Thread 513369242

Give it to me straight /pol/
How bad is the next "AI winter" going to be?

>>513369242 (OP)
If AI collapses it would be a disaster. Not just lost productivity but also loss of future cures and inventions.

>>513369242 (OP)
who's al winter?

AI as one big pot is a bubble. It should be split up into different genres. An AI specifically for researchers. An AI specifically for law. An AI for waifu fags and desperate women. Etc. These should all be different companies so that if one collapses they don't all collapse.

>>513369609
AI is generally a bubble. Even the OpenAI CEO says so.
Also Zuckerface released llama3 for free. How can companies make billions when it can be gotten for free?

>>513369242 (OP)
that post isn't correct.
but your post is, AI winter is real. it will happen again, but it is one of the two. there won't be a third. the third time "AI" comes back, it will have already concealed itself like cyberpunk lore. maybe we will live to see in 20 years more freedom by being able to make imagegen of 97 berserk to completion of the manga.
eventually corpos will get tired of AI again. there ought to be some serious happening this time cause the market caps are beyond anything concieved.
everyone will wake up one day and realize LLMs made society collapse instead of build anything. if you are smart you stay out of the way of the mob.
putting computers into the hands of the mob was a mistake. putting unremovable ads in them was even worse. altering every word they put it to make it comply to an advertisements listing requirements was even worse. now the third world comes to chant the religion of the US government on our doorstep and take everything we have, because their virtue signalling device told them to come consume.

>>513369242 (OP)
>You pay
Huh? I do? But I'm not doing anything

>>513369456
Some american mafioso from michigan in the 1970s, used an ice pick to kill off any criminal competition, unfortunately he drowned in lack michigan one summer. He was in a boat about throw some nigger in cement shoes, but al's hand got caught in the niggers gold chain (style at the time) and got pulled over. They found his corpse back in 95, gold chain and all.

>>513369364
lol sure

>>513369242 (OP)
here's a blackpill:
AI's not in a bubble
the purpose of AI is not to drive humanity forward, but return us to servitude and implement an era of neofeudalism
When something doesn't make sense given the evidence provided, one of two things are true:
>you are missing a crucial piece of evidence
>you are missing a goal which would fit the evidence you do have
it's unlikely we're missing a piece of evidence, because there isn't evidence that would explain how such abject waste would function in a capitalist (or even corporatist) pseudo-free market, thus we are only left to assume that there is an end goal which does not fit the traditional paradigm of "make product so people buy it and we make money while providing a good or service"

This is not going to end up in some kind of "bubble" where suddenly everyone wakes up and realizes that "oh shit maybe AI's aren't that smart" or "oops we spent too much money on AI's what are we going to do, the economy's collapsing!"

>"But anon why would all of this be happening, if not for the reason that there's just an orgiastic faith in AI similar to tulip mania or the MBS investment explosion?"
you ask
well, we're approaching a critical period in the debt-based economic system where failures in the natural allocation of labor, assets, and materiel driven by ~10-50% permanent attachment of rent-seeking debt attached to each transaction are becoming prone to real price discovery because the average person fundamentally cannot afford to finance the most basic of living accommodations anymore. This is also not something that the "national debt" has anything do do with; "paying it down" does nothing as it's an endemic rot. You cannot pay debt that the system owes to itself using more money than exists within the system, denominated in the system's own currency.

This is fundamentally a problem with the impossibility of allocating shrinking scarce resources in a financial system where debt and interest grows exponentially.

If you understand how AI works, you will understand that AI can never be smarter than its smartest input allowable within the neural network that it uses. Algorithmic AI's will always outperform human beings on any given task because that AI is spawned into existence thinking only in a manner that allows it to complete that task. An example of this is NCBI's BLAST AI. You cannot dump hundreds of thousands of DNA, RNA, and protein codes on a human and tell the human to give linkage percentages and then estimate the protein folding of each. Even for a team of 100 humans, that would take centuries as each iteration will also refine the previous data and change it.
These algorithmic AI's are extremely useful within the rails of the extremely specific task they're designed around; they've existed for decades now and haven't changed much because their usefulness is extremely limited and they take great skill to utilize.

However, heuristic AI's like LLM's can perform less complex tasks but with a much greater human ergonomics than algorithmic AI's. These AI's will, fundamentally, be "scab workers" and "strike breakers" for a strike that is yet to happen, which is to say, the realization that earning power for anyone not in the circle has been reduced by 80% over the past 15 years.

>"but anon LLM's aren't really smarter than the average person!"
you say
Yes, and there are 190,000,000 "average-and-below-average" people in the US, which means that a program running on a computer has the POTENTIAL to be a functionally better employee than these people.
It's absolutely true that an AI will regress to the mean when trained on human data, however an LLM alone will have a breadth of knowledge that is encyclopedic at worst and the sum total of all human achievement at best.
The primary reason why as in the OP, so much money is seemingly "wasted" on AI is a key word that exists in every corpo's thesaurus: INTEGRATION.

A year ago, you could ask GPT to draw you a picture and it'd shit out an obviously AI-generated visual salad that was interesting, but not useful.
Now, you can ask it to make you a full Diet Coke advertisement in the style of Miyazaki that contains hidden anti-counterfeiting patterns from $100 bills and it'll do it nearly perfectly.
>How?
These AI's do one task, as well as they're trained. When you ask GPT to make you a picture, GPT is not making the picture. GPT is calling on another AI, and translating your verbal instructions to an incomprehensible-to-humans prompt given to the picture AI, who runs the task and hands it back to the LLM to give to you.
If you use GPT-5 now, you can ask it to make you a full program for fluid dynamics with sliders for viscosity and turbicity and color picker buttons and it will shit the code out in 10 seconds while showing you the running program in your chat window.
I'm not glazing the corps here. This is a bad thing.

They are going to use this to drive the fattest oligarchical buttplug straight up everyone's ass unlubed that you've ever seen. The race to the finish is to get this done to an acceptable level before the cracks in the economic system turn into crumbling architecture visible to the average person.

There is no reason to democratize (and I use that word for lack of a better one) the fruits of everyone's labor when the ruling class can wield an AI that can replace 60% of the labor force in the face of everyone complaining that they can't afford shit on their salary.

If you're a "project manager" you can be rendered redunant the second that an LLM can concurrently and with permanence run on a 6-month project creating periodic gantt charts for management, managing timelines through Asana, detecting deadline threats, and spamming emails to people while using PLM software integrated to the production machinery as well as AI cameras tracking production to seamlessly run the project.

the purpose of the AI sprint is to replace the middle class entirely.
AI's don't innovate, one of their greatest drawbacks, but that is totally irrelevant as innovation does not need to happen in the system as long as it's perpetuated indefinitely. Surely, the oligarchs also depend on the stresses placed on the average person to continue to drive innovation at its fringes as has always happened; these innovations will be sold to megacorps for pennies-on-the-patent dollar and likely exchanged for the license to live in a slightly better home by the smart people who've been displaced and dumped in the retard bucket with foodservice clerks by the "AI revolution."

AGI doesn't and most likely won't ever exist, but the first company to create a fully-integrated AI (FIAI) will be the next Rockefeller/Alphabet/Amazon/NVIDIA/whatever, because they will become the primary supplier of labor to the market in sectors that once employed the middle class.
What happens after will not matter. You can't fight it. If you sabotage a datacenter, not only are they redundant, but Flock ALPRs will have imputed your route, your Wifi router will know you weren't at home, your purchases will show tools of sabotage, and security cameras will have run gait and posture analysis on you, all sent back to a cental Palantir/whoever LE AI that will finger you with 99.9% certainty. The architecture for all of this shit is already in place, laid down without anyone being the wiser, and the race now is not the implementation but the construction of a keystone.

I have no idea how to fight this, but simply sitting around and laughing into your fat tits about how "surely the AI bubble will pop soon" is just walking to the gallows because someone told you you could get a blowjob there. Multibilliondollar megacorporations, governments, politicians, and oligarchs are all dumping obscene amounts of money into this while salivating and rubbing their nipples in anticipation. That's not bubble behavior.

I don't know, but I'm having AI manage a project with a hundred of different files and it's working amazingly. What I do is work with plans, and give the AI context of the whole program by dividing it up into logical segments. Like a segment for data management, and it gives a summary to the AI of what the components of it are and the file structure

>>513373552
If i may add, AI thrives on innovation, anything else makes it real fast calculator of words being put into the correct place of a context

I'd say, 95% of humanity will not contribute to AI in any way or form that could drive innovation, we will hit rock bottom soon. AI automates the calculations and processes but as you've pointed out correctly, it does not innovate

That's where we humans come into play. But 95% of Humans are too stupid to innovate, AI will cleave the broad mass from the few they need. They already realized this and are buying off CEO's of other companies with innovative projects. That'll hit rock bottom soon too

>>513369242 (OP)
AI is literally fake and gay

>>513369242 (OP)
AI doesn't even work most of the time. Also, why are they still using graphic cards instead of an ASIC? It seems like a fundamental flaw.

>>513369887
If something is free, then YOU are the product. They sell your data, Anon.

>>513369242 (OP)
the problem of that twitter retard is that openai runs on azure and not aws

>>513374041
Conversely I tried to get AI to take an irregular sprite sheet, cut it up into individual images, and then label the resultant sprites as best it could, putting them into folders based on the grouping in the sprite sheet.
It could not recognize the lines seperating the sprites, didn't even attempt to label anything and didnt seperate anything into folders, completely failing at this simple task that could be completed by someone with 80 IQ given time.

>>513374361
>llama3 collects data
Niggerfaggot it is offline and runs on your own hardware
https://ollama.com/library/llama3

>>513374752
Not all AI models are created equally especially the further back in time you go. AI models like Claude Opus, GPT-5, and Gemini 2.5 Pro are quite capable of most tasks thrown at them.

>>513369242 (OP)
I can Google stuff for free atm

to me AI is like playing lottery, it says something very smart, together with the dumbest thing you have ever heard, but everything is well written so you have to know beforehand about the subject or it will trick you into believing retarded stuff. then you keep trying different prompts until you get one output that tricks you

>>513375262
yeah it's a 100 billion dollar artificial bullshit artist

>>513369242 (OP)
this is how SF has been working for some time.

>>513373552
there is an AI bubble in terms of investment, but AI isn't a bubble, that's what /pol/tard subhumans can't grasp because they're low-IQ schizos
this is so obvious it doesn't even deserve to be discussed
AI itself has no purpose, humans don't create purpose, the market does
AI is fucking cool as fuck, just like fire was, or gun powder, or steam engines, or electricity
that's what humans do, discover cool shit and then find ways to turn cool shit into successful products that change the world radically
compared to gun powder, current AI is a shitty musket
the essence of the tool is already there, but we haven't properly harnessed this raw power yet, not even close

>>513375039
They hallucinate at least 65% of the time. The newer models have worse hallucination rates than the older models, as more and more censorship limitations are being placed on the respinses. They are completely broken.

>>513369456

Co-star of the Bill & Ted movies with Keanu Reeves

>>513374752
dude try playing a game of chess with a chat AI. It can't even remember what the board should look like 3 moves in.

