Thread 107052346

AI IS NOT A BUBBLE!

STOP COMPARING IT TO THE DOT COM BUBBLE!

SAARS, may we have a better source other than twitter's anus?

>>107052346 (OP)
Why does he need to tell this?

>>107052346 (OP)
Well it's much worse because during the 2000s the economy was doing fine overall.
This is going to be more like 1914 / 1929

Great now he officially hexed it

he doesnt want it to pop before hes out and have it be his legacy

>>107052346 (OP)
Cry about it in your BUBBLE.

>>107052346 (OP)
AI isn't going anywhere

>>107052570
its already everywhere
shatbots are shatbots doe
what theyre actually capable of is kinda written on the tin
the bubble is that people counted on chatbots to compute the answer to everything
it wont disappear but oh lawd is there a market correction incoming
and its gonna be glorious

he's right, it's much, much worse

>>107052713
>speculation sink
>the fucktards who pump the economy into being based on a whole heap of nothing, one of the major causes for inflation are gonna get the cookie jar lid slammed on their fingers
from a macro-economical standpoint that bubble may actually be a good thing

>>107052346 (OP)
>Technology is viable, but costs are sky high while revenue are minimal
It’s actually a worse version of the dotcom bubble

>>107052673
i think chatbots will largely go the way of the dodo because theyre way too inefficient, but smaller ML projects will remain as they have for 15 years
always found it funny when people said engineers are going to be replaced, ignoring that no one is going to let an AI sign off on a project like that for decades, solidworks has had ML and automated matrix math for 10 years

>>107052346 (OP)
>>>/biz/

>>107052346 (OP)
he's right
AI bubble is not the same as .com bubble
Probably worse

HAHA BUBBEL MAKE BOOM

>>107052876
chatbots are a good interface.
you can build/train em to emit labels which then get fed into various frameworks

and ai engineering already exists
its just that people who use it dont share and its a serious technological investment to build one
https://en.wikipedia.org/wiki/Model-based_design
https://www.youtube.com/watch?v=VdG4gUTowXc

picrel has been designed by ai.
not a chatbot, obvioulsy
the pipeline goes like this:
engineer gives a set of constraints: dimensions, materials, (manufacturing methods in a generalized engi ai)
the ai then iterates bw generating a model conform to the constraints and testing it, attempting to remove material each iteration
so in a way it is gen ai, it even uses something akin to a generation/evolution algorythm, but without the statistical aspect of ML. its fully symbolic

>>107052346 (OP)
That Powell fella seems on the up and up! Better believe what he says.

I'm 25% cash but the market keeps booming. Is diversification for chumps?

>>107053062
this is not how booming looks like

>>107052958
NOOOOOOOOOOO THE US GOVERNMENT'S $1,000,000,000,000 DONATION TO SAM ALTMAN!!!

>>107052758
No they aren't. We're all going to pay to bail them out.

>>107053072
My 1 year returns are 17%. The market is booming.

>>107053230
i dont think its realistic
were not talking several billies here
its trillions
at least a couple, bc applel, ngreedia are vastly over-valuated + openai & co if they go full bankrupt, thats something like 3-4 trillions
and theats before counting the derivatives
which get speculated on too, each layer of repackaging gradually adding a layer of speculative value

its gonna be glorious

>>107052758
When has a bubble ever been a good thing? The AI bubble bursting is going to kickstart the Great Depression 2.0

>>107053260
>bases prediction on past performance instead of looking for signals that allow to predict the future
dumb money. but enjoy yourself, somebody has to buy the bags after all

>>107053297
>When has a bubble ever been a good thing?
id argue- every single time
the great depression hapenned because a whole nation didnt do their due diligence if im not mistaken

the definition of a bubble is speculative value getting annihilated
its shadow value that skews the valuation of a fiat currency into inflation
i think its a necessary evil + a wake up call

>>107053304
Not making predictions, just talking about right now. If you want to make predictions: I'm 25% cash and regretting it what are you investing in to profit off what's coming?

>>107053346
im supposed to do 5-30 mins trades on crypto, and ive been on a sabatical year since 2 years now, so im not actually informed about the current state of the market, i dont follow the numbers, and i play in the junior league anyways
buuuuut
gold is always a refuge in a period of turmoil
and it will probably spike once the bubble will pop
beware of paper gold though, because i know theres a swindle with gold papers (etf/etc?)
once the demand will explode people will actually look into the actual availability of gold and all the swindles may come to light.
theres a scenario, and i think its moderately likely where gold explodes twofold: first bc suddenly everyone wants to hedge against the bubble, then once the swindles come to light, and solid gold value goes through the roof for a moment

>>107053304
>predict the future
KEK

Gold: probably safe during crash
Real estate: in a bubble already, would avoid
Bonds: treading water, will survive crash
Cash (high yield): Same as bonds
There has to be something better. VIX?

