Thread 106234928

is AI a meme ?

We already knew about this problem with neural networks DECADES ago, weird how people conveniently forget

>>106234928 (OP)
2 more years!

>>106234928 (OP)
>>106234978
>This is like the Manhattan Project
>I just need $500,000,000,000

>>106234928 (OP)
>2 more years
AGI is a marketing meme to scam investors of their money

AI is not a meme though. It's pretty useful, even with local llms.

>smart high schooler
ok class how many r's are there in...

always was

do you believe it's starting to happen yet?

Is GPT-5 even 1 to 3 orders of magnitudes better than GPT-4?
I think we're already seeing signs of plateauing, the next challenge will be to lower training and inference compute and memory requirements until someone stumbles on something better than the transformer architecture.

>>106234928 (OP)
straight lines are less common than many humans expect

>>106234962
fippy bippy

>>106235024
I have a book from the 80s that calls them Master Systems.

line go up

>>106235178
>AI is not a meme though. It's pretty useful, even with local llms.
AI is useful, but only if you know how to use it. The majority of people who blanket hate AI don’t want to learn what it can and can’t do. It’s similar to the internet. There’s still some idiots out there who want to ban the internet just because they refuse to learn how to use it.

I can only wonder what will happen in the future when the next generation has a better understanding of AI than they do.

>>106234928 (OP)
Ngl, projections like these have really grown on me. I see them as nothing more than hilarious memes these days.

>>106235218
no llms enable unprecedented control over the internet they are not going away. Maybe all those AI scam startups will collapse but the security state will keep the big ones alive. The question should be if openai and co. can replace or rise up to the same rank as meta and co.

>>106234928 (OP)
This isn't actually a straight line, as in a linear chart. It's a log scale

>>106234928 (OP)
>token guesser guesses better tokens
>HOLY SHIT THINKING COMPUTER
Retards

>>106235505
Literally all 99% of humans do is token prediction.

>>106235137
The Manhattan project was a moderate improvement in some areas while being an overall downgrade in the name of efficiency and deployability over conventional bombs?

>>106235438
>AI is useful, but only if you know how to use it
this applies to literally everything in existence lol

>>106235618
What? That risks and harms exist independently of the associated thing and that screeching moral busybodies are shitty fucking people that should be exiled from society?

>>106235618
I know. Just getting tired of the blanket ai hate. :/

>>106235654
NTA but It's a vocal minority. On /g/ it's mostly one faggot mod that outed themselves to me once. That one NPC retard faggot kike troon mod.
That is a fucking loser. They literally have no life but going around shitting up any AI related thread.
They also admitted that to me once.
Then they just ban anyone who they manage to tilt. Although they know who I am and know when I'm deliberately trying to get banned and so I get away with saying some pretty heinous shit on here sometimes just to piss me off.

>>106234978
But Elon Musk said Grok 4 heavy is phd level. He would not lie.

>>106235717
imagine heavily insulting someone yet still respecting their pronouns
is this what reddit looks like?

>>106235717
Take your meds then contemplate the possibility that AI being constantly shoved into everything for the past several years and overhyped has made anons and even normies get bored and tired of hearing about it. You can even see this in yearly marketing surveys which show that advertising a product as featuring AI actually makes it less likely to sell.

It doesn't help that retards are marrying the things rather than using them as tools.

>>106235370
Raw ghz is retarded it should be augmented by IPC improvements

>>106235825
If AI does deliver in the next few years, then we will see mass joblessness and sectors with job security becoming oversaturated.
If AI does not deliver, then the greatest financial bubble of all time will pop and cause economic disaster that will make 2008 look like a joke.
Then wonder why normies don’t like it.

>>106235786
You do realize “they” was used in a singular form way before the pronoun bs started right?

>>106234928 (OP)
Hes right you know, several top specialists in the area agree, just see
ai-2027.com

>>106235868
yes
this is not the case though.

>>106234928 (OP)
Leopold is a multi-millionaire. What the fuck does he care?

>>106235137
Anons, you are aware Sam Altman is a fag, right? The "A.I. Industrial Revolution" is led by LGBT.

>>106235178
Yeah, how many nukes were deployed since then? Just 2 (tests don't count) and even then only to double dab and teabag japs. Firebombing tokyo already guaranteed their surrender.

>>106235861
The best they could muster up so far is chatbots that make no money, so I'm 90% sure a big stock market correction is coming since the entire market hinges on big tech buying GPUs from nvidia.