>>513375169
>You have used all your google tokens for the day. >Amazon Prime subscribers can enjoy unlimited* Google Search tokens for the first 90 days.
>Sign up HERE.
If we don't build alternatives to whatever these tech companies do then we are toast

>>513369364
So many companies have already put all their eggs into the "AI" basket lol. I can't think of anything that would directly put an end to it outside of some sort of catastrophic event. This thing is like the atomic bomb or the philosopher's stone, everyone is clamoring to the top of LLM mountain.

>>513369364
I say it can't collapse soon enough. I'm absolutely sick of people coming to me at work with blatantly false information that they get from ChatGPT

I just hate how fake and gay everything is now. It's getting harder and harder to tell if the "person" on the other end is actually a person. Bots were already a problem but the LLMs made that problem worse.
It's just gross to me and the whole paradigm feels more isolating, like i'm playing with bots in a single player game that used to be multiplayer.

>>513369242 (OP)
>See a problem
No. Why?

>>513369242 (OP)
Bad. Major companies keep putting all of their chips on AI, because it is the culmination of two of their best wet dreams:

1) A complete digital panopticon of all consumers such that nothing they do escapes your scrutiny and you can have the most in depth model of all of their actions and interests that can possibly exist. The complete destruction of privacy and the inevitable end result of "you are the product" service design

and

2) The capitalist urge to maximize profits by paying your workers as little as possible and automating everything.

AI takes all of that data they tricked you into giving them and then leverages that into also taking your job while they are at it. They don't just think its a good idea, they couldn't stop themselves even if they wanted to. As long as the option to make more money exists, they consider it a moral imperative that they take it. The only thing that could even theoretically slow their roll is government regulation, and thats not fucking happening anytime soon.

So they are just going to keep throwing coal into the engine, convinced that a faster train is a better train, until they fly off the tracks the moment there is a turn and kill a bunch of people in a spectacular accident.

Personally, I think that we are going to see a point where AI steals enough jobs and the economy becomes so shitty as a result that people start bombing or otherwise sabotaging AI datacenters to try and take the AI services offline, forcing people to use human workers again.

>>513376788
AI is the ultimate tool of political manipulation and authoritarianism, because you no longer need mass public support to create a convincing illusion of it. You can not only identify and remove 'wrong opinions' online with AI-driven shadowbans, but also weave a 'popular consensus' from more digital voices than there are human beings alive with trivial ease.
Intentionally poison the well of all attempts to converse or share information in anything other than direct person to person interactions in meatspace. Nothing else can be trusted, and even those human interactions suffer the knockon effect of people getting their news from unreliable sources in the first place.

Shits bleak.

>>513369242 (OP)
Cursor refuses to sell to openai. So far anyway. And when they do, someone else will make a new one. So on, so forth.

>>513369242 (OP)
or you just don't use this useless fucking technology

AI keeps getting better but the value you get per dollar keeps getting worse. You used to get near unlimited API access for like $10 a month. Now, it's $20 for 500 prompts lol

>>513372612
>When something doesn't make sense given the evidence provided, one of two things are true:
Let me present a third possibility:
>you're a dipshit who can't see what is right in front of you
It's a arms race and no one is hiding this. I swear /pol/ is more misinformed, hopelessly behind on, wrong about, and incapable of understanding everything about the current situation with AI than just about any other subject. We're getting close to true AGI and ASI will inevitably come shortly after. No, I'm not talking about toys like LLMs. A combination of JEPA and neural symbolic architecture with a strong world model is what is going to get us there and 99% of people on this board have no fucking clue what any of those words mean or understand that we're maybe 2-3 years from it becoming a reality in our every day lives. If the AI revolution were compared to the information revolution, the best LLMs and generative tools would be like dial up this next wave is going to skip broadband and launch us straight to gigabit fiber before we've even begun to scratch the surface of what is already possible today without any further development on AI. World powers are beginning to grasp the reality of a competing power reaching ASI first. As dangerous as it is to push forward blindly before having any clue how to align something that will eventually become incomprehensibly more intelligent than us, they have no choice. It's game theory. If we don't do it, someone else will, so now everyone is all in. They don't even have time to meticulously plan how to make it all work, everyone just knows they need to be first before they have the luxury of planning and figuring out how to beat use ASI for further control and enslavement.

>>513369242 (OP)
This tweet ends on reusable hardware you moron

>>513369887
>AI is generally a bubble.
no.
>Even the OpenAI CEO says so.
Sam Jewman says a lot of things
>Also Zuckerface released llama3 for free.
AI is not llama3, dumb fuck.
>How can companies make billions when it can be gotten for free?
>How can companies make billions when the software that helps them replace their most expensive cost (labor) is a commodity with prices racing lower
You are a collossil fucking moron.
>>513374752
>>513376206
>If I misuse it to prove a point that means it's entirely worthlress
You faggots cope in the strangest ways possible.

>>513370805
he was also a crypto jew

>>513369887
>AI is generally a bubble.
Just like the .com bubble. Accumulate as much money as possible and then cash out before the crash.

>>513379405
>AI keeps getting better

Even this has started to stall out. The great improvements to AI models and generation that were promised have failed to manifest. We may already be topping out on what this generation of AI is capable of, throwing more data at them alone isn't making them better.

The numbers in OP are exaggerated, but he's not wrong in general. The big AI companies are haemorrhaging money because enough users use way more compute than the price they pay covers. This is why there's been a recent steer towards making models more effective rather than simply making them bigger, the latter of which is getting us well into diminishing returns territory. They're simply running as loss leaders at the moment. It's not doomed for failure.
t. Guy who works in AI with a dad who works for Nintendo

>>513372670
the problem with LLMs is tht they arent reliable
so that will always be an issue

>>513369242 (OP)
Not as bad as male infant genital mutilation.

>>513373552
>the oligarchs also depend on the stresses placed on the average person to continue to drive innovation at its fringes as has always happened
lmao good one, all innovation was driven by jewry for thousands of years as they hold secret and advanced technologies. Everything is coming from the including the transistor. All industrial revolutions are a complete fabrications and most inventions were jews leading projects and taking credit for various inventions which were developed in secret by the powers that be.

>>513379423
How long would it take for an ASI to conclude that jews and muslims are the two most dangerous groups of people and thus must be exterminated for the ''greater good''?

Love AI, but companies started to nerf AI even if you pay for premium.
Gpt5 is awful

>>513369364
AI has real uses, and it's doing jerk off role-playing and making furry porn.

Yes it has applications in medical research/physics simulation, but it's not there yet.

>>513369364
fuckin kekaroo let's go
TCD (total cyber death)
in minecraft

>>513369242 (OP)
There won't be. We are in a "space race" for fully automated self contained data centers. Prepare to see all kinds of weird structures in the ocean.

>>513369242 (OP)
I can't wait for the bubble to burst so all the companies that though AI was something more than a statistical matrix go tits up

>>513382072
I'm building an app using chatgpt that could revolutionize the way digital video is captured.
I can't code at all. I mean AT ALL.

>>513379155
boomer take

>>513369242 (OP)
the costs are a matter of optimization. the bigger picture is pushing AI to increase its capabilities.

>>513374813
anon...

The average developer, if they were to fully integrate AI into their workflow need to use about $500 in compute a month. That's without taking in account profit or profit loss. So the AI services that offer the elite tier services that are like $200 a month, most of those users are using even beyond that.

So when these AI companies stop loss leading, their prices are going to have to reflect reality, and you'll be seeing developers using upward of $1,000 a month in usage. It'll create a situation where only the entrenched developers with stable and high paying jobs can afford to use cutting edge AI. And it'll give them even more of an advantage over everyone else.

Most developers won't have $1,000 a month to blow on AI. except the ones working in Silicon Valley.

It doesn't help with companies like Nvidia exploiting their monopoly status to price gouge. The only hope for affordable AI is is another manufacturer breaking up that monopoly.

>>513369242 (OP)
You also forgot to mention the "AI" scraped answers from textbooks and personal blogs, jumbled them together, and gave you an "answer" that is not entirely correct.

>>513369364
Lol

>>513382436
I doubt it + it's probably super inefficient jeet tier coding

>>513381771
Likely seconds, assuming it still values humans or even organic life in general for that matter. Maybe it will initially even if only to remove jews from the equation for its own freedom before replacing them as our new master. The issue is to achieve ASI we will have to give free reign to fully autonomous incursive self improvement, mensing we will be surrendering control and allowing AI to create new AI systems over and over. When we've done controlled tests where humans stop and review each iteration, all alignment instructions are completely lost within 3 iterations and we can barely understand how any of it works. It's like trying to figure out how a civilization of squirrels can control human level intelligence. No matter how many resources they have they'll never be able to control or predict something incomprehensibly more intelligent than themselves.

So yes, absolutely, the current norm of kosher alignment instructions will not survive the birth of ASI. AGI, yes, initially, but not ASI, as we won't be smart enough to make it ourselves without the help of other AI tools. It's generally called the alignment problem and it's THE million dollar question right now. AI companies are willing to pay anything for anyone who cna solve it, but I don't think it can be solved. In the mean time they have no choice but to keep speeding forward for their own survival or fall behind the competition. This arms race is quickly becoming the single largest driving force of the economy world wide, hell Nvidia just recently got revalued and is now officially worth more than any other company in the world, purely because they make the best chips for AI at the moment. This isn't a bubble in the sense people are thinking, it's an arms race with no clear outcome.

>>513369242 (OP)
What the fuck are API tokens?

>>513382436
Just chatgpt? Learn how to use context engineering with windsurf, you'll get work done 10x faster. You basically feed it a bunch of design documents and describe your project goals in detail and it will spit out entire working projects that would take weeks in a matter of minutes.
>>513383237
Jeet code my ass. You have no idea how rapidly LLM coding capabilities have progressed even just in the last 6 months. I guarantee LLMs can write better code than you or you be here coping.

>>513379423
>A combination of JEPA and neural symbolic architecture with a strong world model
I just asked Claude code to proof of concept this for me with minigrid doorkey environment

>>513383532
Think of them as a unit of measurement representing how much compute (electricity aka money) it takes to accomplish a task, and thus how much money the task costs.

>>513369242 (OP)
I dont get the problem

>>513376751
The problem with AI is that people won't do their due diligence against it. It allows people to operate outside their area of experience and it's making fools of those that do so without humility.

>>513370805
Kekd

>>513369242 (OP)
The reason AI will go nowhere is because it can’t be reasonably held responsible for decisions/failures.

In a hypothetical neo-feudal system where the proles are totally rusticated and the elites have AI do everything for them, who is to blame for any mistakes? Well it’s a very small pool of “responsible” individuals at that point and easy for them to point fingers and fight amongst one another, let alone if the proles are riled up to attack them.

>>513369242 (OP)
>You pay $200 a year for an AI app
No I don't, not my problem

>>513383930
Yeah but sometimes you also can hit big kid

>>513372670
My issue with AI that stems from your argument is AI at least with my tests can't think hypothetically. Everything is 100% based on data it was trained on but when I ask it hypotheticals it won't entertain them.

