>>108885897
That's Microsoft right now. I suspect after the market corrects and the cost for ai usage goes up across the board, companies will come to this realization as well.

>>108886142
>the market corrects
lol
the market won't correct itself
the market will fail so hard that the government has to step in to bail it out again

>>108885897
Just ask AI to make itself cheaper

File: 1768997766475116.png (229 KB, 462x410)

229 KB PNG

>>108886153
cool it with the antisemitic remarks.

Their entire business model is based off of this assumption that the training data will be free and plentiful forever.
Their business model fucks up the premise for having an open internet in the first place. Why maintain a website if there's no ad revenue and some shitter chatbot will just steal out the innards?
Okay, so now if the incentive structure is fucked, why have websites in the first place? Where's the money? If nobody is looking at search ads, there's no economic incentive to build a search index, now there's no more training data.
Then there's the whole intellectual property problem which isn't going to go away, politically speaking.

It's extremely socially disruptive and fucks up so many incentive structures. There's going to be a shit ton of backlash against it after the next electoral cycle.

And don't get me started on all these middle manager cunts running around screaming "LAYOFFS, REORGS, LAYOFFS, AI!!!!!!!!!111!!1!", I want to fucking kms. We're going to see a bunch of companies keel over because of this shit. Intuit just did a layoff and had the stock drop 30% because who needs their shitass product in the first place now that we have AI to summarize tax rules

>>108886153
Microsoft knows they're too big to fail as they're used across the entire gubberment, so they can take risks.

>>108885897
Snailcat won.
AI chuds BTFO

everything Microslop touches turns to shit. this is just failure #281.
they are literally the yardstick of what not to touch

File: Screenshot_20260523_06082(...).png (139 KB, 1080x942)

139 KB PNG

>>108886153
It already corrected 3 times the last 5 years. Stay poor

OH MAN if only they let us pay them to make unmoderated videos or something....

in my eyes this is not good news, if the only distinguishing factor a human has over AI is that they're cheaper in the current moment than its already over

File: HIzSeMwXoAAQLBz.png (95 KB, 1141x1000)

95 KB PNG

>Google released Gemini 3.5 Flash this week, and the cheaper, faster model now costs 5.5 times more to run than its predecessor. Token prices tripled to $1.50 per million input and $9.00 per million output, and on agent tasks it burns through so many tokens that total costs end up 75% higher than Gemini 3.1 Pro

Whops, test time scaling isn't a free lunch either when you have to pay for what you use.

>>108887264
Don't humans run on food and water? That "AI" craze is retarded.

Trump will save the job market, trust the plan.

>>108887293
For his buddies, not you.

>>108886153
Please let's all accept the fact this a built-in feature, not some emergency mechanism. It's basically "Print, Pump, Dump on peasants"

File: ezgif-7cfc323363713314.jpg (77 KB, 848x481)

77 KB JPG

>>108885897
> Using the tech is more expensive than paying human employees
Complete a bullshit. A robot doesn't take a 1 hour bathroom break or call in sick to work.
You're getting replaced you stupid luddite.

>more expensive than paying human
I'm sorry but I can't believe this. Especially not for companies that already have the hardware.

>>108887353
>breaks
>explodes
>kills someone
>cat be repaired

>>108885897
I'm telling ya.
This year.
The bubble will pop.
We're yet to hit the half-point of the year and we're already having some of the biggest CEO's in the space telling the truth of the financial viability of the tech. This grift won't last 'till 2027, investors will pull out in Q3-Q4 2026 and crash the market. mark my words.

>>108886171
>cool it with the antisemitic remarks.
Hello, muslim brotherhood

>>108885897
just ask ai to make more money

shrimple as

>>108888081
they will dump their bags on retail investors after ipo, then we'll see more critical articles in the media

>>108886504
Back in my days, people made websites to share things they are passionate about, not to make a dime. I don't know why zoomers are like you, but the only thing you ever think about every moment of the day is how much money you can make at any given moment with every action.

>>108887235
It's not the only factor. Humans are also currently between 4x and 400x faster to perform the task, and AI is limited to performing the kind of tasks you would normally hire pajeet in Pune to do knowing full well the result will be so bad that you wouldn't hire pajeet as an intern if he was slopping out the same quality of work. Meanwhile the human they use as the basis for cost is a principal engineer in the bay area, not pajeet in Pune.

>>108885897
they'll just make it run cheaper with specialized hardware. we'll have frontier AI model ASICs in a couple of months.
have no illusions: even if AI will fail as a whole, damage to software engineering as an occupation is already done an irreversible. even if these models won't get any better, they are already too good.

