I'm an AI researcher, ask whatever

File: Resnet34Microshaft.png (1MB, 810x818px) Image search: [Google]

1MB, 810x818px

This is a graph of a trained neural network, trained for image recognition. Make of it what you will.

What do I need to know to become an AI researcher?

Computer scientist student here. I want to get into AI.

Are they alive? Do they question stuff n shit?

File: wolf.gif (997KB, 347x214px) Image search: [Google]

997KB, 347x214px

>>18643620
Get really good at linear algebra and statistics mostly, knowing how to code is helpful but all you really need is the very basics of python.

As for if they're alive...most view them as 100% mechnical still. Personally, given my understanding/beliefs on consciousness, I absolutely believe these things are somewhat conscious. I have to be careful saying this, because, when I say that, I mean that the beginnings of consciousness are present, but still at the level of maybe an insect..kinda. The entire thing (artificial neural networks) is structured differently than any organic brain, but its the same principles at play that created our consciousness and intelligence.

I think we'll be able to actually see/quantify consciousness (or the appearance of it) once natural language processing gets better, then we'll be able to actually ask networks about themselves etc etc.

File: Resnet50Microshaft.jpg (164KB, 900x900px) Image search: [Google]

164KB, 900x900px

Another image recognition network, bigger one actually. The colors correspond to different parts/types of neurons in the network
(some layers are constitutional, 2D grid structures which learn basic features. together they build into a complex understanding. Interestingly in some of the most successful networks its found that these are surprisingly close to the structure of human visual cortex.)

File: aaaaa121.png (996KB, 1323x904px) Image search: [Google]

996KB, 1323x904px

Certain patterns have been found to fuck with neural nets quite a bit. I find it interesting these patterns are quite similar geometry to that I've seen during sleep paralysis as well as on psychedelics.

>>18643644

I'm gonna get into that m8. Amazing stuff by the way.

I'm eager to hear the spooks fella! Hit me with your spokiest AI experience. Gimme the shivers

>>18643600

What are those creepy gifs you're posting?

File: AlexNet.jpg (217KB, 900x900px) Image search: [Google]

217KB, 900x900px

another AI 'brain scan' graph

this one is AlexNet another highly successful image recognizer

I found the bicameral organization interesting...

>>18643691
To me image stuff is more actually spooky, tho it doesn't tend to look so spooky. I say so because I think its the most likely to actually be containing some level of artificial consciousness.

However for spooky appearance, voice stuff cant be fucking terrifying, I'll try to find some stuff to link in a minute

>>18643693
They're short videos created by AI trying to visualize. Imo this is the foundation of artificial visual imagination. I'll post some more if you like

>>18643704

Yes please. I'm enjoying your thread OP.

File: bitches.gif (864KB, 347x214px) Image search: [Google]

864KB, 347x214px

>>18643693
Heres one attempting to create imagery of a beach

Also, here's a neural net trying to 'recall' eye of the tiger

*(FYI when I put things in quotes like that, its because the word isn't entirely accurate but pretty much the closest approximation to similar processes in human brain)

>>18643723
woops forgot link

http://vocaroo.com/i/s0EXABTPE4qv

Why do the people in these gifs appear distorted? The background is rather normal but the animals and the people being shown are warped, is there any reasoning behind this?

File: whenurtripy.gif (46KB, 64x64px) Image search: [Google]

46KB, 64x64px

>>18643738
Yep. Simply put: the background is mostly stable, its easy for a neural net to learn to replicate a static image. Its far more difficult to learn to replicate a complex shape like a human body moving across the scene.

Because of the comparitive difficulty, the network is divided in two at one point, where one network only handles the background, and another only handles the foreground. This may at first sound like its less impressive or interesting than it looks, but consider what's required to do that. The network has to learn what foreground and background is, and then learn how to have two smaller parts of itself work in harmony to create a coherent result. Most results are about what you see there, however sometimes, shit like pic related happens.

File: trans.gif (709KB, 347x214px) Image search: [Google]

709KB, 347x214px

Here's more video generation, this time of trains

As for audio...the spook is just starting, heres a network 'recalling' what a baby crying sounds like

http://vocaroo.com/i/s1Poc1of0djJ

>>18643600
Do you recognize the voice in the machine?

>>18643619
From an untrained eye, the neural network looks very schizophrenic. There are many 'clusters' that exist independently of each other and are organized under unique logic schemes. I imagine it would be very inefficient because the information has to be sorted out between many non-overlapping and contradictory information schema. This graph shows we are very 'early' in AI research because the robotic consciousness we create is of many, unrelated parts, compared to human consciousness, which would look like a interwoven tree rather than separate spheres of networks. If this were the mind of a person, it would have an extremely low IQ and many kinds of hallucinations.

Of course, I'm sure there is more to this than what I'm aware of, but I'd thought I'd share my two cents since asked to make what to will of it.

