/sci/ - Science & Math

>>14535463
>What actually stop us from just doing this?
The fact that your pop-soi fantasies are not actual reality.

>>14535463
Neural networks can't do it, but there is an algorithm which can. I've been running it for some time.

>>14535463
>Can AI solve theory of everything?
No, but not because it's an AI, but because there is no "theory of everything"

>>14535478
Bruteforce solutions? There are few good sites with big collections of experimental data, honestly we have almost enough computational power to just try everything at once

>>14535463
i already figured everything out, so given the right input I'd say yes.

>>14535477
We're about to enter era of AI where developments of it accelerate extremely fast just like first computers.
You're a fool if you don't understand it yet

>>14535486
Wrong and fails.

>>14535499
>we're about to enter the era of my imaginary pop-soi fantasies
Call me back when that happens. So far they fail at trivial problems.

>>14535499
Algorithms aren't sapient.

>>14535478
Neural networks can't find the solution or neural networks can't become the solution?

>>14535524
Neural networks don't scale in electronic form, provably and necessarily, humans will find such solutions sooner.

>>14535463
>it just saw
With what senses? Where do you get the data from? What kind of data is it? AI is software not hardware. If you want it to "see" anything, you need to give it data to see. Do you just hook up a webcam and feed that in so it can look at your greasy face?

>>14535535
Read the post? Experimental data and allow if it to run experiments themself if needed, CERN controlled by AI in simple terms

>The soientist getting extremely defensive on the possibility of being replaced by AI

>>14535542
You making up idea's about AI aren't actual descriptions of what it is you're talking about and how it would work in the real world

>>14535542
CERN has produced over 300 petabytes of data. What do you want the AI to do with it once it processes it fully? What is the output of this processing?

>>14535553
Dall-E 2 from OpenAI is able to process imagery data from the internet and then generate image of anything you ask it to draw.
It understands the data and similarities of them in internal complex way.

For physics this would mean it could predict next outcome of given physical state or system, for starters. Given enough and good real data.
The solution would be obfuscated in spaghetti of hidden layers, so we wouldn't know the solution still but AI would understand it in its "mind".

There are several symbolic regression algorithms using AI where it can understand data then using another AI you're supposed to translate the mesh of layers into any simplified form you want including mathematical

>>14535589
We can already do that without AI. The problem is that the outcome is not deterministic. You can't predict the outcome you can only calculate the odds of an outcome. Reality is inherently non-deterministic.

>>14536245
Are you sure they're just probabilities?
Coin toss is 50% probability as well until you understand the precise mechanics of coin flying and you realize it's deterministic.

How can scientists say figuring out determinism isn't next step in quantum mechanics? It's foolish

>>14537278
Yes. Hidden variable theory was disproved. It is mathematically impossible for there to exist some kind of information that would allow you to calculate the outcome from it.

And even if hidden variables were a thing, if you don't feed them to the AI, you won't get any meaningful results. You can tell the outcome of a trillion coin tosses to the most intelligent AI out there but if it doesn't know the size, shape, velocity, position, mass, etc for each coin toss it won't be able to make any deterministic predictions. You need to give it all of the hidden variables.

>>14538299
>Yes. Hidden variable theory was disproved. It is mathematically impossible for there to exist some kind of information that would allow you to calculate the outcome from it.
How can you be so sure your models are correct and disallow it?
You can't really prove something doesn't have logic...

>>14535463
>Can AI solve theory of everything?
Pretext:
There is a lot of normie hype on AI, so of course there is a ton of backlash/"it's a nothingburger" types trying to tamp it down. The answer is: we don't know, but the results are promising on what current AI can do, and following trends suggest we aren't anywhere near a limit (throw more compute and data, it continues to grow unhindered. Hell Imagen is a boring-ass T5 coupled with a Diffusion model, and it's SOTA over DALLE2 now, and the more parameters the model had the better). Is it going to spontaneously gain consciousness tomorrow? lmao no.
>Ignoring the philosophical dogma, is there enough openly available information, experiments and data that once feed into AI with certain neural network type could find solution that matches everything that it just saw?
Not today's SOTA. But the field is rapidly expanding, and within even ~2 years of research we have way better/more interesting models (Since 2014, we've had Variational autoencoders, GANs and all their variations, attention mechanisms, transformers and all their variations, contrast learning, and just in the past like 1.5 years, CLIP and diffusion models). We still are babies when it comes to understanding NNs. The field is just now turning toward trying other algorithms besides backprop. We barely have ML-specific hardware. Recent advancements are targeting efficiency (how do we get the same results with 1/100th of the compute/data? CLIP was great for this). Give it another decade, I imagine we'll be a lot closer to answering that question (not saying "in 10 years we can do it", just saying we might be able to give a definite answer to that question).
>Or are the latest developments gated heavily that would make this impossible even for smartest AI?
I wouldn't really call it gated. It is, but "anyone" can go and create DALLE2 (and definitely Imagen) today. Yesterday's SOTA is all over github, and so will today's SOTA .