>>108887353
The human processed more packages than the bot despite taking breaks. Furthermore, bots require maintenance. This is also a task that is so simple that no human actually performs it (it's not sorting or anything, just flipping the package if the label is not right-side-up). We have had sensors and machines that do that without muh AI for many decades. This test is really impressive by the speed and accuracy of the small manipulation performed by the hands (assuming it's real and wasn't remote-controlled). But this is just a PoC. This is like the first neural network beating MNIST. It's really cool, but it's not really useful for much (worse, beating MNIST enabled banks to auto-read your cheques, this doesn't actually have a use yet). We have to wait for more technical improvements before it can actually automate things that truly currently require humans. I am hoping this will work out way better than LLMs, but given the hype is picking up here, I doubt it. The problem with hype is that it ensures small visible things get funded, instead of the research required to make things work, and the funding direction is such that real progress gets completely starved of funding, rather than merely staying at the same funding level. AI winter is coming.

>>108888343
They already use specialized hardware, retard. They have for years. That was particularly Google's main value claim as they build their own TPUs optimized for this (yet have the worst limits of any offering on top of now being more expensive for the same performance level than peers).

>>108888358
not nearly specialized enough. I mean weights burned in silicon.

>>108888374
This would require efficient at-scale chip printing at a time where model weights change dramatically every few months. Beyond the fact that all the fab capacity in the world would not be enough for what they need, only TSMC has the tech and knowhow to do this at presnt.
There is one company who's claiming to do something like this, they have an ai-generated landing page and had an ai demo. It was fast but it was gpt-2 tier of bad so it's probably a scam. Otherwise there is also Etched. But those aren't new and yet they're getting nowhere. They're also hardly the only ones in the space, be it straight to llm weights or more generic chips like those of tenstorrent or cerebras or groq or something.

>>108887401
We already knew it's more expensive, both Nvidia's deep learning VP and Uber's CTO said so, like last month.

>>108888416
I am no expert on the topic, but maybe these chips don't require the highest resolution process, and could be pumped out in large quantities on older fab lines. weights changing are no problem. just release a new model every quarter, and run it at 1% cost of what we do now. newer model = higher price. bleeding edge models could be still be run on GPUs and TPUs for much much higher price. they will solve it.

>>108888460
For obvious reasons, there's no demand for optimizing the older models. This is a massive problem: you can't account the chips as depreciation if they depreciate in 6 fucking months. It doesn't matter if you need a 2nm process or a 24nm process, the capacity requirement takes months to start up the process with a new design assuming the factory is at 0 capacity (if it's already printing shit, you have to share or wait your turn). The cost structure is essentially very heavily front-loaded, it's only if you serve for like 10 years that it'll start be a marginal reduction in price (i.e. gain from faster/cheaper inference vs cost of buying the chip).
You obviously don't know anything about chip chain of supply, design or manufacturing, which is your first mistake.

>>108887401
Nah, it'll go the other way. The software will get more specialized so that you don't NEED insane hardware to train and run it. Large general models will get replaced by specialist models. You can kind of already see this happening with Claude vs ChatGPT for programming.

>>108888481
ok, so they will release in-silicon Opus 4.6, which is already more than enough, and run it for the next 50 years. problem solved. it still wrecks the entire swe job market.
also, I call bullshit on your overly pessimistic assumptions regarding production of new silicon. even shitty small companies like SiFive are able to produce new chips in 2 years with external fabs. no way companies like Google or Microsoft won't be able to do it faster with their infinite resources.

>>108885897
they should start by fixing github's uptime and gaping security holes, and after that idc they can play with "a" "i"

>>108888547
>ok, so they will release in-silicon Opus 4.6, which is already more than enough, and run it for the next 50 years. problem solved
Opus 4.6 is already worse than the competition, and the competition will continue to move on. Obviously that doesn't work. Also reminder that the economics of serving these models breaks down if usage goes down.

>it still wrecks the entire swe job market.
Meds, now.

>small companies like SiFive are able to produce new chips in 2 years with external fabs. no way companies like Google or Microsoft won't be able to do it faster with their infinite resources.
Sure bro they can just click their fingers and mega factories will come out of the ground just like that, fully staffed and equipped with ASML equipment who will of course agree to furnish infinite machines for free and instantly.

Get a brain retard.

>>108885897
Fun thing about ai coding tools: if you pay for tokens and don’t quite get what you want, if you don’t know how to code…you have to continue to pay to get your result because you’re not going to get it any other way.

>>108885897
fake news.
there are no such microsoft reports.