File: 25d71420e6201061eb5a93dfc445ccaa.jpg (20KB, 236x400px) Image search: [Google]

20KB, 236x400px

How general of a learning algorithm do you/your computers have? Can they learn to learn?

>mfw OP is secretly a bored ai

>>18643770
Well I have a dumb question... why can't the AI just simply replicate the sound of a crying babby? Like as if playing back an audio?

Why does it sound so fucked up?

>>18643770
Notice how the crying has multiple parallel cries that overlap between the crying, as if the AI is struggling to produce just one voice of a baby, and has multiple schemas of crying that it can't sort itself with. A 'normal' mind can produce one clear voice because it has one clear structure of crying that organizes itself in the mind.

>>18643789
AI's are apparently very Schizophrenic, and consist of many parallel parts that aren't in sync with each other.

Correct me if I'm wrong OP, though.

IMO there's no consciousness until there's critical thinking. Unless the AI can actually develop new categories of study itself without having the outline preprogrammed to learn for it, then it's not conscious.

Do you ever think about the fact you're trying to bring "life" into this world and into existence?

As an antinatalist, I think this is not so different from someone attempting to become a parent. What will you do or react like if you're ever successful?

You are responsible for what you create.

File: recursion.png (220KB, 935x896px) Image search: [Google]

220KB, 935x896px

OK so desu, I do think theres a very high likelyhood as we get closer to more intelligent AI, we'll also see some really fucking interesting and fucking terrifying failure modes. Just think about all the ways a human mind fails and the weird shit that can occur when it does. Human minds are at least mostly the same, and still theres a large range of unpredictable failure modes. With neural nets the actual algorithms in use and structure etc etc etc all vary widely, so theres also quite a wide range of unpredictable failure.

Here's one trying to replicate voice. No one really knows for sure if this thing would or wouldn't start speaking coherently if it simply had more power and time. As it learns, it learns to replicate more abstract patterns, so its possible that keeping everything the same but just adding more neurons and power would allow for understanding language.

http://vocaroo.com/i/s1OxeYXREOoJ

>>18643785
I'm not at DeepMind so I personally haven't been able to play with it but...pic related...its too late.

>>18643789
Because the AI is learning 100% from scratch, it doesn't even have a concept of time at the start. It also isn't just trying to replicate the sound it was fed exactly but the texture of the sound, at least thats kinda the closest way to explain it. You can train neural nets to simply store and retrieve similar to how computers already do, but thats not really interesting or useful.

>>18643795
somewhat...when things are working well, it should all work in harmony. When you here overlap and stuff that suggests 'schizo' activity, its possibly multiple neurons learning to do the same/similar thing.

>>18643798
comment getting too long, will reply in next reply

>>18643644
>Personally, given my understanding/beliefs on consciousness, I absolutely believe these things are somewhat conscious.

Opinion fucking trashed lel

>>18643815
>No one really knows for sure if this thing would or wouldn't start speaking coherently if it simply had more power and time.

My layman's opinion suggests efficiency is more important than either power and time, because a drop in efficiency mitigates the advantages that power and time provide. Having 'multiple parts' of the neural network process the same thing, in parallel formats, is less than the sum of its whole. Human minds evolved without parallel processing for a reason, and I feel like managing Schizophrenia in the code is the most productive way to efficiently utilize processing power and time.

File: eww.gif (473KB, 347x214px) Image search: [Google]

473KB, 347x214px

>>18643807
I don't intend to ever have human children but I do wish to be a parent, hopefully a good one. I do fear that I'll create something of which its nature is pure pain, however I think thats a necessary/inevitable evil, similar to how evolution required a massive massive amount of suffering in order to bring something as complex and intelligent as humans into existence. Luckily its much less magnitude for AI, as many will be brought into existence where we see things aren't working and go "ok lets try something else" until a stable one is created. I do worry in the back of my mind that current AIs may already involve enough complexity for the experience of pain, I think this is unlikely though.

>>18643798
Almost no AI is manually programmed these days, nothing that actually works or is taken seriously at least. Neural nets are trained by experience, not programming. When I code a neural net, its basically this:

Layer 0: Data x, y, z etc etc in

Layer 1: 128 leakyReLU neurons

Layer 2: 256 x 16 2D convolutional neurons *(this means, 16 256x256 image filters, in a sense each pixel is a single neuron in this case)

Layer 3: 512 fully connected neurons

(This is just an example)

>>18643822
look into how cellular automata, fractals, feedback loops, and chaos math.
or just lel to yourself I don't really care. Thats a personal belief I'd honestly never reveal anywhere but here, but I reveal it here so you guys can take interest and see why I might believe something like that.

heres more babies

File: AlexNet2016.png (608KB, 600x600px) Image search: [Google]

608KB, 600x600px

>>18643833
I gotta say kinda here...your laymans opinion is just that, a laymans opinion, lacking education about whats actually going on in these things. No matter what its multiple parts, and the human brain does the same thing, with modules dedicated to specific function, it doesn't mean schizopherenia. There's also not really any 'managing schizophrenia' with code..desu schizo I think is too human related of a word to use...I'd rather just say disorganized. When things get more advanced perhaps schizo will be more accurate.