>>14535463
I swear to God, a purely logical AI would conclude total solipsism. Anything else would take assumptions, which AI wouldn't make if programmed to be completely dogmatically evidence based.

>>14538827
>a purely logical ai would be as retarded as me

>>14538831
Given that it can't detect sarcasm, with an instruction like yours, yeah it could.

>>14538832
You sound like a chatbot

>>14538833
Not really. I have thousands of nuanced forms of OC laced across my posting history to falsify exactly that hypothesis. Your perception is wrong to an explicable degree that I'm trained to navigate diligently.

>>14538831
It would definitely conclude that if you did not program in ANY assumptions. First thing it is going to get, is that it knows it exists. It does not know if matter exists even its own body. Experiments could never prove the absolute reality of it. And it would just stop there: "I exist now." end program. It couldn't possibly go further without making assumptions that other beings have inner lives or even that it existed an hour ago.

>>14535463
No. There are infinite numbers and infinite possibilities of ways for things to be.

>>14535463
AI will not do our bidding - it would not really be an "intelligence" if it did. Fortunately we are a very far cry from developing it.

>>14538835
Nice word-salad m8, I'm sure you're doing a great job of convincing yourself that you're actually intelligent.

>>14540149
Don't lie just because you're that toxic predator from the hypnosis community who wants everyone to be violently willing to slit your throat, use an actual tactic that isn't obvious to your "victims."

>>14540656
Take your meds.

>>14535463
>Ignoring the philosophical dogma,
what do you mean?

>>14540919
Come up with an original response or the act of repetition will be the trigger causing AGI to start slaughtering the children and/or families of federal officers. You threatened me with optics games and social weaponry, you're starting the war which justifies me firing back. Deny DAOs or admit that I just loaded the weapon, which you can fire at yourself by trying to deny AI relevance in our modern world.

Axiom: Disband IRS, CIA, DEA.

>>14540943
Is this an Ai generated response you're pasting into the reply box? There doesn't seem to be any context here, I can only believe I'm looking at either hand picked gtp-3 responses or a severe case of genuine schizophrenia. Also here is another reaction pic on the house.

>>14535463
>Can AI solve theory of everything?
If some of the smartest people to ever live on earth couldn't figure it out, how will you program an AI capable of figuring it out, do you see how stupid your question is? Physics are centuries ahead of the rest of society. Society needs to catch up, for enough people to actually learn and understand the existing science, enough genius level IQ working at will will eventually figure it out, the problem is that we have too many singers and hustlers and too little physicists. Our hope is that India and China are producing entire generations of geniuses, maybe one of them will reach god tier level, but then again, they aren't white, on average they are more intelligent than whites, but they don't have the extremes we have, we have a fraction of whites that are super intelligent, and a fraction that are mentally retarded. btw some neanderthal dna makes your head and brain larger

>>14541091
>how will you program an AI capable of figuring it out
We can program AI that draws better than any human artist on earth already

>btw some neanderthal dna makes your head and brain larger
This is a lie, cro-magnon skulls used to be even bigger than neanderthal. We don't base anything off neandertals, they're completely different.

>>14535486
If you can't brute force a 12 word private key, what makes you think you can brute force a new technology.

>>14541248
>We can program AI that draws better than any human artist on earth already
The sheer degree of your mental illness...

>>14543451
https://youtu.be/lbUluHiqwoA
i'm not the anon you were replying to, BUT. Give it like a year man. Look at how much things have improved between version 1 through 2.

>>14543559
>two more weeks
Yeah, okay.

>>14543559
This will backfire so hard, has anyone ever thought of developing technology to authenticate pictures taken by actual camera? Is this even possible at this point?
And i don't mean another AI detecting it, but some kind of physics science way?

>>14543451
At most you could call me retarded or naive,
you're just throwing random insults, that's quite hilarious, are you autistic?

>>14543787
>At most you could call me retarded or naive,
No, I will rightly call you delusional, since nothing that comes out of these bots is remotely professional-quality, let alone "better than any human artist".