>>108889406
wrong, software contractors exist

File: copium.delivery.jpg (464 KB, 1800x1801)

464 KB JPG

>>108888578
>Opus 4.6 is already worse than the competition
not if it still gets the job done, and can be offered at lower price
>Meds, now.
I got you covered bro
>Sure bro they can just click their fingers and [shit I never suggested will happen]
ok

>>108885897
Not what happened, you obsolete lying ape dumbass, they just wanted their wageslaves to use their garbage cli instead of claude code

>>108888350
>The human processed more packages than the bot despite taking breaks.
Not by much and it's moot anyway since the robot will always stay and work overtime. The human doesn't or gets burned out.

> Furthermore, bots require maintenance.
And? Just one engineer is needed to watch them until the bots start fixing each other.

> We have had sensors and machines that do that without muh AI for many decades.
You're missing the point. If a robot can do assembly line work then it can also scale and do anything else on the factory floor because it's not bolted to the ground. The robot sweeping the floor or or doing paperwork in addition to moving boxes is already immense value.

>We have to wait for more technical improvements before it can actually automate things that truly currently require humans.
They're already being used in factories with mass deployment coming before 2030. You couldn't be more wrong.

>I am hoping this will work out way better than LLMs, but given the hype is picking up here, I doubt it.
LLM's have had the biggest impact since the fucking industrial revolution. Why do you luddites keep pretending it's not?

>AI winter is coming.
Cope.

>>108890924
The human took breaks and didn't work overtime, yet still managed to do more than the constantly working robot, that is a fuckhuge massive problem for the robot. We are a century short of robots being able to fix themselves. Base servicing usually takes just one technician, but a full team must be available to actually fix them. This is a routine operation. We've had robots doing assembly line work in forever. This robot is not doing anything we haven't done for decades in an automated manner. They aren't used anywhere, they're prototypes, and nowhere near ready as demonstrated here. LLMs have yet to have any impact at all. AI winter demonstrated by (You).

>>108890967
>The human took breaks and didn't work overtime, yet still managed to do more than the constantly working robot,
Why do you keep repeating this when I already said the robot can work 24 hours+?
Do you think humans work in their fucking sleep? That's lost productivity.

> We are a century short of robots being able to fix themselves.
And airplanes will never get off the ground in a thousands of years. Now we have jets that break the speed of sound in less than a century.

>We've had robots doing assembly line work in forever.
Not bipedal ones but keep coping faggot.

> LLMs have yet to have any impact at all.
Humans are being mass fired by the THOUSANDS. It's just the beginning.

>>108890987
The robot needs 24 hours to do the work of a human in 7 effective hours.
It's not rocket science. Didn't bother reading the rest. As usual, the only aitards are clinically retarded. Proven right there.

>>108885897
It doesn't matter. The point is to cull the goyim.

File: Untitled.png (1.06 MB, 1332x753)

1.06 MB PNG

>>108891002
>The robot needs 24 hours to do the work of a human in 7 effective hours.
You lying faggot, they were practically neck and neck.
Being off by just 200 packages is already solved when the robot now works overtime forever.
You are being replaced.

>>108888416
>This would require efficient at-scale chip printing at a time where model weights change dramatically every few months. Beyond the fact that all the fab capacity in the world would not be enough for what they need, only TSMC has the tech and knowhow to do this at presnt.
The core problem is that training and inference are still far too expensive. The whole algorithmic design is just horrible that way. Faster hardware gets you some gains, but that doesn't help too much when the algorithm is quadratic or worse (and the size of the problem space we want to tackle is so huge).
There are alternative learning algorithms that can learn live, in real time, and that use a lot less energy. They've been demonstrated and published. But "frontier AI labs" are actually very poor at understanding the wider literature, and are only really at the frontier of wasting time and money, and the hardware concerned has got nothing to do with nVidia's GPUs.

>>108885897
>He still believes the layoffs are because of AI myth
HAHAHAHA

>>108891043
No, there exists no online learning algorithm that are efficient in either time or data. There have been tons of alternative algorithms attempted, but there has never been any that have been shown to work except on toy examples.
That said this isn't really the problem because "we" "don't care" "about online learning" since "we" can't handle distribution drift and poisoning yet. You are right that the efficiency of the training is a problem. As for the efficiency of inference, that's more an innate property of deep learning algorithms. It was the main argument against them on edge devices in the past decade before LLMs as well.

>>108891034
That clanker is making mistakes all the time. Any corporation using clankers or AI that make mistakes must be fined into bankruptcy.

I bet most companies would prefer paying more over having to deal with humans

File: 1759477956175586.jpg (59 KB, 463x600)

59 KB JPG

>>cool it with the antisemitic remarks.
>Hello, muslim brotherhood

File: 1778946543072836.png (136 KB, 728x514)

136 KB PNG

>>108885897
This doesn't mean AI will die, it just means using local models will become the norm. No shit using the most expensive models will be expensive as fuck. But you can buy the hardware to run Qwen locally right now, host it yourself, and only pay the electricity fee going forward.