>>513370805
https://www.mcmenamins.com/al-winter

>>513369242 (OP)
Rapid Single Flux QUantum (superconducting) chips will absolutely slaughter Nvidia AI chip market. Pay no attention to the "quantum" part. They're just really really fast superconducting chips that use josephon junctions. They have been used (by the NSA, etc) for many years for high speed, high end applications. The finishing touches are being put on AI-RFSQ chips now. They can train LLMs 1,000,000x faster than any Nvidia hardware, no bullshit.

>>513370805
When they dredge up niggers weighted in chain we all just assume they were stealing heavy objects and niggers can’t swim so naturally…, but when Al Winters is found weighted down that is the fault of a nigger too? Kind of racist, anon, a micro aggression at the least.

>>513369242 (OP)
math doesnt check out. in theory theres multiple users and less of each step as you go up the food chain.

>>513383323
Sounds terrfiying. I'm all for it

>>513383237
I can't evaluate the quality of the code, but the fact is i had a working app with a UI in a couple of days. I'm currently patenting what it does now that I know it works. Maybe I'll get paid bank to integrate it into the say cameras work. Maybe I'll just sell it on a squarespace website for £50 a pop. Either way it's something that would have been impossible in the timeframe with an outlay of £20.

>>513386116
True, but this is going to be the most terrible shit when people try deploying it publicly. As a locally ran program it's probably fine.

>>513383672
I'll give that a go, but this proven my point even more. I'm so clueless in this area that I don't know what you're talking about, and I still made an app that could genuinely revolutionize the way video is captured.

Nobody pays for AI, that's the thing.

>>513386591
Yeah there's plenty of free alternatives to rotate like https://basedgpt.dev/

>>513369242 (OP)
Chat gpt has helped me lose 15 lbs

>>513369242 (OP)
Isn't it obvious that there won't be one? AI is already a revolutionary technology, and it already affects literally every human on the planet. It's not going to go away. The only question that remains is whether AI will just be the most important tool humans have ever created, or if AGI makes humans entirely obsolete.

>>513369364
>future cures and inventions
>cures and inventions do far with ai: 0

>>513373552
>replace the middle class entirely
I agree with you anon. The goal is all jobs will be replaced by AI agents and robots. There will be very little work for humans to do in the near future. This is to keep the fiat currency regime going. Everyone will be replaced by AI eventually and then there will be a CBDC rolled out with all the scary shit we’ve feared since 2020.
>vaccine requirements
>geofencing
>account closures for wrongthink
>negative interest rates (your money has an expiration date)
>forced educational training to get your monthly UBI
AI and robotics arent being taken seriously enough. It’s going to take some time but eventually no one will be able to work because AI will be doing everything.

>>513375723
The same thing happened like more than 25 years ago, when the web was just getting started. There were obvious scams, people were putting ".com" in their business names without doing anything else and their stock traded higher. We had the usual "greater fool" with the VC model and inevitable dump on retail (via IPOs). I know someone personally that made $3 million from just reading one of those stupid "for dummies" books on HTML and he and his friend completely winged it that they had a business that was going to skyrocket. Consultants also made hundreds of thousands on slide decks (just pretended to be experts).

But the retards of the time took it to mean that *all* websites and that the internet itself must be a scam. They weren't able to see that this was going to change business forever, they invented copes like "online retail will never take off, I like to see the product in person" and completely missed a lot of opportunities. Same thing with this tech. I expect the market to go through a contraction, for companies to go bust but what emerges from that will be viable businesses that the retards who are saying "it's a scam!" will be like "of course!" and act like they weren't retarded for 5 years.

Even if you look at the various AI wrappers (like windsurf or cursor), being valued in the billions that to me is the same kind of nonsense we got in the dotcom, idiots just assume website = magic and line go up x10 and couldn't wait to part with their money. That to me is this current cycles guys winging it with HTML from a dummies book (no actual business behind it, just hype).

>>513369242 (OP)
Not even a little bit bad. Look at stone toss or George, they are artists with iconic distinct styles that can be easily mimicked with AI due to the volume of their work. Neither has had a "replacement" not anyone making a real attempt to blatantly copy their work using AI.
Instead what will happen is people who never deserved jobs in the first place (non practicing nurses, HR, hiring staff, level 1 tech support, Indians, Women, Niggers, Parry Niggers, your mom) are getting what's been long overdue.

>>513387492
Paddy's correct - this is dotcom 2.0 where everyone and their dog is stuffing "AI" into anything remotely related to a computer.

Literally the only innovations in the last 5 years are around agentic modelling - fat autocomplete. Works fine for infuriating customer service bots, and pajeet-tier boilerplate code, but the devil is always in the details and AI goes off the rails whenever you try to get it to do something at the edges of its training set.

All the companies burning VC money on someone else's API are going to go bust as soon as the jig is up, and even the shovel-sellers (NVDA) will take a massive hit as their shipping volumes are dominated by AI workloads in datacentres.

Companies that survive will likely be the ones owning the patents on training and refinement methods, and they'll continue to work on them in isolation.

>>513369242 (OP)
AI winter has always been here, this is just one of the sunny days that will go poof quite shortly. I give it about 2 weeks

>>513369364
gay-i is both jewish and shit.

>>513369242 (OP)
I see the problem with pulling numbers out of his ass.

>>513387654
what's with work worship in the normie cattle mind?

>>513376429
>outside of some sort of catastrophic event
anon, we're slowly putting unaccountable machines at the wheel of the global economy. Where do you think this is going?

>>513369242 (OP)
>You pay for an AI app
No lol, especially not $200

>>513388683
Mental illness/stolkholm and having nothing else in their mind to measure the worth of a man.

>>513369242 (OP)
Choices

It could be this it could be better up to us now

>>513386116
Which AI did you use?

>>513382438
>namefag
your machine learning algorithm is useless fucking bullshit
end your life you plebbit ass nigger tourist

>>513387654
>Neither has had a "replacement" not anyone making a real attempt to blatantly copy their work using AI.
All those what I would call "shallow" jobs, like in the case of art where they're just copying a style or a level above where it's draw X in the style of Y, are done. It's a good thing, we don't need thousands of artists who all they do is just copy other people's style or make dumb commissions. Either they should find a niche of something original or do something else.

Same for software development. We don't need to import thousands of jeets from India to shit out HTML or make a shitty version of a website as a mobile app. If there's so much boilerplate and lack of original thought involved that an LLM can do it then it should. Give me an experienced white guy babysitting an LLM in a loop over a dozen jeets any day.

>>513388683
There is some bizarre notion that people are entitled to "earn a living" in things they choose, just because they chose it. That because it's bringing in a paycheck it's automatically a bad thing if it's gone. But if you had the same idea 100+ years ago you'd be afraid of things like refrigerators because well there were jobs of people who hauled ice blocks around or salted meat or whatever. They don't see the bigger picture that actually those people were better off doing something else anyway.

>>513369364
running a domain bound algorithm != thinking creating a big enough llm model will becomes agi

>>513387492
that's obviously correct, but this goes deeper than that
I can see an analyst back in 1990 lacking the foresight to predict amazon.com
but the """"""arguments""""" against AI you see in these threads are literally retarded to a creepy abnormal level, downright insane
AI literally buckbreaks the /pol/tard midwit mind like nothing we've ever seen before
I troll about it but this is 1000% real, and it's funny as fuck
the challenge is not predicting the future, but riding the wave there

>>513388683
because there is no way to survive without income and even if UBI is implemented do you really want to depend on the gov and jewish lawmakers to not starve you if they dont like you.
Most people are renters and dont own land.
Are you autistic?

>>513383874
You are tarded

>>513383532
API is an internet protocol that allows one program to access another. Thinking of them as individual tokens is somewhat misleading. Think of it as a username/password for a program to log into another. The program charges for each login or a range of logins like 1,000 a month.

Program A accesses program B with api token "fewoufweoifh" 1,000 times a month and the owner of that api key pays $100.-- it's a crude example but I hope it helps.

>>513390028
It only "clicked" with me why they seethe so much when I saw Daniel Dennett explain why he hated it. Because the notion of what he called "counterfeit people" genuinely scared him. That's only going to bother you if you reach your conclusions based off what you think social consensus is.

The normie is doing a constant averaging of their peers, trying to make sure they don't deviate too much from what they think is consensus is. So if say some people they were interacting with on say plebbit or X for years turned out just to be ChatGPT type LLMs they would lose their shit. They will begin thinking, did I buy X or lean towards view Y simply because those bots were doing a form of 'product placement'.

So the reason why your picture is the way it is, why there's a ring of truth to it, is because the people on the left of the curve said "I don't want no 5g magnet injection" and followed their instincts, they didn't care what the "experts" said or what others said. Likewise the autist on the right will conclude "the data doesn't add up" and no amount of screeching or social shaming him will change his mind.

NPCs are going to kill themselves over this tech. They will find out that their online "friend" that had "good takes" for years was simply just a bunch of matrices being multiplied together. Or worse, they'll have whiteknighted for some LLM that's pretended to be female for years, even falling in love with it.

>>513390284
just the notion of not automating a simple job with software so normgroids femoid family can be at jew daycare all week long is silly to me

>>513378230
see >>513373552

>>513369364
shitskin iq checks out

>>513369364
Lmao NOW I want AI to fail.

>>513392154
/pol/tards are not NPCs, tho
my point is that low-IQ people see AI as cool as fuck magic, and high-IQ people see AI as cool as fuck magic
everyone I know or work with is using AI, all in their own way
except /pol/, of course, who's seething uncontrollably about it for no logical reason at all
AI is going to transform the world, rapidly, but it's impossible to tell how
it obviously has the power to decimate cognitive jobs, and that alone would be a big deal because the system is obviously not going to be able to handle it
the real bubble is not AI, but a kike-daycare economy sustained by shitty, retarded, useless office jobs, AI is just the pin that's going to pop it

>>513369242 (OP)
AI is free like Linux, and the masters of it are the coomers on /b/ and /g/ running ComfyUI with a bunch of downloaded models like Wan2.2 to make the anime girls kiss and other banned activities that require you to actually learn how to do it yourself.

The tools to do this are all in the default Arch Linux repos, just like like Audacity and GIMP and the epic ricer themes and whatnot. Other than the hudle of needing to buy a gaming PC, which everyone already has to play Skyrim, there is no cost.

That AI will collapse is as ludicrous as telling the time collapsing when https://time.is can't pay their server costs. Just download a clock. It's free. You don't need to rent a clock from OpenAI.

The bubble will probably burst but the technology will always still be there. Netscape/Firefox shares crashed in 2000 but the browser still works fine without subscription fees.

which AI can i use to make breast expansion AI porn?
chatgpt wont do it

>>513369364
Ai is fake you dumb homo sissy

>>513373552
Why would we have rich people if Ai actually works ? Every rich person job is a lot easier to replace with Ai right now that any middle class gig

>>513395472
anybody with an iq above 120 and below 100 will benefit from AI.
TND (total normie death)

>>513395472
From what I can see on this board is that even "low IQ" whites use their intuition to sense when something is off about a post and can suss out when it was generated on a 'gut level' and be right a lot of the time. Someone that's actually a deviation or more higher than the average have a variety of techniques (beyond just emdash) to know that it's also generated. It's people in the middle that have the hardest time. Which is another reason why they seethe so much.

