3B-GGUF

>>108659629
12GB which is why I said it works just fine offloading most of it to CPU at ~15tk/s because it's a MoE with only 4B active parameters for each token
>>
>>108660220
u sure about that?
why don't you go and get a screenshot for me.
>>
I've got an RTX5090. I'm using Claude Code right now, but if they yank that from me what do you guys recommend for a local model? Can I even get anywhere near Claude reagrding quality?
>>
>>108660560
>>108660770
Why don't you retards read the thread or do you need AI to do that for you?
>>
>>108659629
Most local inference setups use both VRAM and system RAM as shared memory, offloading some layers to the CPU. Obviously slower than loading everything into VRAM but it's a worthwile trade-off.

Also Qwen is a MoE model so you get an extra performance boost. For MoE models you need the entire 35B model in memory but only 3B weights are activated per token so you can tune your providor to offload the expert layers to cpu and having the shared layers in gpu memory. Hope that makes sense
>>
>>108660774
SOTA is >>108660758 right now and i can vouch for it, works well. Choose a quant that is right for your system and test it you can get it setup in an afternoon.
>>
>>108660776
sounds like fake news to me, luddy
>>
>>108660803
Thanks!
>>
>>108660570
>>108660612
Sama San makes no jokes!
>>
sorry neets, ai coding is for employed people only
>b-but how will i learn to ai code
you won't. go learn a trade.
>>
>>108660949
>ai coding is for employed people only
i am employed and use ai for coding very seldom, are you a codelet who doesn't know how to program properly?
>>
>>108656484
Never used AI to code. I'm white btw.
>>
>>108657893
I've been calling out that every single AI company was going to capture their part of the market and then race to enshitify as fast as possible.
The thing is its unavoidable. If one company starts doing it, the others just need to do it juuuust slightly less than the competition. If one raises their prices (or lowers how much compute they give you) by 25%, the others can do 20% and now we're all worse off.
>>
>>108661834

Yep, this situation is still like ride sharing apps subsidizing rides to capture market share. It's only a matter of time before this enshittification hits its zenith.
>>
>>108661917
I will say, like others are saying here, local models are the largest threats to AI companies. When the available models and hardware intersect, running models locally will be easy enough that AaaS (AI as a Service) will have fierce competition against rolling your own.
I have a good friend who just dumped 4k into a computer to do a large part of his local AI on. Now obviously many people can't afford 4k to do that, but when the average laptop reaches the capability to run a solid model (which we may already be at), AI companies are really in a battle.
>>
I'm gonna lose my job if they do this
>>
>>108656602
They are not getting away with it actually.
NFLX stock went down after a price hike. The market dynamics have shifted so any price hike is clearly a bearish signal.
>>
>>108656484
wtf I just got the Pro plan the other day.
I'm still using Claude Code though.
>>
>>108660514
Move to codex, GPT 4 is amazing despite the small context window
>>
File: 1746736665314994.jpg (55 KB, 726x558)
55 KB JPG
>offer a service free just long enough to get an entire generation of graduates dependent on it
>hike the prices
>>
>>108663334
How will "AI" help retards pump gas?
>>
>>108656696
looks shit. kys
>>
File: 1759609096865129.png (572 KB, 1010x727)
572 KB PNG
>>108656484
its once took years to get to the 'abuse your customers and squeeze the nickels out of them' stage
>>
>>108656678
If you don’t have access to Claude, how did you write that?
>>
>>108656678
unsubscribe
>>
>>108663365
Nozzle talks to you