>>107053346
>>107053382
dont go all in on gold either
diversify. gold will overheat at points, and there will be corrections
diversify away the fluctuations
maybe invest in some hard value like housing, idk, land

idk man, do due diligence
but gold might be a thing to look into

>>107052346 (OP)
>>107052363
Once OpenAI shutdown then there is no customer support and the skeleton crew who is still working at support will be overwhelmed or crashes the system because 500 calls to one center is not a good idea.

Emails for Openai will be sent to real people who may take years to finish their job.

There will be no more updates until corporate deals with angry employees who were fired or need to be retrained. New hires to replace the old will slow down updates.

So called artist will try to use local machines AI but soon realize how much money they are losing to electricity. They will never achieve the same results as openai because the real open source models suck.

AI is a bubble, it will hurt only USA who over dependent on it. This is your American economy problem, not everyone else.

>>107052346 (OP)
How could it be a bubble when demand is far, far outstripping supply?

>>107053426
simple rule of thumb:
you take the netto yearly revenues of a company
and you divide by its valuation
this gives you how many years would it take to recover your investment
if you were to become the sole owner of a company at its current valuation

then you compare to the rest of the industry/financial landscape
when a company yields less than a savings account
yeah, its a bubble

Frankly, if you take away my ability to use AI, my life will be severely impeded. Everyone I encounter also talks about how they use AI to improve their daily life. It changed the world and is an essential utility now like electricity and water.

I'm not going back to google searches
I'm not going back to writing boilerplate code
I'm not going back to authoring PDFs from abstract data
I'm not going back to photoshop or fucking gimp for memes
We're just not doing that shit anymore

>>107052346 (OP)
He's right. It's much worse.

>>107053443
Listen, very carefully: change, or you will fail miserably.

>>107053458
i thought i already said im a high-ish frequency trader, so i have a big tolerance to risk
and im quite versed in risk mitigation techniques.

and even im gonna say: ai is overheated as fuck. limp dick / 10, would not buy atm

>>107053458
>>107053469
cont
smart money invested in ngreedia & co when covefe started and people got gibbes
now that train has sailed, buying into it now would be buying high to sell low

>>107053469
lel if anything it's underbought because of fud like this
you don't understand tech trends

>>107053535
>looks at failure rate of vc capital
lolle
yeah i dont understand the hype
and thats a good thing. i dont intend to change that

>>107053535
>sar please buy the AI very smart

>>107052346 (OP)
>everyone gets into AWS cloud services until live is jacked up and everyone needs to scale back and find cheaper alternatives
>this definitely won’t happen with all the AI trash we’ve invested in for 2 years tho
Job security, simple as.

>>107053552
I look forward to taking your bag in 3 years lmao

>>107053625
ill diversify your mom away in exchange for goods and services

>>107053449
Time to let go of your toy and face the future.

>>107052346 (OP)
Yeah, it isn't the same; it is multiple orders of magnitude worse.

>>107053753
not gl its gonna kinda suck if/when chud gpt dies
its a powerful search engine, you can give it complex queries, you can partially bypass the paywalling of scientific documents
i kinda like it

>>107053449
You can just run qwen3 locally. It only needs like 4GB of RAM. There's no way this goes away now.

>>107052346 (OP)
Based. Total luddite death

What can you even do with gigantic GPU farms running LLMs if it turns out there's no sustainable commercial demand for LLMs?

>>107053881
We're about to find out.

>>107053881
crypto?

>>107053831
>you can partially bypass the paywalling of scientific documents
I can fully do that without using LLMs; no, I don't do it by relying on Sci-Hub or Libgen.

>>107053881
>>107053889
you sell em
we might see a tsunami of heavily used enterprise gpus, but at a bargain price
cryptobros would have a field day, thats for ones
maybe users are gonna massively buy em for local models purposes
maybe theres some agent-class ais being developed to become opensourced, that would be based
fukken, who wouldnt like to have a jarvis with voice recognition and generative abilities running locally

>>107053914
i didnt really think about that issue
because using an llm to do that is just so much more convenient
and i dont need the actual texts either. in my usecase it boils down to finding a figure, or getting a high level overview
an llm with browsing capabilities is sufficient for that

>>107052514
well fuck.

>>107053928
>agent-class ais
Lol, lmao.

>>107053980
???
llm into labels
the labels are actually commands
slap a speech to text on top of it for mechanime maximization purposes
this could actually increase productivity.
and would be cool as fuck

Don't worry, they just need another few trillion dollars and then they'll finally be profitable!

>>107053980
>>107054009
aaand with some generational capacities bc its still an llm
its gonna happen
and its gonna get opensourced at some point
but i dont want to wait 20 years for that to happen :/

>>107054019
they hit the wall
its completely delusional to attempt to scale up at this point
its hard maths.
consider two tokens, in the context of, well, context size: a and b
now consider a unit of operations, an abstraction of the computing needed to build a two token correlation map
this gives us 2 token context size => a->b,
1 unit of computation.
now lets consider 3 tokens, a, b and c
no we have a->b, a->c, b->c
3 units of computation, 3x the compute for a 50% increase in context size
the computational needs grow pyramidally with the increase in token size.
this is the problem
and beyond a certain point, its just completely hopeless

>>107053928
>you sell em
idk they might shred them right before their loans get called. You're overestimating public corporations.