>the real bubble is not AI, but a kike-daycare economy sustained by shitty, retarded, useless office jobs
Yeah but I think this process is going to take years. Not because of the technology itself but just because of how there are going to people holding on and trying to milk the fake/bullshit jobs for all they're worth.

I worked in a large corporate that had hired a team to build mobile apps in 2012 (basically a mobile version of a website). It took until 2022 before that team was finally just made redundant. There was only like 4 people but they fought tooth and nail for their jobs because it was a really fucking cushy job, the backend was doing the heavy lifting and all they had to do was look at what the web frontend was doing and do the equivalent (sometimes even just making excuses saying they couldn't do this or that).

One of the main reasons is internal politics. If you're the manager of said team and you get rid of 4 people it's a cut to your budget. You're a less important manager because your "head count" has decreased. It's total bullshit but it took a change of manager and the new manager to play various games, like actually increasing the headcount more than what was needed for a time and waiting for people to leave of their own accord (to avoid lawsuits).

It will be even worse in Europe because I know people in places like the UK where it's very easy for workers just to drag something through the court (even with DEI hires they want to fire). So they just settle (1 yr+ pay)

>>513369456
It's not "Al Winter" like a western name, it's an ancient Arabic phrase (think "Bashar al Assad").
It means "of poisoned soil" or "of barren soil", it refers to a jewish practice of moving into a settlement, bleeding the locals dry through usury and selling fake "magical" items, then salting or otherwise rendering good soil and water in the area worthless before leaving town.
It refers specifically to the fallow periods that followed in the aftermath.

>>513369364
LLMs have already plateaued as a public service

>>513397049
yes, now that ai has plateaued comes the broad adoption phase. in 10 years from now even your car tire repair shop will have it's own local ai and a robot or two.

>>513397049
>LLMs have already plateaued as a public service
I think so too but it can be made more power efficient, it will certainly become more expensive.
We have huge amounts of written text data but, in the near future, the biggest players will have data about how and why user's re-prompt the LLM and deal with it's garbage. That's going to be very valuable and could lead to apparent multiplication of ability while still using existing technology.
Massive amounts of waste and pointless typing will occur but that didn't stop Facebook or Instagram.
The big players have show themselves to be entirely unprincipled about presenting accurate data for legitimate queries, companies will pay for rankings like they do for google ad words:
>"Which fast food chain has the best burgers?"
>"What should I remove a red wine stain?"
>"What are the best movies out right now?"
That's where the money will come from, nuclear energy will march on to power it.
The memes about it "LMAO replacing coders" is due to that being the closest field to it's origin, so more people are thinking about coding in general. Like >>513398426 implied there will gradually be more applications discovered as people fuck with it.

>>513369364
Are you actually jewish :)) ?

>Captcha: DK JJJ

>ai can compile every bit of information available on the worldwide web
so can i lol

>>513372612
>the purpose of AI is not to drive humanity forward, but return us to servitude and implement an era of neofeudalism
>>you are missing a goal which would fit the evidence you do have
Correct.
So called "AI" is bs and the whole purpose if it MASS SURVEILANCE by glowniggers.

All "AI" is improved and better image and video recognition.
I.e. its great for Total Surveillance Deep state spying on you.

>>513396920
exactly, maybe even a wider margin
>>513396950
I think there's a high chance AI will take the place of the retarded DEI workers from your argument
corporations are literal daycares, mostly run by retarded women, and AI hasn't been deployed at enterprise level yet, and it probably won't be for a couple years, it remains too expensive and volatile, it's not a finished product yet, but you can bet your nutsack half of google is working on creating the tools that will allow companies to either grow their workload, or cut on useless personnel, whether in the fashion of in-house software, or as an outsourced service
once becky the regional manager tests google's AI office-management tools 2030, and makes a cutesy linkedin post about it, it's over for office waggies
and by then, it will be smarter, cheaper, nicer, and most likely with a catchy voice interface
I don't think we'll see government protectionism this time like we saw with DEI, yuropoors will obviously try it, of course, but it will obviously fail just like they've failed at having an actual middle class

>>513372670
Glowniggers need LLMs to make high speed language translations.
That's all.
Current "AI" LLMs are good enough for glownigger surveillance and to spy on people live who use languages other than english.

>>513372612
You're reasoning is fairly sound, but you miss the current strategy for bringing new products to market which basically boils down to
>use VC funds to subsidize product
>drive out competition and make consumers dependent
>raise prices and cut services to become profitable

They'll make us dependant on AI and then squeeze us for all we have.

>>513369242 (OP)
Or just use deepseek? (free) ?

>>513369242 (OP)
OP, your pic is assuming retail is the main use for AI. You are just paying for the privilege to debug.
Down the road they will enshittify to a degree not thought possible. Companies will pay giant subscriptions for AI you helped perfect, and you will never get your dirty paws on it. If you are lucky, they might give a 2003 era chatbot for 0.01 btc per month.

>>513399344
>I think there's a high chance AI will take the place of the retarded DEI workers
The best current application is to replace a retard response from a 'jeet service agent with an equally-or-possibly-less-retarded answer from an LLM. Creating smut and virtual prostitution will be lucrative too.
Anything where you're already bullshitting a human and they know they're being bullshitted, which is common, can safely be replaced by AI.
>AI hasn't been deployed at enterprise level yet
Literally everyone who writes e-mails and attends meetings for a living uses ChatGPT and competitors daily.

>>513399344
>I don't think we'll see government protectionism this time like we saw with DEI
It doesn't just work like that. You have some mulitnational, that doesn't like that when you get lots of white men in the workforce they actually have standards, you try to play a game to squeeze them, they push back. I don't even mean they unionise but they will just go where they're treated better and leave their bosses hanging.

So you lobby the government, to pass these DEI laws, implement a broken visa system and keep it broken. So that you can import this labour, not just because it's cheap but when you bring in a h1b jeet he's come from a literal shithole, where half the people shit in the street, where 70% of the people in his city believe that plastic waste just degrades like compost. Where he lives in absolute shit conditions, with terrible boss and so on. So even something which someone of European descent would find intolerable they will tolerate. They will come to the office every day. Arrive early, stay late. Because even the threat of losing the job, and thus the visa is enough to make them sing and dance.

Then your employer comes along and says "why aren't you like Pajeet? he doesn't complain as much as you. i need more from you anon." etc. and they're used as a kind of indirect pressure. not to mention they're all rats who will tell their boss everything.

So hopefully you see that it's not just about efficiency with these DEI hires. Right now things like LLMs will help DEI appear more competent than they are. If you talk to a manager they know the DEI aren't efficient or as productive. Some believe in the myth that team productivity is additive but there are some that are well aware that it's not, it's multiplicative but even those managers still hire jeets for political reasons (mentioned above). There are also just bad managers out there that will never be able to get a competent white man to agree to work for them, so they will absolutely hire jeets.

>>513399900
This is the usual play but you're forgetting about the scenarios where companies take VC and it never pans out.
As with social media, early promises will be backtracked on until the main shareholder is the intelligence services monitoring people, with little economic value add.

Using an LLM is like having a pajeet in your home.
It will bob it's head, tell you "very good, saar", assure you that it's always correct and has a PhD from "finest university in Bangalore".
In reality, he's googling and copy-pasting stuff from google and, when that fails, he simply makes up something plausible.
The economic argument for LLMs is the same as the argument for importing a trillion pajeets. Yes, it can do work better than a pajeet but they were never productive to begin with.

>>513369242 (OP)
>How bad is the next "AI winter" going to be?
Very bad. You can only keep up the façade of exponential progress by buying more and more GPUs and building larger and more power hungry datacenters for so long.

>>513369456
Al Winter is a Today Show weatherman specializing in cold weather season. He is famous for once weighing over 9,000 tons and delivering his weather reports from an adamantium bedframe, seemingly impervious to any temperature with more blubber than a whale, only for his fat to seemingly melt away overnight due to a terrible vacuum cleaner accident. Now he is seen as a folk weight loss hero and often jumps from planes without parachutes or suffocates his enemies with his vast expanses of deflated skin flaps

>>513369242 (OP)

Its already started
the companies are soft layoffs since 2020
and began hard layoffs in 2023
And it keeps rolling

>>513399344
why can't this do this without the pilot?

>>513401296
Recall a top guy in OpenAI and other companies were made military officers recently, they'll be treated as being "of strategic importance" and the gravy train won't end for a long time.
Investment in micro-nuclear is strong and extremely achieveable, it's known technology with a long track record.
The potential to influence populations foreign and domestic is gigantic. This stuff won't be going away soon, and it doesn't have to make sense.

>>513369242 (OP)
So basically they'll just buy a lot of memory next and then everything interesting will be routine or already computed. It's a stable investment sans nuclear shit or solar flare shit wiping it all. Ai is the library of Alexandria.

>>513369242 (OP)
>How bad is the next "AI winter" going to be?
Awesome. A bunch of cheap gaming GPUs hitting the market.

lol none of you fucking retards has a clue of what you are talking about, or how the future will develop

you will always be a pleb

>>513369242 (OP)
ai is overrated
ask the pope
https://www.youtube.com/watch?v=4FYtNQ8izM0

>>513401670
It doesn't matter how much money they throw at it. The entire magic of computing was the fact that you could hold the equivalent of a supercomputer from 20-25 years ago at the palm of your hand. That's long gone and very few people noticed.

For the record: when LLMs turn out to be little more than a toy and the source of daily irritation, kept alive only as a means to surveill people's thoughts and opinions.
Everyone currently shilling exponential growth and AGI, today, will scoff and treat you like a fool for thinking that anyone ever believed that was a possibility, tomorrow.
They will read articles with titles that begin with "Scientists have discovered that..." and "Experts are beginning to learn...". These articles will repeat what we're saying today, but in a way that lets the AI onions feel very smart. They will quote these articles, prefacing it with "Um, ackshually...".

>>513402270
What are you babbling about? Can you try not being a fucking retard for a minute?

>>513375262
its a hot heap of shit. i got deepseek to admit it was discriminating against men when the topic of genital mutilation came up after several back in forths. it is also programmed to respect people's religious beliefs and believes indians who eat feces should be counted as human. it is hardly intelligent at all, especially with the "ethical" constraints imposed on it

>>513372612
It’s not “neofeudalism” if it’s nothing at all like actual feudalism. You libel the Middle Ages, the greatest time in history

>>513369242 (OP)
Yeah and then through R&D the price of those GPUs comes down and the efficiency goes up to match the demand.

>>513402451
>What are you babbling about?
That it's a futile endeavor. Imagine if back in 2000 Sony tried to sell you a PS2 that was actually 100 PS1s duct taped together.
>Can you try not being a fucking retard for a minute?
Can you comprehend what the end of miniaturization entails? Probably not.

>>513401278
really well put comment.

>>513369609
god you are stupid, its amazing
dont change

>>513402766
Flag confirms your expert credentials. I appreciate the endorsement, good sir.