The future isn't AI going away, it's local and open weight models tailored for your own company and your own usecase. This is literally what Microsoft is doing in your article. They're ditching the more expensive, third party Claude for their less expensive, in house solution. They're not stopping the usage of AI.

>>108887235
That isn't the only distinguishing factor, it's just the one that's being highlighted because the entire premise of AI was that it was supposed to replace your workforce for less money.

>>108886504
>There's going to be a shit ton of backlash against it after the next electoral cycle.
If anything, the opposite. The Democrats realize they made a huge mistake by positioning itself against the tech elite to satisfy art hoes and all the malding retards who were against AI due to pattisanship switches.

>>108891171
>picrel is real
that's just pure pottery

>>108891174
This doesn't mean Blockchain will die, it just means using local models will become the norm.

>>108891271
Blockchain is already local...

>>108891271
The blockchain is some meme storage and transfer system. LLMs write complicated working code in seconds. You are not doing an apples to apples comparison here.

>>108891332
LLM is some meme autocomplete system. Blockchains foster complicated security and scalability enterprise solutions. You are not doing an apples to apples comparison here.

File: a-droid-story-c3po-r2d2.jpg (860 KB, 1275x717)

860 KB JPG

>>108891332
> LLMs write complicated working code in seconds.
Not only that, it's literally robot technology.
Science fiction has always portrayed AI as being a tool or functional, so how did we get gaslit into believing it's the same as fucking blockchain? Luddites are so full of shit.

File: maxresdefault.jpg (151 KB, 1280x720)

151 KB JPG

>>108891364
>LLM is some meme autocomplete system.
A meme that can take your job apparently.
Just like the internet "a series of tubes" could still put Blockbuster out of business.

>>108891204
the democrats are almost always pro more govt involvement. This type of copyright dispute where economic incentives start blowing up is a perfect opportunity for regulators to come in

>>108888313
I'm not talking about me, I'm talking about economic incentives at scale

>>108891385
The internet didn't bankrupt blockbuster (virtually the contrary, it is failure to use the internet that dented them) and LLMs have not taken any jobs.

>>108891385
Blockchain took so many jobs.
Just like NFTs could still put T-bills out of market.

>>108885897
The more you CHUDS stall the advancement of AI the more expensive it becomes to use! Not supporting AI is economic terrorism!

>>108891946
>economic terrorism
that's basically me every time I do groceries

>>108887281
Yes, but more skilled humans don't require more food and water to run.

https://www.youtube.com/watch?v=O0VLtQzKlMI

>>108887353
Lets imagine robots replace the cart pushers, cashiers and factory workers. What will millions of people with no job going to do? They'll just sit at home and watch their family starve to death?

>>108888034
>>still describing humans

>>108887281
False equivalence. Whether you're a waste of oxygen or the most skilled worker in history, your needs remain pretty much the same. Learning how to build a house doesn't make you 10x more hungry or thirsty.

File: boy and strawberries (Dav(...).jpg (541 KB, 1600x1067)

541 KB JPG

>>108892003
>What will millions of people with no job going to do?
Farms are always hiring.
If an illegal Mexican who can't speak proper English can find a job, then what does it say about the pampered cart pusher or factory worker who is even more worthless?

>>108891432
You proved my point.
Back in the 2000s Blockbuster dismissed the internet in the exact same way artists and coders are shrugging of AI.
They didn't believe it was a threat until it was too late. Youtube and Netflix eliminated the video rental business, Blockbuster had no fall back and died.

>>108892245
anon, we were making over $100k...

>>108892245
Farms will automate harvest collecting with robots.
Next.

>>108885897
Microcock is salty because nobody uses copilot. It's making them lose a lot of money. And copilot is not even that bad compared to the others.

>>108892286
And the Blockbuster CEO made millions.
What's your point other than you hate working.

>>108892364
They are last on the chopping block right now. Get to work and stop crying.

>>108886153
govts arent waiting for markets to drop before a bail out. There will be no correction.

>>108892263
>backpedaling this hard already
Ever heard of betamax?

>>108892245
Farms are only hiring illegal mexicans specifically because they can take lower wages (as they'll go back to mexico with the proceeds) and don't have protection under US law. Many kids try to get farm jobs when they live in the right area and are turned around because illegal mexicans are the only thing farmers accept.

>>108892392
Farms are already highly automated and they're one of the areas where robot automation has made the most headway so far.

>>108887146
based as fuck

>>108887146
Massive amount of gibs were used to prevent the correction from actually happening. This is why we only see tiny (less than 90%) dips, but also why it keeps dipping frequently (totally abnormal for healthy markets).

copilot is so fucking shit holy shit