[ Home ] [ wiz / dep / hob / lounge / jp / meta / games / music ] [ all ] [  Rules ] [  FAQ ] [  Search /  History ]

/lounge/ - Lounge

The Wizard's Lounge
Email
Comment
File
Embed
Password (For file deletion.)

  [Go to bottom]   [Catalog]   [Return]   [Archive]

File: 1781000287008.jpg (910.35 KB, 2048x1152, 16:9, 4366525252.jpg) ImgOps iqdb

 No.324033

I don't think AI will be around for long, and not for the reason people think.

AI is very old technology, the LLMs utilized are from the 1950s. When computers were first invented they tried using artificial neurons that learnt (the Perceptron). The problem was back then memory and scaling, you could get one node to learn but by the time it went to the next it started to forget. The hardware wasn't there. Because of this AI bots relied on binary tree if/else statements and weren't programmed to learn for decades.

All that happened in the 2010s was that hardware became available to stack LLMs on top of each other. The technology behind AI is veeeeery old, it's a hardware problem. If you're some nerd that reads into tech innovations you'll find the same is true for most inventions. Wood pulp paper wasn't suddenly eureka invented in the 19th century, it's that the steam engine made the processing of wood through mulching possible. The automatic threshing machine wasn't eureka invented, it's that there were dozens of invisible small innovations that fed into it that went unthanked, like the complex mathematic epicycloid calculations that went into the gear functions that some mathematics nerds and clockmakers perfected decades ago.

Anyways, the actual supply chains that are currently feeding the current top of the line GPUs and chips are unsustainable. They're spread across so many countries and it's like a precarious inverted pyramid, each part of the chip fabrication is a monopoly because it's stretching human ingenuity and ability to its absolute limit. And for every part of the input (the lenses, the robotic arms, the chemical acids, the light diodes for the lasers, etc) there's like another 5,000+ companies making bespoke tools for specialists. It's quite literally not cost effective or arguably even possible to scale production further. Like the Concorde and other supersonic jets, the rising costs are not due to economic constraints, but human. The concorde went out of service, and likewise AI will too.

The AI bubble will burst once they try to recuperate costs, once they find out that businesses don't want to pay $100,000 for the slop, they'll get a government bail out. The government will bail it out thinking they're getting a palatinr tier surveillance engine. This will turn out to be a white elephant because the AI data centers built will require replacing every 3 years due to the GPUs burning out, costing hundreds of billions annually, further feeding into the debt crisis. There will be too much nosiy smog of irrelevant data for it to do palantir tier Klaus Schwab surveillance, so that will eventually be abandoned after governments realize they were swindled into an unproductive asset. The chip market will then crash, and with it the ability to make expensive top of the line chips at a reasonable price as the market isn't there.

Eventually people will realize the GPU/CPU price problem isn't going anywhere and the high costs are here to stay, and start shifting to older, cheaper nodes. Smartphones will largely begin to become a luxury $3,000+ product, bricklike smartphones will become the common option. AI will become too energy expensive to host on computers and become a luxury product, as tech will regress to about 2012 level hardware.

An economic recession will set in and the advertising revenue will dry up as nobody is buying shit, and all the accumulated tech data on everyone will become regarded as too expensive to host and be deleted. It was only ever hoarded as a speculative asset. Big tech's business model will collapse and videos/pictures on social media older than 6 months will either get deleted or get lowered in quality.

It will be like the Concorde, a supersonic jet launched in the 1970s that flew significantly faster than all passenger planes today. It got retired after it was seen as too uneconomical after decades of service. AI will be like that, it will start to be regarded as a luxury product using chips we can no longer manufacture, then after 2 decades it'll disappear.

 No.324034

so you're saying the world will regress one day?

 No.324035

>>324034
I'm saying yeah, very likely. A globalized vertically integrated supply chain struggling to break from a market monopoly, with literal superpowers economically unable to condense the entire production within their borders, yeah, it's unsustainable.

A single major geopolitical crisis would topple that fine inverted pyramid of expertise and it'd arguably take another 5-10 years to scale back up again once collapsed, even if it could scale up again. And even if it doesn't collapse, it's flatlining. The shit about "specialized architecture" is just computing reverting to how it was before the silicon chip got invented and they focused on universal scaling. Moore's law is well over at this point and the 2nm-3nm shit isn't true for the entirety of the chip's wafer length, it's just specific parts and it's used to market improvements that don't exist.

Economic specialization is like an onion, one machine tool leads to the production of finer machine tools. This is likely the final reach.

 No.324041

>>324033
>as tech will regress to about 2012 level hardware.
I'd be okay with this if it also led to a push for optimisation on the software side. Most theoretical performance improvements I'm supposed to see from newer hardware don't translate to real world usage because operating systems and apps have become more bloated and gobble up the overhead.