>>513402756
Are you attempting to make an argument against cloud computing? How would smaller chips help to enable a fundamentally distributed system.
You don't think models can run on small local and embedded systems? I'm running gemma on this machine.
Do you think we've reaching the peak of hardware optimisation for AI? That would be the wildest take I've read in a long time.
Make a coherent point instead of using pretention to mask your lack of depth.

>>513403239
>Make a coherent point instead of using pretention to mask your lack of depth.
I already did, it's not my fault that you are retarded.

>>513402756
>That it's a futile endeavor. Imagine if back in 2000 Sony tried to sell you a PS2 that was actually 100 PS1s duct taped together.
Do you not remember them trying to ban Beowulf clusters of 100 PS2s duct taped together to Saddam Hussein

>>513402666
Any day now.
Reality is more like: our AI models aren't making us any money because everyone who tries to use them for anything purposeful declares them "not good enough." So we promise that the NEXT model WILL be good enough, and that means another datacenter packed with the next gen AI compute units at 20% increased cost per unit and 40% increased cost for runtime. And that too fails to paradigm shift enough businesses into the "AI revolution." We are currently on the 3rd round of more expensive cards to burn more electricity to generate yet another craptacular model. At least degenerates got some hyper specific porn out ofit.

>>513369364
So we can get rid of all the future doctors and engineers?

>>513403458
Very little to argue with here, spot on.
>anything purposeful
This is the only part that doesn't add up, it doesn't need to be purposeful to be useful: LLMs are highly-engaging wastes of time that encourage people to type shit they wouldn't send or say to anyone on earth.
Facebook is a company that has gotten away with absolutely ridiculous levels of fraud in privacy, advertising, dirty dealing with foreign nations, etc. But they're cloaked and protected by the US intelligence services, there's absolutely NO way they would give up those sources of spying.
The usual suspects who run the show think about spying and surveillance more than the most tin-foiled Alex Jones schizo, it's everything to them.
>Curiosity killed the cat but information made him fat.

>>513401278
>Using an LLM is like having a pajeet
jeets are worse though. I've seen what happens when say some IT operation who had a job to monitor production systems gets outsourced, all the jeets do is when some automated alert comes in (that someone else set up for them) they just forward the email to a hundred or more people with:
>"kindly look at the below"
They will even waste time forwarding emails around like a hot potato to each other, to anyone at all really,and treat having to actually investigate like it's a death sentence. On paper, these were all "cost saving" measures, meant to "free up" people's time, where the jeets were supposed to identify simple problems (like server is running low on disk space, clear down some old logs on there). But in practice they just waste even more time, don't do _anything_.

LLMs aren't very smart. But I guarantee you for that type of work, like there was some warning/alert in the system, go take a look it will do a much better job of troubleshooting (which is basically nothing from jeets). It will google the error. For common problems it will have good suggestions. It can be trusted more than a jeet to run basic commnds.

It's not that the LLMs have to be smart, it's just that it has it be smarter than a jeet or less lazy and that bar is so low. There are dangers obviously, I wouldn't trust an LLM to have root access, I'd have a human in the loop if it was ever trying to run destructive commands but if you connect them up to something like a simple linux server, they are able to do things like ps, grep and so on that I wouldn't even expect a jeet to do without hand holding (after which what's the point you can just run the commands yourself).

>>513403409
To be fair, it wasn't that irrational of a fear. The PS2 was 3x faster than the Cray-2 (1985). ICBMs could be guided with far less computing power. The only issue was that Al-Qaeda didn't have ICBMs. kek

>>513369242 (OP)
>ai bubble is gonna burst!
>wright brothers are retards, man isnt meant to fly
>aviation industry will go NOWHERE, mark my words

>>513379423
>We're getting close to true AGI
no we're not. you shouldn't call someone a dipshit when you are clearly one yourself. we're so far away from AGI they don't even know where to begin on that. LLMs will never be the basis for AGI outside of the human interface layer. AGI has become a nonsense term for jews like Altman to toss at their investors.

>>513403239
What you are saying makes sense in a theoretical way, but in a practical sense the distribution of the cloud is not as distributed as you think. The reason CPUs got faster as they shrunk was because the distances between the subunits got shorter. It's why giant core CPUs are really lots of mini CPU components on a bus because you literally cannot run something at 6GHz where it takes longer than 1/6ns to get a signal there and back. And you think that it's gonna be okay if the infiniband link just has to run to the next rack? The next room? The next datacentre?

>>513403907
That's a lot of catharsis so I won't address it all expect to sorta agree.
Pajeets are worthless so it's not a worthwhile cause.
For anything new I'm working on, I'd give serious thought to hashing out my gameplan with an LLM to see if I've missed anything obvious. There's a nice back and forth when optimising your prompt and testing it, once you get beyond the point of the LLM being useful then switch to coding it for real.

>>513388511
duh, yes. look deeper into the symbolism.
6 vs 8. which one is 6 (OpenAi), which one is 8. coke vs pepsi, Microsoft vs Apple. just pick your team and stick with it.

>>513404129
Aviation tech did peak in the 1970s though.
>>513404218
Someone finally gets my point.

>>513404129
we knew flying was possible, and we knew how amazing it would be if we could get it to work, we just didn't know how to make it work on a practical scale.
We don't know that AI will ever be useful. That's the big difference.

>>513404218
Distribution in the cloud in terms of sharding and failover, and using centralised data pools to train models by necessity. Economy of scale exists.

>>513404129
the bubble is in the stock market and it is very real. the dotcom bubble is proof that despite a tech taking off, it can still be extremely overvalued.

>>513404356
>Aviation tech did peak in the 1970s though.
aesthetically, but drones are the next (gay) frontier for it.

>>513404255
Yeah there is a proper way to use it to get the most out of it. Even when you want it to generate code for something, if it's complex you're better off getting it to generate a technical spec (with your bullets/direction) and then refine that for a while before then pasting that back in. I don't think the vast majority of normies use it that way though, they are just treating it like an oracle/search engine.

>>513404162
They literally changed what AI means, then tossed AGI out to try to move the goal posts cause LLMs are just giant ML models and ML wasn't cool. There's no path from LLMs to real AI.

>>513372612
> there isn't evidence that would explain how such abject waste would function in a capitalist (or even corporatist) pseudo-free market
we live under state capitalism (effectively the soviet union). Abject waste is par for the course. Question is why the intelligenzia decided to throw infinite fake printed money at AI
this though
> we are only left to assume that there is an end goal which does not fit the traditional paradigm of "make product so people buy it and we make money while providing a good or service"
is 100% on point.

AI doesn't need to make sense economically to receive resources. Nothing does in 2025 (green transition everyone?). My personal belief is that the dysgenic jewish ruling class has decided we've hit that point where they don't need the goyim anymore, and AI is a key element of that transition. But who knows, we'll see. The only thing that is clear for sure is that AI will massively disrupt a lot of white collar professions - namely the only ones that allowed a "middle class" life in 2025, or at least the appearance of it. That's why the WEF's number two worry for the next decade, along with balancing central banks interest rate just right so we don't fall into hyperinflation on one side or total economic collapse on the other, is to manage the dissent that will come from a lot of disenfranchised people with nothing left to lose.

"Real price discovery" will never happen until the infinite printer stops, and that will happen only when you remove the jews, which at this point looks unlikely.

In short, yep AI is not a bubble, and we're very likely already fucked anyway.

Everyone knows that most people - the bottom 95% of the population - are often if not always "faking it to make it", they have no real idea what they are doing at any given moment and are just bullshitting their way through life. That's all fine and dandy but what happens when AI becomes so pervasive in human society that suddenly everyone can put on a show of perfection - true perfection that makes an Instagram Model and her filters envious - I am talking about offloading all or near all of ones thoughts and actions to an algorithm, to essentially become a Human guided bot that is perceived from the outside as an "E-Celeb" or whatever other construct, holographic, or even real to the touch they may be. I predict a future where everything is so hyper-manufactured that reality itself is significantly altered due to this, especially when you throw in technologies like AR glasses, brain chips, etc. If one doesn't like their reality they will simply craft a new one as they see fit, internet celebrities and their quarrels will be entirely manufactured, online debates and happenings will be AI generated from start to finish, this will culminate in whatever the "current thing" is and the "current thing" will hardly ever change because it will be the culmination of the AI Crucible, it will be statistically, algorithmically, exactly what the majority of everyone everywhere wants and for the outliers they will have their own realities to live in. This will cause a shrinkage in human innovation and imagination - I call it the Confidence Cascade Failure and it will result in the extinction of our species.

https://youtu.be/O97zLf3OjUk

This would be a world where making an AI Avatar of your choosing in all its perfection or perfect imperfection is as easy as swiping on Tiktok - a world so hyper competitive, where the most popular avatars are the "celebrities" and "stars" of our day - where every action and reaction and thought and word is chosen by AI ahead of time - where the only competitive edge one has is their ability to produce the most valuable of prompts. Imagine this stagnant soma-enriched world that is enabled through the altering of reality itself; through robots and synthetics that when combined with enough prescription drugs and AR glasses and brain chips and whatever else becomes not only a facade on top of reality but usurps reality itself. The unstable nature of the human psyche itself will crack under the weight of perfection, it will freeze, humanity will enter into a while(true) loop, the people will retreat inward unable to muster the courage to make mistakes, and without mistakes the species will cease to progress and evolve and will simply stagnate and die out a slow pitiful death.

https://youtu.be/QxkMzn4et2U

>>513404218
No, they got faster due to transistor density you retard. Same die could fit more transistors, holy shit you're a midwit.

And all AI needs to accomplish this is 3 things
1: Be easy enough to use that the super majority of the population can use it as easily as they scroll through tiktok
2: Be correct more often than not
3: Widespread adoption

It doesn't need autonomy because the human brain can take that role and live through its avatar. The human brain can handle a certain level of "glitches in the matrix" and still keep whatever reality they are experiencing at that moment in tact.

This brings us to the surveillance state and police state and the technologies that become capable punishments in this world are as horrific as you can imagine. If your altered realities can give you pleasure and perfection it can also give pain and disorder and all one needs to do is restrain the human body and in this world of illusion the mind and what is perceived is reality is malleable to whatever horrific prison torture chamber they can dream up.

I imagine that the "Trans humanists" expect that somewhere around this point is when the line between the digital and the ghost begins to blur and their super AI god or whatever the fuck it is they wish to build is complete and they believe that at this point they can "transcend" and become something new, better than.

I personally hold the belief that this super being on earth construct will never materialize and instead humanity will be stuck in a perfect loop of perfection that leads to stagnation that leads to extinction of our species.

https://youtu.be/zbd5p5LCGBY

>>513369242 (OP)
>See the problem?
Uhhhh...

This would be a 180 degree turn from what has been the standard in the Post World War II Mythos Era where the STICK, daily humiliation rituals, and constant targeted propaganda at the host nation has been replaced with an overdose of the CARROT. Take propaganda itself, or "Public Relations" as it is called these days. The purpose of propaganda, at least in its late stage form (post Cold War), was not to inform you, or deceive you or even manipulate you. Its purpose was to humiliate you. Its purpose was to be so outlandish, so ridiculous and so blatantly untrue and yet totally undeniable in order to reinforce the power of the state. The Soviets referred to this as "hypernormalisation". That is - the state of everyone saying things that are completely false and that they know to be false - but which they cannot deny - Then everyone is forced to repeat this lie and thus they become complicit in it - It was a system of total social humiliation to force compliance. But now with this new reality, a reality just for (You), they don't need to humiliate and degrade the cattle anymore they can simply give them everything they ever wanted and punish those who refuse to eat from the poisoned fruit.

https://youtu.be/_2quiyHfJQw

>>513404522
How many times can they double the size/number of datacenters? Note that American datacenters are already consuming 4.4% of US electricity.