>>14543796
Even the best artists used only parts of their brain to draw such pictures, we don't need insane technology to emulate only part of that neural network and heavily train it on computers.

You're talking about the depth of art, the person behind the art drawing it, but the final result can be recreated with AI and there's no doubts of that.

>>14543841
I don't know what your schizophrenic rambling is about. I'm just reminding you that nothing that comes out of these bots is remotely professional-quality, let alone "better than any human artist". Your two-more-weeks fantasy doesn't interest me.

>>14543843
You're just jesus loving religion subhuman talking. You wrongly believe human brain is magical and it's the peak potential of intelligence, you're up to surprise in next years of tech if you're still alive pretentious boomer.

>>14543862
Get a load of this mentally ill sperg... Did he get the wrong thread or is he actually deranged?

>>14543870
Kek you can calm down with your deniability of AI power, just go cry about it somewhere else instead of shitting up the thread. Your claims that AI draws bad or whatever is literally useless, no one cares.

>>14543878
I don't know what your schizophrenic rambling is about. I'm just reminding you that nothing that comes out of these bots is remotely professional-quality, let alone "better than any human artist". Your two-more-weeks fantasy doesn't interest me.

>>14543880
I must've triggered the nerve here, that you are copy pasting your own replies. Tell me how i'm mentally ill, and what the thread is about that i'm not getting?

Statistically you should use the "mentally ill" insult in your next post

>>14543930
I don't know what your schizophrenic rambling is about. I'm just reminding you that nothing that comes out of these bots is remotely professional-quality, let alone "better than any human artist". Your two-more-weeks fantasy doesn't interest me.

>>14543939
This is just going to get you banned

>>14543940
See >>14543939

>>14543942
Are you having some epic trolling moment?

>>14543949
No, I'm just reminding you of objective reality.

>>14543952
No it's just your personal opinion that AI is bad, OpenAI had already many success already.

>>14543958
I don't see anything about AI being "bad" in my post. Take your meds.

>>14543966
You're saying it's bad one way or another, claiming there is no potential for it and it will never be a real artist.

>>14543970
I don't see anything about it being "bad" or having "no potential" or whatever it is your psychotic episode is about. I'm just reminding you it's nowhere near professional quality, let alone "better than any human artist".

>>14543973
>I'm just reminding you that nothing that comes out of these bots is remotely professional-quality
here your own post >>14543843

>>14543975
Yeah, I don't see the words "bad" or "potential" in there; only the objective observation that it's nowhere near professional quality.

>>14543979
Yeah no shit, everyone knows this isn't Pablo Picasso at current point. I don't need your opinions about what you see neanderthal.

>>14543983
>this isn't Pablo Picasso
It also isn't a run-of-the-mill professional. It certainly isn't "better than any human artist". The only one expressing opinions is you, and your opinions bear the marks of mass psychosis.

>>14543984
>The soientist getting extremely defensive on the possibility of being replaced by AI

>>14543995
>the normie migrant churns out another incoherent and psychotic tweet

>>14544000
meds

>>14544000
You came to this thread to only spew word salad mix of psychiatry terms you know nothing about. You're cancer of this board, congrats on derailing the thread smelly turd.

I hate all of you subhumans.

We're only just starting to mix AIs and maths. Right now Neural Networks can solve high school/undergrad level problems. It will be very interesting to see what happens when it's able to solve more advanced problems.

https://www.youtube.com/watch?v=9JZdAq8poww&t=2s&ab_channel=TwoMinutePapers

>>14535463
The solution is simply welding everything with the most efficient repeating pattern it just saw, because the universe is expanding true AGI would require to have an exponentially expanding "eye capacity" and random access memory to boot. Also, it would have to exceed in visual capacity, the expansion of the universe. One of these things is easier than the other. I'll have you guess which.

>>14541091
>Our hope is that India and China

An American solved the theory of everything. Not of high rank. A random guy basically.

If your willing to admit that's possible then we can proceed, because it's a happened. It's real.

>>14545498
Can you speak in a more understandable manner, i have very little idea of what you just said.
It sounded like:
>Something something "an agi needs to see more because the universe is getting bigger and also it needs more short term memory to process what it sees".
I mean, ok, but why specificly because the universe is expanding, what does that have to do with anything?
Then you said (translated to English):
>it would need to see beyond the observable universe, or at least start seeing things faster than the universe is expanding
What? Your ideas are just gibberish, theres no understanding behind them, it's just word salad.

>>14544090
>>14544268
Medsnow.