 No.324044

I believe AI is the ultimate enslavement of humanity…
https://www.bbc.com/news/articles/crrpggegwe0o
I don't trust them at all

 No.324051

An

Ai expert who probably wasnt even aware of Ai any, pre 2020 lol.

Ai is now unstoppable,

by 2035, possibly sooner, Ai will become smarter than smartest person to of ever lived.

At said point ALL white collar occupations will be automated. Ai is now, MID 2026 startin to dip into occupations of:

scientists, doctors,

its solvin unsolvable equations. Smartest people already see writin on te wall

as to manual labour already a move to build robots to do it.

tryin to come ere of all places to plant

le Anti flag lame and weak

 No.324052

>>324051
What the fuck is this typing?

 No.324053

>>324052
Beyond Reddit Spacing

 No.324055

>>324052
African. And they better get smart about it before that continent is just one big building for servers.

 No.324212

The latest nodes of semiconductor manufacturing have escaped monopolisation despite certain attempts to limit production.
This tech isn't going anywhere.
There's also lots of experimental tech beyond EUV.
Worst case nothing beyond EUV becomes cheap enough to mass produce and the eternal spec sheet escalation grinds to a halt.
AI is software and suffers from scamming/bad marketing. Still you can't deny it's at least slowly advancing even if it isn't the revolution it's marketed as.

 No.324214

More like taiwan will fall under chinese control over the next years, the US will not be able to compete and will become like an african nothing country where crime and rioting is a common thing, needless to say it will not be able to fabricate at the level of china

 No.324247

it's funny that AI vibe coding is encouraging sloppy coding that eats more RAM, while simultaneously costing more RAM. Eventually this is going to be like the oil market where it will become economical for new players to enter and undercut everyone. That's probably what China is working on.

I don't think this bubble will burst. The rich will make it go the way of self driving cars, quietly forgotten. They got way too much invested and don't dare admit they made a dumb decision. China doesn't have to win militarily, they just gotta take advantage of this dumb shit to crash our economy.

 No.324277

>>324035
What about China?
The got everything they need on paper.

>>324033
Maybe I'm a retard and missed your point entirely, but again, why would the whole world or AI in general go the way of the dodo?
Consumer grade hardware which is plenty and has a wide market can run insane stuff now.
The whole palantir thing doesn't need as many resources as you seem to think either.

I could see chip making as a whole being hit if a couple big companies get hit in some major war or something.
Idk I'm just some retard anyways. Don't see this happening.
Even if ASML and taiwan as a whole gets obliterated I think a lot of other companies now have the means to produce such hardware.
Maybe not top0.00001% level, but at this point isn't computing already way stronger than we need anyways?

Smart people can just come up with better code, better math etc.
Chinese are already trying to kill USA AI with their own free/open source stuff and they seem to be succeeding in some ways showing people that paying USA companies to hand them all your business data might not be a good value proposition…

 No.324328

>>324247
Self driving cars… forgotten? Lel wut?

 No.324346

I don't have anything to add OP because I think the same is going to happen. It's just a matter of when.
I'm more interested in the (for now still theoretical) quantum hardware.

 No.324350

>>324346
>quantum hardware
Fake except will be used as an excuse used for crypto money stolen (all of them will become worthless) lol

 No.324369

>>324033
Same here. I've been following this guy Ed Zitron for a few years, he's been one of the first LLM skeptics that got traction. Think he even appeared on some american news stations recently. LLMs haven't proven in any capacity that they are actually better at any tasks than humans in the real world.
Benchmarks are set up specifically to measure arbitrary work loads that are made so an LLM can understand them. They use metrics like "lines of code written by AI" because it's impossible to measure the quality of said code or how maintainable it is in the long run (though experienced coders have talked about how much of a cluster fuck it is due to limited context windows and lack of awareness for things outside of the current task). LLMs are incapable of generating new solutions for anything, just mix and match existing stuff that sounds reasonable to everyone in the lower 50th percentile of knowledge in any given field. Hallucination will not be solved as it's just a fundamental quirk of how these things work. The curve of improvement is starting to flatten, just throwing more training inputs is no longer producing better outputs. Datacenter build outs are already stopping, big tech companies pulling out of their commitments or putting them on pause because there's no demand. All of it is just propped up by the years worth of backorders between AI companies, chip manufacturers and datacenter builders, which is a circlejerk of venture capital investment.
I don't know if I truly see a regression in tech but maybe that would be nice. It's not like the increase in raw hardware power has been used very well by the software in most cases. But since the chip fabs are sold out for the next few years we will definitely see stagnation on the consumer side. Maybe if enough PC parts manufacturers die out in the meantime due to lack of demand in the consumer space (((they))) will try to make everyone shift to cloud PCs. All of this LLM shit is so retarded.



[Go to top] [Catalog] [Return][Post a Reply]
Delete Post [ ]
[ Home ] [ wiz / dep / hob / lounge / jp / meta / games / music ] [ all ] [  Rules ] [  FAQ ] [  Search /  History ]