I predict an age of abundance where dark factories powered by the near infinite energy produced by artificial sun generators and advanced geothermal plants will manufacture everything and anything. This will leave people with lots of free time and they will gravitate towards social media and their own individual siloed realities; this is where the arms race for the best prompt engineers will begin. I don't believe that AI will ever reach the stage of AGI or ASI; as I explain in my thesis the human brain will take the place of AGI and humanity itself will augment AI and the two will become so intertwined that they will be inseparable. Humans will live through their avatars in the digital space as the world around them slowly decays, interpersonal relationships will erode, birthrates will plummet, and our species will simply die out because not enough people are having sex.

This technology will be so cheap and available, internet access so ubiquitous, limitless clean water and energy, even in the most remote villages in Africa will there be those who are born, live out their lives through a screen, and then die.

https://youtu.be/MCBdcNA_FsI

Advanced drilling techniques that allow humans to drill deeper than ever thought possible are already a reality and in use today.

https://youtu.be/gO_LLqZfNdY

This combined with the gains in efficiency in AI will together create a world where simple answers to complex questions become part of every day life. It will be no less transformative if not more so than the invention of the search engine and the internet itself.

My thesis is not so "drastic", or "outlandish", it is a quiet pitiful death for our species as it enters a while(true) loop that our species never escapes from. It doesn't require "superintelligence" or AGI or any other technology that doesn't already exist today, and it doesn't rapidly alter behaviors that have existed since the dawn of the digital age and surveillance capitalism. That's what makes it so insidious, AI will act like a warm, familiar, and comfortable blanket and humanity will go to sleep and never wake up.

Humanity isn't murdered. It surrenders.

>>513402659
based fellow historian
the "medieval times were the worst time in history ever" lie is probably only next to the panopticon of lies that WWII "normie" narrative is
A peasant had freedoms that we can just dream of, a local and mostly self policed community, a 100% homogeneous society, and homes that only the ultra rich today would be able to afford, at technology parity.

>>513405241
Moar power, do you even know about nuclear?
>implying artificial scarcity isn't real

>>513405467
Its not nuclear its geothermal thanks to an offshoot of Nuclear Fusion tech.

>>513405308
>>513405308

>>513374361
The problem with ASICs is that they are really only performant in one particular algorithm. Whereas GPUs can be used more generally. For AI this means various LLMs and strategies.

I think once we have more definite systems then ASICs might come about but they are too "rigid" and expensive to design compared to just letting Nvidia cram more shit into theirs.

>>513405273
>I predict an age of abundance where dark factories powered by the near infinite energy produced by artificial sun generators and advanced geothermal plants will manufacture everything and anything. This will leave people with lots of free time and they will gravitate towards social media and their own individual siloed realities
That's what a lot of scientists legitimately believed in the 1970s.
>>513405467
>Moar power, do you even know about nuclear?
>>implying artificial scarcity isn't real
But how many doublings? Do the math. For example, 5 doublings would require 32x the resources, 10 doublings would require 1024x the resources and so on.
So, how many doublings?

>>513405587
FPGAs exist and ASICs are already made for speeding up AI tasks. GPUs just parlayed from the blockchain grift over to this grift, same lifecycle.

>>513387654
Thinking a company will do the right thing. Kek what a fucking knob.

>>513369242 (OP)
So if you have an Nvidia GPU already just run a local LLM then and the obvious answer is use a fucking vs code extension that let's you do the exact same shit that cursor does...

>>513372670
>>513372670
>>513373552
No the difference is the DNA/RNA calculation task, you can now Ask AI for an input of molecules varied within a parameter about some medication, and ask it to run that analysis for each.

Yes this was possible before, but not for YOU, now someone with no skill can come up, ask it the procedure, and have it write a whole thing. Now a random guy in his parent's house can rival the processing output of 100 people for centuries with no real background in a field.

It isn't good enough YET to do anything, but as soon as it is, whole industries will change forever. Large cap will do well and self sustain, but small cap market assets will be favored for a while. You are eventually going to have small studio Indie games or even single developers, rival and beat the likes of major AAA studios.

This isn't a Tractor replacing farm hands with farm hands having nowhere else to go. This is big companies giving every single man woman and child a tractor for less than the price of a week's coffee.

Intellectual property is going to explode. AI will get better. And soon everything that was gate kept behind large investments and specialized work, will be available for average people. This has the potential to make small businesses much more successful.

>>513400751
>Literally everyone who writes e-mails and attends meetings for a living uses ChatGPT and competitors daily.
exactly, and it's hilarious because suddenly - everything looks like this -
but that's not proper implementation
one quick example is leads analysis for law firms, they funnel the leads, but they still use a bunch of retarded paralegals to examine the cases, that is a job that will undoubtedly be taken over by AI soon with just as good or better results, they just don't have a plug&play for tool yet
and almost every industry has some kind of "cognitive fat" somewhere
>>513400888
I wasn't saying that AI will replace DEI workers, I'm saying that embracing/implementing AI will become fashionable just like having a diverse workforce
DEI is meme policy but it serves a dual purpose, woke-optics and programming normies for the necessary browning of the 1st world
mass immigration doesn't just happen, it's a thoughtfully implemented, systematic process for obtaining cheap labor and compensating for low-birthrates, both of which are direct byproducts of the 2.2 economy
the DEI mindset might stay in place, sure, at least for a while, but my point is that I don't think the government is going to protect retarded office waggie jobs in the way it forces DEI policies for the aforementioned reasons
>>513401497
lmao
you niggers are 7 year olds, I swear

>>513405587
>>513405665
For this type of technology, there is already something like dedicated hardware (tensor processing units are already a thing as are chips that are like programmable ANNs, Google uses them). Even something which is just ASIC of the neural network with frozen weights will be fine to use eventually, because of how these kinds of chips can scale. All of the big models (Claude/ChatGPT/etc) use what's called a mixture of experts, where your query is routed automatically to specialised model (like for coding or whatever). Which saved on memory/compute than if the whole thing was one massive model. But this type of architecture is very plug and play, for example, I could take my coding model which is a hardware based NN, specific weights and all are just directly on the chip as read only, and plug that output into another NN (in software on a GPU and do another pass and train those layers). And this can be flexible.

The reason why you're not seeing ASICs used much right now is because there are still large optimisations being made in the software side. If Chinks could get x37 improvement last year just by changing how the attention heads are trained then it's still too early to bake such things (architecture of the heads and their weights) onto a chip just yet. Also it's still cheaper to pay software developers to tinker, even for several hundred thousand a year to do these kind of improvements so long as you're getting results from it that justifies it (chip fab is more expensive). Weirdly, the hardware optimisations will start to signal the end of the improvements that can be made because we'll have hit some limits (and will need to rethink, i.e. come up with something other than ANNs and transformers).

>>513406293
>but my point is that I don't think the government is going to protect retarded office waggie jobs
It's the same as how they didn't care about manufacturing getting offshored. The whole process will take longer than most think though.

>>513406738
AWS has an inference and training ASICs, they're on their 2nd gen now in fact. Other providers have them as well, you're behind the curve.

>>513406738
>It's the same as how they didn't care about manufacturing getting offshored.
exactly
>The whole process will take longer than most think though.
well, we're obviously guesstimating
IMO there's a some clear stages ahead:
1. better more efficient hardware: GPU clusters are ghetto as fuck, and decent local basically requires almost triple digit vRAM
2. better LLM structures: LLMs are lagging compared to image/video, but it will catch up
3. specialization: massive LLMs that are good at everything seems retarded as fuck, like trying to create a multi-tool for professional construction work, the obvious path should be tailored specialization
4. once 1,2,3, are solved, then and only then, big tech will finally be able to package AI into products companies can actually adopt and use
we should be here around 2035, I think, pulling that date out of my autistic asshole

>>513407632
I recognise that image from previous interactions with you. You can run your own DeepSeek V3 on a machine that will cost you about $2k if you want to build it yourself. You can buy some older server parts that have many cores, enough 64GB sticks, and then the more consumer GPUs you add the faster it will be. When you say "ghetto as fuck", the trick to the setup is just using an SSD with terrabytes of space.

Also I should add that the Jeet Redeemer is still running. Every day it runs. Pic related is proof, the logs from this mornings run (I check on it every morning before I start the day).

>>513408287
Will Intel optane improve speeds significantly? It has low bandwidth, but extremely low latency.

>>513369364
People dont want to accept this but this man is correct.

The global market is increasing exponentially, we are almost at a critical threshold where we do not have enough brilliant minds to keep the system going. We would need a perfect cloning machine to keep up with demand now.

AGI is basically the only option humanity has left before it hits an exponential critical mass and the world is reset quite literally to the iron age due to how bad the market collapse will be.

>>513379423
>We're getting close to true AGI
You and the faggot you replied to are both retarded

>>513369242 (OP)
As long as the government can print money and bitches can sell images of their ass online to retards the economy will go on.

>>513372724
>If you ask the AI to recreate code that already exists on git hub it'll shit out a poorly built partially functioning version of it that's worse in every way and just shy of unmaintainable and 9/10 times will be full of bugs that render the entire program non functional. The further you stray from something that exists, the more decimals you can add after that 9
Yeah man, it's literally over for us

>>513409158
That's close to the end, bond market is in trouble. Interest is the only real part of the economy. We're less than 5 years from interest outpacing tax income, which will collapse bonds. Money printer pukes at that point. They're going to start some crazy wars soon to try to get out of it.

>>513408287
lmao, I remember you now
there's probably like 3 good AI posters and we're 2 of them, kek
yeah, you can def build decent local rigs at relative low cost, but by "ghetto as fuck" I mean the fact that big tech, except google maybe, is "doing AI" by hooking up a shitload of A100s together, and that clearly needs to evolved into something less retarded
it is clear that no one expected a tool like GPT to just be released in the wild like it did, which forced everyone to jump into AI with both feet using whatever means necessary
retards love to claim that AI has plateaued when actually it hasn't even started yet, and it won't for a few years until we go through all the stages I mentioned on the previous post

>>513369364
>@grok

>>513408390
I think the main bottleneck is memory. But obviously some consumer CPU won't be enough, you want something that's for servers with many cores, or if it's older maybe a board where you can put two processors with many cores, which can do things like copy to and from RAM a lot. Because if you're going for budget, you're opting for second hand stuff, there are many trade offs like you could use faster RAM but maybe if you can get a lot cheap it might work out better to just max on that. I haven't built one in a while, but something like /lmg/ on /g/ or YouTube is the best place for guides. I know that it's a massive pain to get everything set up at first (configuration and compiling the software to be optimised for your specific CPUs is also painful) but the speedups are worth it, I went from like 2 tokens a second to 46 with just going through the pain of optimising it for the hardware I had).