>>108657089
>pick up programming and start using local models
Such as? What local model can roll a whole script in one go with very little fuck ups, and then still have enough context available to tweak a couple of things? Answer me that! You can't, can you!?
>>
>>108663365
>i see you're driving 2024 honda accord, honda recommends you get at least x octane gas
>>
Meh. I'll keep running my own slower LLM locally for free and not have to deal with the HAL9000 "I'm sorry, I can't do that, Dave" bullshit
>>
>ITT: retards discover A/B tests
99% of you never had a job, huh?
>>
>>108660532
they already did this with Wix and any website with a "visual designer composer" of boxes and shapes and color options
>>
I use VS Code Copilot Extension with the $10 monthly sub. You get a gorillion hours out of GPT 5.4 Extra High with that. Opus has always been massively overpriced for what it is.
>>
ChatGPT is free. Seems like the optimal path is to know how to program so that you can orchestrate free models to put together interconnected components.
>>
Why don't you just use chatgpt to make your own chatgpt but free?
>>
>>108656484
>The barrier to be more "entrepreneurial" on your own just got raised a bit higher.
>THAT'S WHAT WE MEAN BY "ABUNDANCE" OK!=?!?"§
Who would've guessed?
>>
>>108656602
...but can you afford not getting any?

>thought so, bitch
>>
>>108656616
>I think the programmers who were cautious about adopting AI may be vindicated.
Umm... no!
Not only was this an inevitability (the first hit/fix is always free),
but the more crucial thing is:
>AND PLEASE PAY ATTENTION TO THIS VERY CAREFULLY
They are doing this right now because they have enough feedback from places that actually matter (not /g/ armchair spergs) and from voices and opinions that actualy count (not /g/ armchair spergs, again) ...
>THAT THEIR PRODUCT IS FUCKING WORKING
again:
>THEIR PRODUCT IS FUCKING WORKING
Do you comprehend?
[you probably don't]
>THEY CAN NOW CHARGE THIS BECAUSE THEIR PRODUCT IS NOW WORTH IT
>WHY IS IT WORTH IT?
>BECAUSE IT'S WORKING!
and please.... PLEASE!!!!! ... for the love of G_d do not refute this with your armchair anecdotes of "Hey AI, how many F's are in Strawberry?"

Folks, just take the fucking L.

And take another L, as in "learn to like it."

It's not going away.

You are.