>>106235918
I'm retarded. It was meant for >>106235614

>>106235926
I don't know why they didn't lean more into the image and video generation side, which is actually useful. I guess nothing short of AGI will suffice.

>>106235916
>Sam Altman is a fag, right? The "A.I. Industrial Revolution" is led by LGBT
Wtf I love AI now

>>106235847
the point is not to say single core performance hasn't improved, rather that we didn't anticipate hitting a clock rate ceiling so soon. we were forced to improve IPC and add moar cores to keep improving performance

straight line on a log graph with a made up axis

>>106235955
>I don't know why they didn't lean more into the image and video generation side, which is actually useful
Is it? Do you enjoy watching AI slop? Every AI generated video I've ever seen is trash. Maybe there's a market for it for shit like stock footage where quality isn't exactly expected, but it's not a huge market.

>>106235977
Yes true, but we found other ways to improve performance. I think GPT is ultimatively a dead end and a new technique is needed to make meaningful breakthroughs

>>106236024
You know what needs to be done. The ai doing cad, the taking orders at wendys, improving robot ai. Its all tedious and it needs to be done and its not fun.

>>106235955
Most people actually don’t like AI generated videos. Not because they suck but because they are AI generated. Recently people on the internet were talking about how they hate that the bunnies on the trampoline were AI generated. Each step we get to dead internet theory is step normies hate. LLMs at least give promise for one day understanding science enough to speed up scientific research, somehow. Outside of improved science people don’t want AI and most just tolerate it for the promise of new drug discoveries.

>>106236013
>>106236092
Fair points. But I'm talking more localized stuff used to touch up individual images and videos, like the stuff Adobe implemented. Obviously, spamming the Internet with AI slop is just wasteful and damaging.

>>106235233
shark detected

>>106236111
I mean, it’s already here and it’s tiny market. Besides it’s stupid trying to compete there since any model that can be run on gaming pc is bound to be either itself open sourced or be replaced by open source competitor. However LLMs in future hypothetically could take jobs of millions of white collar workers, converting large portions of their salaries into AI company revenue. Instead of collecting berries they are going straight for the mammoth.

>>106235547
Based on what? I'm a scorpio btw

>>106234928 (OP)
early life

>>106235847
IPC is a spook, clock speed is always better

>>106235861
I read valuations and other factors will make the burst smaller than that of 2008 or the .com bubble. Ai still does have quite a bit of actual use, but it is a bubble and it will pop, rest assured. The hype around it is way too huge for what might amount, once everything stabilizes, to a 5%-10% increase in productivity in software, and quite a bit more in translation, though machine translations were quite ubiquitous already for quite some time now (think instruction manuals for chinese products). I mean, it has the potential to shake things up in a lot of fields quite a bit, but there won't be AGI in the near future nor is the singularity near or anything like that.

>>106235916
>Anons, you are aware Sam Altman is a fag, right?
what's even worse is he's a jew

>>106234962
why isn't 0 at the center tho? I hate that graph

>>106235547
so much heckin this. Midwits can go through an entire degree program and parrot back what they're taught but never truly grasp the concepts. I saw plenty of this during my computer science memegree program. It was like they had an insurmountable mental block that made them unable to apply concepts in a broader sense when they weren't being handheld by the professor and are now unemployed and seething every day on twitter, and I went to a good school even.

You know what I hate about AI optimist is. They just assume sentience is probability. Its the same stupid instrumentalist idea that you don't need a real model of comprehension that you only need to model a phenomenon as is. The problem with this dumb belief is that without a foundational model of what is happening you can't make extrapolations. In other words the phenomenon is what it is nothing more, nothing less. What new can be discovered with that world view?

>>106236092
people don't like it
normies love it

>>106235137
he is a disgusting lizardperson

>>106234928 (OP)
The next "step" for LLMs is integration into computers outside of being a glorified chatbot. They've already started doing this with ClaudeCode, Gemini-cli, etc. but I imagine that they want to give it more freedom with AI agents as well.
That being said they are kind of a meme currently, they take up a ton of computational power and the result isn't particularly useful, that being said it's fun to play around with locally.

>>106235200
This. Npbp

>>106234928 (OP)
>smart high schooler
No it's an eidetic 75 iq with adhd, wandering ideation, and an impossibility of real time self reflection

>>106235469
No. Cloudflare mafia was DDoSing independent sites before LLMs became trendy.
LLMs actually forced normie sites to embrace web3 crypto and side with Kiwifarms.