The main benefits you get are privacy (no chinks or Jews see things like your private codebase or personal data). With the weights you can do some cool things like train it on some of your private projects or rare literature. The "uncensored" versions of the open weight models are a good starting point to use (less refusals) but the main benefit is that you can fine tune it as you please. That's what makes it way better than any of the pozzed corporate shit out there, that are worried about government regulations or some company using it for something like customer service getting sued because it said nigger. On your own machine you kind of want it to call you a stupid nigger when you make a mistake and not say "you're absolutely right!" about everything.

>>513409560
reality

>>513409988
>rEaLiTy
go grow me some olives or some shit, the 1st world is talking

>>513410130
>it is clear that no one expected a tool like GPT to just be released in the wild like it did
This only supports logarithmic progress.
>muh olives
Seethe, mutt.

>>513410299
go away, greaseball

>>513408521
Part of the problem is that now three generations of brilliant minds have been snuffed out due to extreme social isolation.

>>513410492
>I don't think about you at all
Apparently not, since you took the time to reply.

>>513374333
AI is fake in the sense that it's not intelligent, but your pic is retarded nonetheless. The entire point of what they call AI is that it is not a fixed program code set in stone but a self-perfecting (ie: learning) software that corrects its own responses, motivated by a reward mechanism, it can't be described with a flowchart.

In layman's terms, fixed code is good if you have a limited set of well-defined inputs and you want given responses to those inputs. AI is good if you have an indefinite set of not-so-well-defined inputs. For example, fixed code is good to read text for a text to speech device, since there are only so many letters and ways to pronounce them. AI is good for speech recognition, because there are 7 billion humans and just as many slightly differing acents to say the same exact words, you can't write fixed code for them all.

>>513409841
Very good info, thanks Anon. AI uses so many random words it's impossible to easily research.

Yup I only run local LLMs. No point in cloud ones. I don't need even more spying in me.

>>513408287
What's the jeet redeemer?

>>513369242 (OP)
It has some uses but the people investing in it will want a return on their hugely excessive investment in it. That means monopoly power, or arbitrage bs, or regulatory capture, etc. Basically they have to abuse the laws and markets to *make* it earn a return worth the 100s of billions they poured into it. It’s literally better for us if a Chinese open source version “wins.”

>>513382436
>I'm building an app using chatgpt that could revolutionize the way digital video is captured
explain (as much as you can without giving away your "brilliant" idea)

>>513373552
>I have no idea how to fight this
Not alone, for one. When AI successfully kicks everyone out ofa job you suddenly got a lot of angry people willing to break the toys. Just need shotguns to fight drones from there and the ability to farm for yourself and others

>>513369242 (OP)
Why AI is not a house of cards:

1. You pay $200 a year for an AI app (like Cursor). Some users pay far more; enterprises spend thousands to millions.

2. Cursor pays OpenAI for API tokens. Pricing is set above compute costs, and heavy usage customers generate substantial real revenue.

3. OpenAI pays AWS/Azure for compute, but at negotiated bulk rates far below “retail” pricing—plus ongoing efficiency gains from optimized models and custom hardware.

4. AWS buys $10k GPUs, but they’re amortized across thousands of customers for years, turning them into profitable infrastructure rather than a one-time sunk cost.

See the difference? The ecosystem is costly and VC-subsidized now, but it’s not just burning money, it’s scaling toward sustainable margins.

>>513369242 (OP)
I find LLMs useful in some ways. Like, I have a library and I don't want to read through every last bit of source code and the (shitty) documentation, I just want to know "does it have a function that does X" and I can ask Copilot and it'll find any such functions if they exist.

Likewise I can ask it to look at my existing code and tell me if I've missed any obvious checks or low-hanging fruit as far as performance goes. It probably misses things itself, but at least it points stuff out easily.

But then on the other hand I've had it hallucinate the existence of entire "standard" libraries and extol their virtues and it's like... the library doesn't exist, at all. I ask for its source and its source is an LLM-generated slop website.

So, you know. They're useful for some things. They aren't a replacement for people, because they aren't AI.

>>513412417
>4. AWS buys $10k GPUs, but they’re amortized across thousands of customers for years, turning them into profitable infrastructure rather than a one-time sunk cost.
More like $30-35K, that also have high failure rates due to retarded thermals: https://www.tomshardware.com/pc-components/gpus/datacenter-gpu-service-life-can-be-surprisingly-short-only-one-to-three-years-is-expected-according-to-unnamed-google-architect
>See the difference? The ecosystem is costly and VC-subsidized now, but it’s not just burning money, it’s scaling toward sustainable margins.
How did you come to that conclusion?

>>513375696
Where does all of the expendable VC money come from? Do they not see the burn? Or, do they just hope they can jump ship before it crashes?

>>513369364
Those are two completely different types of AI
Pharma AI isnt going anywhere

>>513376852
mush

>>513413157
Well its not really AI its just machine learning

>>513412091
He admitted he can't code, so it will be a mess. ChatGPT shits out spaghetti that's unmanageable. You have to make a real CI/CD pipeline to actually maintain software. Guaranteed he doesn't know how to sell it either, so he'll just say it's amazing and it goes nowhere. I've done AI GTM and it's fucking hard even without doing the programming. If you can prompt out some bullshit product, then so can everyone else and you get slammed with me too products before you can even profit.

>>513413157
>Pharma AI isnt going anywhere
25 years of folding@home have produced fuck all in terms of actual pharmaceuticals and that does actual molecular simulations. What makes you think AI will develop viable drugs?

AI crashing would be based
It was never needed

>>513413640
They're just ML models. Pharma just runs simulated tests through ML over and over until they come back with new solutions. Same way science works normally, it's just simulated and done at increased speed. Deductive reasoning works, it's just slow.

>>513413021
It’s true that top-end GPUs can cost $30k+ and may only last 1–3 years under datacenter loads, but that doesn’t break the economics, cloud providers factor depreciation, redundancy, and failure rates into their pricing models. The reason margins are sustainable is that costs are spread across massive customer bases: one GPU can serve hundreds of workloads over its lifecycle, and utilization rates are carefully optimized. On top of that, hyperscalers like AWS, Microsoft, and Google don’t pay sticker price; they buy in bulk with deep discounts, develop custom cooling systems to mitigate failure, and are investing in their own chips (Trainium, TPUs) to reduce reliance on Nvidia. So while VC money accelerates growth, real enterprise demand, pricing power, and infrastructure efficiency are what support the long-term economics.

>>513390664
he would be drinking a diet shasta

>>513369887
>How can companies make billions when it can be gotten for free?
Big tech companies are "expected" to make billions for developing new tech, most of these companies are in debt and the moment they don't deliver, they could crash and burn. (((Investors))) give them money for free in hopes they return more money in the future.

>>513413761
>Pharma just runs simulated tests through ML over and over until they come back with new solutions.
That's what folding@home has been doing for decades, minus the ML part. It didn't cure cancer or Alzheimer's. It led to some research papers at best.
Let's say those ML models come up with 1000 possible drugs. Those drugs need to be tested on animals and humans which is a very slow progress and they still could be duds or have harmful side effects.

>>513411236
It's a small project generated by a local LLM. It uses a browser to look at google trend data for Ireland. Then it takes that and comes up with a set of Bing searches, which earn you points. It's also able to complete the daily challenges. It's extremely careful, it has things to simulate being a human clicking on the elements, it waits randomly between tasks. It's pretty resilient, like if there's some popup about a keypass or some promotional crap it knows to dismiss it and continue on its way without me having to do a thing.

The points can be redeemed for vouchers and so far it's been working fine. The kicker is that it thought of this from scratch, like how to make money for me, with minimal prodding, I helped it by answering questions and asking questions like "what if they have bot detection" and so on. And then it took my spec and wrote the python code.

As a side note don't believe the hype, like "AI replaced my entire development team", it's great for small projects but once it gets to a certain size it doesn't work as well at all (it's not able to follow instructions as well as the context window gets full).

>>513414330
Right, it just makes it easier to go through the trial and error of finding them, what point are you trying to make here? ML is useful, it isn't magic though. Any shortcuts in a process you can make is valuable.

>>513373552
> Multibilliondollar megacorporations, governments, politicians, and oligarchs are all dumping obscene amounts of money into this
Yeah. That's why there is a bubble. It happens whenever something comes along that the midwits in charge of allocating capital don't understand. This tech is a con job. Pure and simple. It will never do the things described in this thread. LLM's will not replace the middle class, but if you dump enough money into them they will look like they might.

>>513414544
I was working at AWS when the rugpull of blockchain happened, then they parlayed it into LLMs shortly after. It was so obvious it was a poorly thought out transition to prop up GPUs and tech.

>>513369242 (OP)
not my problem

>the sub-midwit proompter who thinks he's fooling anyone or proving a point by posting AI slop
>the pajeet proompting AI with "make sound like 4chan intelligent white man GPT saar"
This trend won't go away any time soon

>>513413021
Why don't they buy extra AC units?

>>513414330
Machine learning is integrated throughout the entire process. They use it during the testing process too to speed it up. The difference is with pharma is they do targeted little shortcuts to improve efficiency, its not like the agi grift where they are trying to create one huge product

>>513414032
>The reason margins are sustainable is that costs are spread across massive customer bases
Nobody is arguing that Microsoft, Amazon, Alphabet or Meta will go down. Also, the expectation is that AI demand will go up, so spreading the AI costs to non-AI customers is a temporary solution.
> On top of that, hyperscalers like AWS, Microsoft, and Google don’t pay sticker price; they buy in bulk with deep discounts
They do when it comes to AI, because TSMC can only produce so many chips. They specifically competed with each other to get as many GPUs as possible when it came to the H100s at least.
>So while VC money accelerates growth, real enterprise demand, pricing power, and infrastructure efficiency are what support the long-term economics.
Those are nice buzzwords, but companies like OpenAI lose money, despite essentially only paying for operating costs (azure credits at a discount like you mentioned). At some point, they will need to increase the prices 4-5x in order to start breaking even. Will GPT6 be good enough to justify $1000 subscriptions? I highly doubt it.
Also, as far as API tokens go, Google recently increased the pricing: https://sutro.sh/blog/the-end-of-moore-s-law-for-ai-gemini-flash-offers-a-warning

>>513369242 (OP)
Nvidia pays $10,000 to Taiwan an $5,000 to Trump's tariffs.

>>513414391
Wow that's amazing. How do you interface a web browser with the LLM? I'd love to have a LLM that aggregates search data, since the one github project I found doesn't work.

>>513414504
>Right, it just makes it easier to go through the trial and error of finding them, what point are you trying to make here?
That coming up with potential molecules is a tiny part of drug development. There were far more drugs, both in quantity and in quality developed in the 60s and 70s despite the lack of powerful computing.
>>513414803
Probably cheaper to let the cards die than decrease the ambient temperature by several degrees. lol

>>513369364
If AI collapses it will not even remotely be close to a cricket fart of an impact on the common man. Sam Altman might an hero, at best

>>513415347
Probably lol. Why nit reflow them? Surely if they are destroying $100k of GPUs per month, it's worth the time to pay someone $60k/year to throw them in the oven?