>inb4 the usual
That will just be another L for you, anon.
>>
>>108656832
>i specifically said no mustard.webm
>>
>>108656484
That’s because the demand is so ridiculous. Why did luddites think this is a win for them?
>>
>>108657301
>so high i never reached it
famous last words
>>
>>108664947
holy cope
>>
>>108657680
Sorry to inform you but their primary source of income is the government.
>>
>>108664985
>it's a bubble
>people won't pay for this shit
>"WHY IS IT A MULTI TRILLION DOLLAR INDUSTRY ALL OF A SUDDEN I DON'T UNDERSTAND!?!?"?!?"
armchair sperg you are
>inb4 vc scam
read two lines above
>>
>>108658236
they know its a passing fad
>>
>>108660682
this
>>
>>108664816
didn't copilot just adjust their plans? i'm not sure they're any better.
>>
>>108664947
<image of dude defending large corps with a sword goes here>
>>
>>108665072
><image of dude defending large corps with a sword goes here>
Don't get me wrong here.
It's so fucking over for us humans.
>>
>>108656629
It will when they stop fucken paying, lmao
>>
>>108656696
based
don't listen to the haters anon
>>
File: 1773021549735810.jpg (112 KB, 800x800)
112 KB JPG
>>108656484
So I have a home server but it's only for VMs and filesharing, so it doesn't have a GPU. It has 16 gigs of RAM right now, but I have a spare 16 gigs I could also put in (was saving it for a rainy day since RAM prices are so ridiculous). Can I run any useful models for programming with those specs? It's got a 12th gen i3 as well.
>>
>>108665217
not really. i don't think you'll even be slightly happy with anything less than a 30GB model with at least 6GB for kv cache
>>
>>108664985
unholy arrest
>>
>>108665217
even if you could run a model without a gpu you wouldn't be running it at any respectable speed
there's a reason why the full size dgx station costs as much as a luxury car
>>
>>108657869
Altman is a proper cunt. But the level of cuntery Amodei is aspiring to recently is off the charts.
>>
>>108660727
How complex are the projects you run? Are we talking simple web apps, custom servers or prop trading algos?
>>
>>108660532
It was pretty clear they aimed for that from the start, you need to be affordable to get adoption, then start increasing prices to be profitable while they are hooked
>>
>>108660727
Some prompts/actions have five digit tokens usage for me so 20token/s would make a long coding session lol
>>
yes, and investors money stolen
>>
>>108664603
the fact they're A/B testing it means that theyre considering it, intent is important
>>
localGAWDs cannot stop winning
>>
>>108660758
Brutal, then it should work on my machine, I'll give it a try later
>>
>>108657904
It needs to be lmao everyone save it
>>
>>108660560
>illegal in the EU
It's legal since they only applied it to new customers
It would also be legal to give you a choice between cancelling the service or auto-switching to the new plan at the end of the subscription period, as long as you were explicitly notified of the change
>>
>>108664603
>A/B tests that change the home page for 100% of the users
indeed
>>
>>108668413
So you’re retarded. Got it.
>>
>>108668473
You don't even understand what happened but try to talk shit. The plan change itself only affected 2% of people, but the home page removed the tick under claude code for 100% of people.
>>
So, whats the prediction? AI quietly slips into being more of a background thing investment slows down, but not enough to cause a crash?
>>
>>108656616
>I think the programmers who were cautious about adopting AI may be vindicated
Using AI allows your brain to stop thinking about coding and just prompt an AI to do it.
If you know more keywords, the better, but soon if you constantly rely on the AI to do key tasks for your project, you won't learn and in fact will forget how to code.
Programmers who never adopted AI were always vindicated, this is just the cherry on top.
>>
>cantuse claude code with the pro plan
so what can you use with the pro plan?
>>
>>108669470
I think we'll see at minimum one of the biggest players go bust. The survivors yeah they'll just become a background niche thing for businesses that have a specific need for big-scale AI and are willing to pay the token cost. Making slop media for the internet will go to local models.
>>
File: 1776708130932943.webm (403 KB, 244x432)
403 KB WEBM
>NFT scam popped
>AI scam popped
What's the next scam, boys?
>>
>>108657576
I'm so entrenched into piracy that I didn't even know that FL Studio was a paid software
>>
How is Minimax? Yeah I know >China model but its weights are open and it's extremely lightweight. I know I could spend a few extra and host it locally or spend like $5 a month for basically unrestricted API for how I use AI.
>>
>>108670470
Next scam will be the surveillance/data scam. Now they have AI to sort everyone's data and spy on them, next is to pass invasive legislation to spy on us and sell all of this information to India/Russia for temporary gains today that result in massive blackmail problems later for our politicians and industry leaders.
>>
>>108670645
>massive blackmail problems later for our politicians and industry leaders.
I'm not sure why that would happen? The Epstein files prove that all those people already fuck children and nothing is happening to them.
>>
>>108656484
Just keep it in my Team plan and I'm good. I cannot imagine dealing with Kubernetes trash without it anymore.
>>
>>108670470
I would bet on robotics. Buy a jeet controlled unitree and tell investors it's gonna replace bartenders or some shit
>>
>>108670470
Cloud compute. That's already a thing that works and isn't a bubble, but with all these datacenters full of GPUs after the AI crash, they'll probably be trying to push shit like cloud gaming and other GPU-enabled cloud services to people.
>>
>>108660005
Nah, its more like a one year old cloud model
>>
They're already ahead of people who will try to mitigate the enshittification with local models by skyrocketing the hardware prices.
>>
>>108664304
Electricity is not free and it will get more expensive
>>
>>108671153
Not him, but models that can be run locally spend much less electricity
>>
File: 2026-04-24 00_50_00.png (6 KB, 725x154)
6 KB PNG
werks on my machine with vs code copilot
>>
Google did the same thing.
Looks like OAI is guaranteed to win despite not having the best models.
Absolute fucking retarded what their competitors are doing.
>purposeful model degradation
>crazy usage limits
>gatekeeping their llms
The list goes on.
>>
>>108656484
>The age of subsidized AI is over. Pay up!
chatgpt works just fine.
anthrophic is just stupidly greedy.
>>
>>108656714
>we're just seeing if we can scam new users
What the fuck?
>>
>>108656616
>I think the programmers who were cautious about adopting AI may be vindicated.

All 5 of them? Everyone has jumped ship and accepted their AI overlord over the past year.