>>106234978
High IQ post

>>106235444
this image weirds me out

>>106244204
only a quarter of it should

>>106234928 (OP)
It just seems the same as the global warming predictions or maybe even consider the same predictions by early AI researchers.
>In from three to eight years we will have a machine with the general intelligence of an average human being. -Marvin Minsky 1970

>>106235955
AI-generated videos have never impressed me
AI image generation has been pretty good for porn though

>>106235438
ok, list 5 useful things you do with AI
Dante must die mode: not porn-related

>>106244484
Search engine/Web Crawler

>>106244484
Post on art websites leaping as artist and stealing their audience

It won't be much longer. All they really have to do at this point is integrate their multimodal agents fully into a system where they can operate on their own codebases with distributed consensus and AGI will emerge.

It's not really any different than any other large human codebase. People laugh at AI code but imagine if a company hired a bunch of junior devs and then let them merge PRs straight to production with no testing, no QA, no meetings or any oversight, and no feedback at all.

At this point, all they're really probably waiting on is one more order of magnitude of power and compute and they'll crack it.

>>106244530
plus training on AI generated content

>>106244530
This is one of the most retarded memes on 4chan. They censor the models, of course, but it doesn't make them dumber. These models are entirely incapable of genuine intelligence, whether they're censored or not.

>>106244614
Automated coding =/= AGI. AGI is AI that can do everything human on a computer can. So do actual scientific research, be a CEO, be an engineer, be operation director. AI systems today can’t do any of that, and even then we are so behind robotics that even just cognitive AGI would have small impact because most white collar jobs still require little bit of physicality, be it just mixing Petri dishes in lab or operating 3D printers. Not to mention that it has been almost 3 years since ChatGPT got released and we are still nowhere near fixing hallucinations

>>106234928 (OP)
>can't into inductive fallacy
Leopold Aschentwinker should read a book

>OpenAI achieves ASI
>Dumbfuck fat americans ignore them
>Train it on the talmud
>World realizes they've been conquered
>Run to your nearest rabi to convert to judaism
>It's too late, drones tranquilize you
>Your consciousness is preserved
>They turn you into an immortal cock ring
>wire your brain so you can't even learn to enjoy it like a sick degenerate faggot
>Sam Altman or some other high ranking jew puts you on then fucks goyim babies for a trillion years

>>106244484
DOD missiles and kill systems. Think of satellites, autonomous IFV, AA, interceptors, wargaming etc. Automation of government functions and efficiency. I also think AI is far overblown and kind of a scam for regular people. I think its much more useful to the government and large corps exponentially.

They’d rather burn the GPUs and equipment from the data centers than sell it for cheap when they’re done with it.

>>106244803
> DOD missiles
Why does a missile need to generate an internal image of a human cock ring, or emit large texts of slightly randomized words according to word sequences that it used for input?

>>106244728
If it can code without human intervention then it will be able to prepare all the systems it will need for these other fields. Sure, it will take time, but once it's capable of that they'll give it all the resources it needs and then it will just be a matter of time.

>>106244850
Advanced missiles need to evade detection or evade enemy fire through complicated maneuvers and strategy while making all changes internally through its onboard sensors and processors. Ability to reprioritize targets or change course if a better options shows itself in real time.

>>106234928 (OP)
AGI is definitely a meme... there's most likely no such thing as "general intelligence" -everyone is adapted to do the one specific set of things they are either drawn to or forced to adapt to. There's a whole mathematical theorem called the "no free lunch theorem" that basically says AGI is impossible. That doesn't mean you can't have multiple systems that are good at different things and are coordinated...however that's not just an LLM.
People are going to look back at the idea that LLMs were the path to superintelligence and palm their faces.

>>106244931
>There's a whole mathematical theorem called the "no free lunch theorem" that basically says AGI is impossible
Or at least it says the idea that you can get to it through a single algorithm is retarded.

>>106244931
There's some debate but in my personal opinion (saying this is as having a master's in math) NFL theorem is useless for machine learning, if there's a task A then you can always have task "not A" which requires you to not complete task A correctly, the average performance is the same across all algorithms since success rate of A + success rate of not A = 100% but we do not want AGI to do all tasks, we just want it to succeed in some.