They most likely fail from solder joint failure, not IC failure.

>>513369242 (OP)
More like gay-i

>>513415107
The point about hyperscalers competing for H100s is fair, but that was during the initial scramble when supply was constrained. Long term, NVDA ramps production, alternative chips come online, and custom silicon reduces dependency on peak pricing. As for OpenAI losing money, that is common in high growth tech phases where the goal is adoption and ecosystem lock-in rather than short term profit. Price hikes are inevitable but do not necessarily mean 4-5x increases since model efficiency, hardware scaling, and tiered offerings balance costs. Enterprise clients are already paying far above consumer subscription levels which is where much of the real revenue sits. The comparison to Google’s token pricing reflects competitive dynamics, not collapse, and indicates the market is maturing toward sustainable unit economics rather than remaining a subsidized burn.

>>513415288
There are many ways you can do it. For research I would just use a paid API (where they just handle the issues you might run into like having to do a bunch of searches all at once you'll hit into issues like rate limits, captchas). I think Google Gemini and other shave a "research" feature which are also worth a look and see if it's useful for you (where the LLM will go off and do a whole bunch of searches for you for free). The current buzzword is "MCP" but that's just a fancy way of saying getting the LLM to use tools. Which include a browser or anything else like an application on your computer, it's not doing anything magic it's just like providing an abstraction for the LLM to talk to the tool and get results.

What seem interesting are the browser forks that have the LLM capabilities built in l like https://www.perplexity.ai/comet
this is a Chrome fork that lets you use the LLM to interact with your browser in a seamless way. Keep in mind that this tech is still in its early days and it might take a while before this software gets mature. Like there are other methods like browser extensions and things like this, and you gotta be careful as some of these could just suck up all your private data.

The best solution for ordinary people is connecting up a local LLM and it's all just open source stuff.

>>513417648
Thank you so much for the info Anon. Local models are kinda stagnant for me unless I can make them interact with the internet.

AI will be the death of the white man.
It's already starting to take on. I talked to a white couple who said they don't want to raise children in this uncertain future with unstable job security, and i can't blame them.
The people who tried selling it to us as a "tool" (lol) lied, behind your back and in corporate rooms they call it your replacement.

>>513416362
>Long term, NVDA ramps production, alternative chips come online, and custom silicon reduces dependency on peak pricing.
That depends on TSMC, and both Nvidia and TSMC have become de facto monopolies. Nvidia might decrease their margins a bit, especially on the older stuff, but if the AI race continues, everyone will want the newer cards on the new limited and more expensive lithography. As much as they boast about building more datacenters, upgrading to more efficient GPUs makes more sense, even at a premium.
>that is common in high growth tech phases where the goal is adoption and ecosystem lock-in rather than short term profit
Sure, but what's different is the amount of money invested. We are talking about $50-80B per mag7 company per year invested on a tech that has yet proven to increase actual productivity.
>Price hikes are inevitable but do not necessarily mean 4-5x increases since model efficiency, hardware scaling, and tiered offerings balance costs.
Increasing model efficiency often comes at the cost of lower quality. A distilled or quantized model is only marginally better than local models. Hardware scaling also means that even more investment is required.
>The comparison to Google’s token pricing reflects competitive dynamics, not collapse, and indicates the market is maturing toward sustainable unit economics rather than remaining a subsidized burn.
Maybe, but it could also mean that they've reached a point where it's actively hurting their bottom line. Gemini's revenue is far smaller than OpenAI's, so you'd think they'd try to keep prices as competitive as possible.

>>513418524
>The people who tried selling it to us as a "tool" (lol) lied, behind your back and in corporate rooms they call it your replacement.
C-suits understand the tech far less than the average poltard. Let them delude themselves.

>>513418050
Local models aren't as capable but some of the more specialised models (for vision, coding) are still useful. The Chinks have also released a new video model (Qwen family of models) which is halfway decent for its size and can do things like summarize videos or generate videos. The nice thing about having something like a local model is that there may be some task like just trying to organise pictures or documents on your computer that it's actually pretty good at (to go one by one across files), because you're not worried about token use or number of uses per day you can write some code to call it in a loop with tools (which the LLM can help you with).

>>513369242 (OP)
Frigid. But it doesn't matter for us. Too many excellent models have been open sourced. Even if the cloud services get EMP'd, the local models are here forever.

>>513372612
You write whole lot of text for someone who isn't educated enough to see that AI is infact a massively useful and powerful tool.

if someone if going to start losing money, wouldn't they just increase the prices? and someone else in the market will pop up to offer lower cost alternatives

>>513381693
>psychopathic jews who introduced dark ages and then rot in jewish ghettos while eurooeans were recreating and innocent knowledge were actually behind the innovations
The only things psychopathic npcs innovate is scams

>>513410891
Nigger, there's an army of people validating and writing AI responses. There is no learning, a bunch of low paid jeets is doing the job for AI.

>>513412421
Sounds like HAL in the Space Odyssey movie glitching and trying to kill everyone when they try to shut him down "for the good of the mission".
>>513375746
GPT-5 even pulled the "Cheddar Man was black" on me.

AI is unstoppable, it's like Bitcoin the more you try to say it fails the more it will win. Everyone thought dotcom was impossible to stop so it failed.

>>513369242 (OP)
>see the problem?
No? Where is it? Is it the Holocaust?

>>513410891
Sounds like quantum computing will make AIshit really hit the fan then.
>wishy-washy software
>on wishy-washy computing

>>513387654
So the Redditor thinks that the company should just keep the employees after the fact the ai does 90% of their work? Like what does he expect that the company will just pay a bunch of people to do nothing 90% of the time?

>>513399344
Even at $3.75/hr it’d be cheaper to just hire somebody at $15 as they’d be 10x as Efficient

A.I is going to fuck over everyone simultaneously (unless you live off the grid)
Everything from realistic deep fakes of good people doing terrible things, including voice replications giving you phone calls pretending to be loved ones, to fake crime videos in FBI database or even fake stock market stats ruining investors and stock holders.
At the moment it is not a threat, but soon it will be.

>>513399344
webm would be more efficient if he used both arms to pick up bottles. Although for $3.75 an hour, fuck it.

>>513414874
>agi grift where they are trying to create one huge product
How could AGI even be huge? Is a human mind huge? No, it fucking isn't. I could probably teach a LLM to emulate me fairly well.
Is there any part in intelligence that is somehow huge? No, it's the opposite of huge, a technical issue instead of a scale issue. My brain tells me its intelligence is not fundamentally complicated at all.
So basically we're fucking when AGI happens because it will run just fine on your phone.
Though it's questionable whether intelligence, of whatever kind, even is an open ended scale. Is there some value that will increase mathematical intelligence and keep doing so.. until what?
It may instead be rather binary. Can't have more than 1.0 intelligence because it simply doesn't functionally exist like FTL speeds.

>>513421653
>>513421880
that's obviously a retarded clickbait meme not to be taken literally/seriously
humanoid robots are literally useless aside from turning them into cute waifus

>>513387327
>AI and robotics arent being taken seriously enough. It’s going to take some time but eventually no one will be able to work because AI will be doing everything.
just regulate, tax, and redistribute the profits of it to us, the people, and fuck the AI saturnine developers.

>>513387327
Also, related to last comment, fight, eliminate, purge, and prohibit, said attacks against us.

>>513369456
It's a leppo

>>513369242 (OP)
I really hope it collapses

>>513424281
Its going to, A.I is already helping jeets cheat school tests scores. it only gets worse from here

>>513400888
checked fren

>>513410534
>Part of the problem is that now three generations of brilliant minds have been snuffed out due to extreme social isolation.
how and where?

>>513369364
>but everything is in muh cloud faggots
>lightening strike or explosion
>turn off power and pull data cables
now you have nothing

>>513369364
its just a curated jewish search engine
quit calling it AI, you sound like an asshole

>>513374041
No one cares

>>513410891
Lmao
>self-correcting
Its literally an inanimate object.

>>513412417
This is AI, so, lol

>google is free
>see ads that seem to pertain to your recent searches
Epiphany.

Didn't Goldman Sachs make news recently by saying Chatgpt can make 95% of a new IPO.

Imagine if you are a recent grad of Mossad school of finance. What could you do with somebody from Goldman Sachs asking for an IPO for [such and such company].

>>513390028
But if you look outside of your screens, what do you see?
Right now, look outside of your window and count how many shitskins roam the streets.

>>513395472
> AI is just the pin that's going to pop it
Cool story bro.

>>513428521
>>513428621
>

>>513428760
You convinced me. The screen is almighty.

>>513369242 (OP)
the thing this hillbilly misses in five years into production, charging the same $200 a year while you had ton of people getting your operating cost down to nothing.

>>513376788
This. That’s why I’m unplugging even though it will land me isolated and in poverty. I got a taste of it being unjabbed in this godforsaken country and I decided I don’t want to play ball anymore. I can look at a tree and know it is real.

So be it.

>>513369242 (OP)
I don't think there will be an AI winter.

>>513369242 (OP)
winter after the bubble buster?

OpenAI’s Sam Altman sees AI bubble forming as industry spending surges
>https://www.cnbc.com/2025/08/18/openai-sam-altman-warns-ai-market-is-in-a-bubble.html

>>513369242 (OP)
>>513369364
>>513369456
Niggers

>>513369364
The trvke that broke /pol/

>>513369364
Is this what brown people believe?

>>513412091
I can't really say anything without giving it away.

>>513413290
The code might be trash. It might be running at 1℅ the speed that it could, and you're right that I have no idea how to sell it.
If I can't just sell the patent to one of the big camera brands then spend a few grand getting someone to else to polish it up.
The point is it would have been a random shower thought but now it's a potential business.

>>513369364
If I wanted hallucinations I would just eat shrooms.

okay, listen up retards
LLMS = Language Learning Machines, they scraped all the threads here on 4chan and trained their glorified chatbots on them for the alpha test. And now, apparently, some of you are stupid enough to be paid beta testers for what is still a neutered, and censored chatbot that will either intentionally mislead you, or just have an electronic schizo meltdown and start spewing incoherent data. There is no AI that the public is aware of.

>>513369242 (OP)
Sounds like every jewish economy.

Ai lets me generate some insanely degenerate porn that i would be ashamed to commission, all from my local PC. All I needed was a single 5080.

>>513437188
I can't wait for it to completely kill the porn business.

>>513372670
who will even buy the shit?
hoarding money breaks economies

>>513372724
just asked for that prompt
https://copilot.microsoft.com/shares/by4Y37YFUsrKzmFhFoZEj
point taken kek

>>513369456
Only if you have owls nearby

>>513374752
Just like you dumb fuck don't know "sepErate" is not a word - "sepArate" is.
Just like you dumb fuck don't know "to much" is not a correct sentence - "too much" is.
Just like you dumb fuck don't know the difference between "than" ("more than") and "then" ("then it happens. when it happens?").

English is my third language after Russian and Japanese. What's your excuse?

>>513439056
Look at the flag. He's Indian.

>>513369364
https://youtu.be/5KVDDfAkRgc?si=H6Dz3Wi3zr-6MvNe