>>107053842
Maybe in 10 years? I only use the most advanced models, and the free shit can't compete.

>>107054201
why would they do that?
if i were the lender id be pissed that my investment is a total loss

>>107054134
trump just needs to print goyillion dollars and the growth will look linear again

>>107053881
Pointless question when the growth of AI has been faster than the growth of the Internet itself.

WE ARE NOT
GOING BACK

>>107054201
>>107054253
cont
in 2008 they didnt destroy the houses
the flooded the market with em

>>107054254
thats not exactly how that happens.
theres extra steps to it and not all of the printed money goes into ai

>>107054280
spend on ai grows exponentially while results are lower and lower, they need to start printing in hyper mode or the whole economy collapses

>>107054305
just few more quadrilly bros

>>107053928
You're probably gonna see companies just buying hardware, dismantle if and sell the components to MSI so they can make regular consumer GPUs

>>107054305
nah
i think you underestimate the size of the us economy
+ letting the economy collapse is macroeconomically the right thing to do
+ that would be a big win for the protestant gang bc that would decimate the jewish gang's assets

>>107053449
You are a subhuman who outsourced his ability to think and create and thus has no reason to exist.

>>107054336
its refurbished then
it would be a major scandal if they tried to sell that at "new" price
i dont think its worth the fallout
if they just refurbished the cards and sell them at a decent price...
that would be nice
or maybe something in the middle: they sell refurbished+downsized hardware to poor markets like india

AI winter has already started. We might have another breakthrough in 10 or 20 years.

>>107054352
inb4 kike altman shills: AGI race is too big to fail, everyone needs to bail them out

>>107053297
Post-industrial, financiarized economies need bubbles to keep the economy growing.

>>107054352
The Jewish gang are going to sell all their paper assets to goys just before the cataclysm hits, anon. Like they always do.

>>107054439
they already did, your retirement money is already in it goy, time for bailout

>>107053449
Why do you need to do any of those things? And boilerplate code has nothing to do with AI. Most IDEs have been able to automatically write boilerplate code for like the past decade.

>>107054378
i think its too big to bail out
its not only the ai companies that would need a bailout
but everyone who didnt sell in time
and some products are derived from ai investments, which will drop the trust factor of many investors which will mean massive withdrawals and speculation down for many investment funds
i dont think everything is gonna happen at once though
its hard to imagine a scenario where one day, everyone sells everything ai related
i think well see an inflection point, where people who buy into ai will no longer outweigh people selling, which will prolly look like a plateau for a moment, then a crash
most ai companies arent even public yet
so theres no curve to follow, properly

>>107052346 (OP)
It's not like the dotcom bubble because the internet provided something truly revolutionary.
AI provides a financially suicidal autocomplete that takes billions of times more processing power to enable retards to generate buggy unfinished code that they can't fix without more prompts

>>107054439
they need bailouts every 5 minutes.
i think youre greatly overestimating them.
people who got fucked over in 2008 didnt do basic due diligence
bailouts ensued.
they influence immigration policies to maintain a predictable market growth. bc theyre retards who can only hodl.
yeah, they have money etc, but give me a familial fortune to start with, plus insider trading information, and youre gonna see im a second warren buffet

>>107054366
Please use a chisel and hammer to build the Empire State Building. Unless you've outsourced your ability to think and create to machinery.

>>107054499
Oh, so you're just really ignorant.

>>107054134
In terms of brute force compute, memory bandwidth, etc, sure. We’re reaching the point of diminishing returns, but there is a long way to go as far as optimizations are concerned. Hardware architecture will continue to evolve as well.

>>107054570
ye ofc but its not happening fast enough and thats how big tech ends up considering the acquisition of nuclear fukken reactors
multistage turbokek

>>107054570
>>107054600
aaaaan even in terms of optimization theres only so much you can do when the root problem is exponential-like
what is needed here is a paradigm change
which means ETA: unknown
could be tomorrow, could be 5 years from now

>>107052346 (OP)
>>>107033485
Oof. Bubble so tight they're making photoshop's clone free to shove AI into it.

>>107053260
You literally described a bubble.

>>107054375
At least there's still lots of ingenious work to do when it comes to integrating LLMs as language interfaces to expert systems and gaming. And it will stay a nerd thing because normies can't be trusted with not killing themselves by accident due to some hallucination.

Big tech investing in nuclear power is a positive thing.
>>107054634
Sure, AI/ML won’t become O(n) any time soon, but we’re still scratching the surface on optimizations. Algorithms today do a ton of unneeded computation, the hard part is figuring out what shortcuts you can actually take. Look at how fast transformers and diffusers are improving. You’re right that we need a paradigm shift, and I think some kind of hardware architectural breakthrough is what will happen. The real answer is finding a better way to avoid time complexity than “just add more tensor cores, bro”