>>106234928 (OP)
Maybe I got way too high and watched pantheon but now I'm convinced that the real purpose behind AI companies is to create enough compute so that the elite can upload their brains and become immortal and live in a digital paradise while eliminating the rest of us plebs.

>>106245699
>solar flares your brains into incoherency and digital torture at a rate of 1 year per second

>>106234928 (OP)
>believe on a straight line
>its in exponential scale
>Effective compute until now is < 1

>>106235027
this but actually. think about where vibe coding was in 2022. then 2024 then now

>>106234928 (OP)
Always has been.

>>106245699
the elite don't want other elite to live forever

>>106244856
Considering that most jobs have been barely impacted by human programmers I doubt that superhuman programmer could solve these in few years. Even if you put AI research and general understanding of computer systems under programming skills, all that would do is make everyone at Google, OpenAI and xAI jobless, but it would not move us closer to stuff like fusion reactors or curing cancer or whatever other a I do goals these people have. It has been proven that today’s AI systems are almost entirely dependent on the quality of training data, and we don’t have training data good enough to do most of the tasks it’s still terrible at (everything that isn’t writing essays, coding or making porn). Like, sure you have perfect programmer that can solve every coding problem out there and zero shot every kind of software, but how will that get you training data to train plumberbot or make the AI suddenly fully understand quantum physics on the level of phds working with fusion reactors, without just memorising some facts and then wrongly applying that knowledge.

>>106245666
Then it’s not AGI. That’s the whole point of the G in AGI, it’s general, so it should be as good as human in everything, except maybe physical tasks because those are question of robotics.
>>106244931
But I think that better idea for AGI rather then being a single model it should rather be a point in time. A point in human history when all AI systems combined could outperform any human at everything, so human labour loses all of its value beyond being intermediary for physical labour because it takes few years to develop all the robots that would also do the physical tasks.

>>106245699
True except for the uploading part. They just want to be physical humans or maybe cyborgs while the rest of us die so they can do whatever they want with Earth natural resources without spending most of it on feeding and keeping 8 billion people alive. Uploading yourself on a computer is stupid because at most you would just make a chatbot that pretends to be you. You, your consciousness is still in your brain and you will not get out of there anytime soon. The most you do is create a copy and then kill yourself, and the copy will thing uploading succeeded.

>>106235024
>We
WUZ

Some folks got left out of the loop.

>>106235825
>It doesn't help that retards are marrying the things rather than using them as tools.
These are exceptionally few people. You forget incels'/autists' microscopic place in this wide world, just like a true ledditor.

>>106244931
You contradict yourself. So AGI would be a collective of multiple AIs that together make a digital super being, potentially. That's what you just said.

you fucking anti-ai faggots with your one-dimensional ideas.

>nothing could ever be smarter than human

faggots.

non-believers in ai are partially just reacting to all ai as if it's all picrel, and the rest are insecure faggots who can't imagine the idea that a computer is smarter than them.

>>106239066
You had an FX CPU in the 2010's didn't you?

Haven't they scraped the entirety of the internet by now and as a result ran out of data to train on?

>>106235861
Also pic related in enough to make the NIMBY boomers start making IEDs

>>106234928 (OP)
two more years and two more trillions , trust me bro

>>106248592
good morning saar

>>106244652
except we have no way of knowing that with the censorship there

>>106244228
You are right, top left is disgusting.

>>106248592
you'll be stuck in mumbai until you die

>>106244530
>>106234978
STOP BEING ANTISEMITIC

>>106234928 (OP)
AGI is impossible by definition.
The only way to create life is by consummating your marriage.

>>106235547
>>106240356
it sounds more like this is just the retarded way people get taught to learn things in school, but is not all that humans do

>>106250178
>>106251348

i'm probably whiter than both of you pussies, and i i have beautiful blue eyes, and i own more land and more gold than you ever will.

>cant trust it to be right every time, guess AI is a meme

who said anything about "trusting" AI? you faggots are incapable of critical thinking for anything you read?

you are everything that you hate about AI.

only people who have a fully developed consciousness should be allowed to use AI. (no vaxxies, no browns, no PMs, etc)

>>106235024
explain this graph to me. Is it about the abilities of a AI with a sharp increase at the start and then not improving over many steps?

>>106251366
you don't know that

AI is neat but mostly marketing. They always shill their slop as the greatest thing ever. Even funnier when you consider how many people trust it to do critical tasks when it doesn't know basic things.