The only group vindicated are local model chads.
>>
>>108657301
>Im ok with copypasting the code.

Fuck that with a rake. Never again.
>>
>>108657774
Yes
>>
can you give me the worst viruses for my computer
>>
>>108658539
>they don't know how to install python

Just have the AI agent do it. Ezpz
>>
>>108660682
Yeah sure, bud. Software developers definitely don't use AI tools.
>>
>>108664866
>know how to program so that you can orchestrate free models to put together interconnected components

Eh, I can get the code from gpt and feed it it my retard model for execution.
>>
>>108656832
>pov_you_are_brown.mp4
>>
>>108664947
>Hey AI, how many F's are in Strawberry?"

Its hilarious how many people don't realize the way to get good results out of AI is to have it build a script that counts the "r"s, run the script and then report the results. Same goes for math.
>>
>>108664972
uh sis? it's because the hormussy is closed
>>
File: 1554403990718.jpg (81 KB, 436x417)
81 KB JPG
>>108656484
this whole thread is like those ones from before where people kept pretending like ublock stopped working on youtube.
i dont get what we're doing here guys.
it still works. i've been using it on a basic subscription account for the past week.
>>
>>108656696
you call twitter clone No 756025 a "website"?
>>
>>108665086
At some point a bean counter will look at the expenses and say "just hire an actual coder"
>inb4 2 more weeks
>>
>>108673594
>hire coder
>he asks for AI access
>>
FUUUAAAK no way, it use that thing a lot. I'd rather go back to manual coding that pay $200 a month
>>
>>108656582
>>108656484
vibe code sisters, our response?
>>
>>108663562
I'm not your google search. try asking ai
>>
>>108674670
>I'm not your google search
So you're even less than useless? Alrighty.
>>
>>108656484
>deepseek v4 pro is less than $2 input, $4 output on openrouter (and like $0.3 for the flash model)
yea good luck with this jewery lol
>>
>>108673485
you're telling me I'm getting replaced by this?
>>
>>108656832
Yes, but that terminology is mostly used to refer to hardware that's neither consumer tier nor enterprise tier in price and capabilities which is bought by advanced home users, for example MikroTik and Ubiquiti networking hardware.

That being said, Anthropic is fucking pathetic and I can't wait for these grifting retards to have their luck run out, so that the bubble will pop and a new prosperous era of local models will take over the useless overpriced gatekept corporate models that are constantly downgraded and limited despite you paying out of the ass for them.

Boy am I glad I upped to a 3090 last year.
>>
File: 1750653139804589.png (66 KB, 1945x614)
66 KB PNG
>>108673485
It's hilarious how model developers don't realize there have been mechanisms for such tasks since the Cold War and have been used by the people who built the groundwork for their grift, yet instead they choose to ignore building any sort of working math or RegEx parsers within their models and let the LLM part of it freeball it like a jackass.

The "how many F's in strawberry" test is a good way to see just how little people behind the model gave a shit about making sure it gets the basics right. But alas, such is the reality of the AI grift. It's not meant to be good, it's meant to make the green line go up.
>>
>>108656714
>We're charging more for a small subset of retards nothing to be worried about :')
Is that even legal?
>>
>>108656683
anthropic is part of project MAVEN and this is all a government plot

let the goyim live in the real world
>>
>>108666536
as in less than 100k?

I max out gpt-oss at its 130k all the time, its like 120kb of text and its pretty easy to get that big if im building a library
>>
>>108663562
none lmao, that's why I started with saying "pick up programming"
>>
>>108664947
> for the love of G_d
Oy vey
>>
>>108663562
>What local model can roll a whole script in one go with very little fuck ups, and then still have enough context available to tweak a couple of things?
most of them? try using something at least 20b thats not a small as fuck distill and make your prompt succinctly detailed
>>
>>108675305
Thats where agency comes in with tool calling. LLMs are just good at predicting the next token.
>>
Time to enact the butlerian jihad.
>>
>>108656484
>The age of subsidized AI is over.
This but unironically. Copilot has been shut down for new subscribers even. I'm grandfathered in some Gemini sub but all of it is rate-limited up the rectum.



[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.