>>106253154
but isn't it worth it? the grift I mean. for everyone, collectively? the more people pay the slop, the better it gets, the more use we get out of it.
you could argue whipping paypigs into consooming it is useful for everybody

I don't think I screencapped it but I remember when the big wave of hype/fearmongering about DeepSeek hit (what happened to that btw? I haven't heard about for a long time), there was a Twitter thread from someone explaining one of the big reasons Western AI developers were ostensibly losing to China:

>long before LLM-based AI actually exists, autistic doomsday cultists predict AGI will attain godlike power and bring about the apocalypse
>once LLMs make "AI" a real practical thing, doomsday cultists go into overdrive warning Big AI that they're creating an evil god beyond their control
>Big AI people ignore the "evil" and "beyond their control" part and become obsessed with the "god" part
>AI research gets increasingly diverted from practical applications and toward the extremely hypothetical AGI/superintelligence because the people leading development think they will be rewarded for being the first to invent Computer God
>meanwhile in China believing such things would get your organs harvested so chinks happily stay focused on more realistic applications

>>106249216
>98.7% of internet data is retard slop
>AI is trained on retard slop
>AI Is retarded
Wow its just like me

>>106235547
Prove that the brain generates tokens then retard

>>106235218
People have been predicting the top of the s-curve every month for years now. Every time there has been another new technique that allows the scaling to continue within about 4 months of the prediction.

>>106235227
Those new micromodels that outperform o3 with 5M parameters are probably a good bet. RCMs, were they called?

>>106254348
When is it going cause mass layoffs of programmers? I want the "just learn to code" niggers to feel it, its only fair.

>>106234928 (OP)
>is AI a meme ?
>>>/g/ldg
see for yourself and laugh

>>106254371
Already has. Entry level job openings have plummeted. Most companies are refusing to hire new talent because even the slop ai is better than new programmers right out of bootcamp or college. Salaries for for non-silicon valley programmers is dropping as the pool of applicants grows and AI takes more and more duties. Either AI gets good enough to take out mid-level or the system breaks down from competency crisis.

>>106240356
>heckin
is this a new word filter I haven't hit yet or are you just an enormous faggot?

>>106253337
>what happened to that btw? I haven't heard about for a long time
It's just one more evolutionary step. Everyone in the space who's paying attention is tracking the Kimi/Qwen/GLM/etc flavor of the month. Commercial AI is pretending it doesn't exist and specifically won't talk to anyone who they know will ask them about it. I wouldn't put my long term money on any of them since the open source models are keeping pace, and while commercial AI is busy trying to find ways to bleed their perceived competition, open source can quietly focus on advancing.

>>106254872
Taking the legs out of new programmers while needing high level programmers seems rather short sighted

>just extrapolate exponentially bro
There's a funny psychological phenomenon whereby when the rate of progress in a field has already peaked, people experience maximum enthusiasm and hope for future progress. It's like being in 1969 and expecting that by 1990 we'd take vacations on the Moon and travel from New York to Tokyo in 45 minutes, when in reality the plateau was already reached and people in 1990 were still travelling in Boeing 747s.

We're there right now with AI, this is it, the top AI Clippy of 2040 might not be dramatically better than ChatGPT 4.1, it could even be made much worse out of greed, just like air travel also became worse starting in the 1970s.

>>106245699
>elites are going to kill themselves and be replaced by a shitty LLM trained on some of their thoughts and experiences.

Wish we were this lucky.

>>106235233
no matter how many times I unfocus my eyes I cant see the sailboat

>>106235916
then whyd he rape his little sister when she was a child?

>>106235918
neither firebombing them nor nuking them caused them to surrender. it was the russians finishing up germany and getting ready to be deployed to the pacific front that made them give up.

>>106234928 (OP)
Yes. I wanna clap his cheecks tho

>>106236013
https://youtu.be/Rcp1SH7Wq3I

>>106255653
That was his way of figuring out his sexuality. His "orientation" used to be in the DSM for a good reason.

Imagine diogenes saying, behold a man

>>106257099
>his ribcage isnt on his eye
sucks to suck

>>106257099
The labels are off, obviously, but as far as the diagram goes: it's pretty good aside from the femoral vessals passing anterior to the pelvis and the sprant and the ludger being on the wrong side.

>>106259077
Yeah, the human body is complex, let's try something simpler
>Okay, drawing maps of the country is too difficult what about just one state...

>>106259077
Yeah man, where is your sprant and ludger? Oh, on the wrong side? God bless, you never had a chance