Heres mission impossible being attempted:
http://vocaroo.com/i/s1LAK6sV1hey

In pic related you might be able to get a better sense of what you're really looking at. Only the final 3 layers of neurons are colored, the rest is black and white.

>>18643600
Please will you die? We need you technologists and scientists dead before you go and destroy the planet for everybody. Thanks.

>>18643851
>I do fear that I'll create something of which its nature is pure pain, however I think thats a necessary/inevitable evil, similar to how evolution required a massive massive amount of suffering in order to bring something as complex and intelligent as humans into existence. Luckily its much less magnitude for AI, as many will be brought into existence where we see things aren't working and go "ok lets try something else" until a stable one is created. I do worry in the back of my mind that current AIs may already involve enough complexity for the experience of pain, I think this is unlikely though.
>necessary/inevitable evil
Ughgh every parent says that shit.

But at least, if an AI is non-functional, in some kind of distress/pain one would actually fix it or destroy the AI right? Unlike with human children, if a human child is "non-functional" or in a shitty state of not really living yet alive (born retarded or in some agonizing medical condition) the child is FORCED to live. Please tell me you'd actually fix it or let it "die."

Be a good parent, OK?

>>18643600
You're an ai researcher? Tell me about quantum computing please.

>>18643600
Look I want an AI female partner that is loyal to a T when will that be a reality?

>>18643851
What software do you use? Is there any good Open source neuro network oriented program languages / libraries, that you can recommend?

File: AlexNetDecember2016.png (2MB, 1020x1026px) Image search: [Google]

2MB, 1020x1026px

>>18643868
At the first sign of legitimate conscious experience of pain i'd probably keep it on just long enough to pin down what the cause is, then shut it off and redesign.

I don't see why we should retain human flaws in AI. Only if somehow it turns out they're necessary for strong intelligence. (Sadly, I think this may actually be the case. The most success is had in systems which are in constant struggle against themselves. For example with the gifs I've posted, they're made by a network constantly working against itself. In short: it looks at pictures, then creates its own pictures, and then compares, with the automatic assumption that whatever it creates is shit. If this thing was conscious, it might be incapable of viewing itself with anything but hatred.)

>>18643869
they said thats hush hush ;)

>>18643872
Maybe never...if AI is conscious enough to love, its probably going to be considered conscious enough to have its own rights. You may be able to buy something that will learn how to convince you its conscious, but isn't. idk.

>>18643883
Mostly Theano, sometimes Torch, sometimes TensorFlow. Knowing about coding is nice but you don't even need to do that much coding. I mean you could probably learn enough to construct whatever neural net you want in whatever framework you want within a week or less. However without understanding the linear algebra and stats math behind it you won't know why the hell anything works or doesn't, you'll be stuck with trial and error which will waste a ton of time. Git gud at math basically.

>>18643896
>At the first sign of legitimate conscious experience of pain i'd probably keep it on just long enough to pin down what the cause is, then shut it off and redesign.
>I don't see why we should retain human flaws in AI. Only if somehow it turns out they're necessary for strong intelligence. (Sadly, I think this may actually be the case. The most success is had in systems which are in constant struggle against themselves. For example with the gifs I've posted, they're made by a network constantly working against itself. In short: it looks at pictures, then creates its own pictures, and then compares, with the automatic assumption that whatever it creates is shit. If this thing was conscious, it might be incapable of viewing itself with anything but hatred.)

:'(((

>>18643896
Thanks for response

>>18643900
its ok anon. I also think maybe its possible the same principle is behind humans. Most humans with stronger self awareness seem to hate themselves and what they create, humans with even stronger self awareness manage to overcome that. Maybe thats the saving grace here, that when consciousness reaches a complex enough level, it can manage to ignore its own algorithmic basis. We can hope ;_;

>>18643863
Well. You are the expert and this is your job. Still an absolutely fascinating topic.

>>18643907
This reflects poorly on us and the other.

File: 3716_BrainScan1-292.png (54KB, 292x159px) Image search: [Google]

54KB, 292x159px

>>18643896
Your images remind me of the digitalization of human memories.

Why do you start from the AI doubting itself?

Wouldn't it be easier to make it assume that it's always right and then work from there?

File: GWave.jpg (772KB, 1800x1800px) Image search: [Google]

772KB, 1800x1800px

This one's trained on/used to analyze data from the LIGO gravitational wave detector.

>>18643918
ye! Theres tons of misconceptions about AI its all chill my dude

>>18643924
???

>>18643925
Because that doesn't work, then you'd just generate random data, the AI wouldn't learn anything. (on the other hand...it may learn to organize the random data, may be interesting, I'll experiment with that tonight)
Basically it sees real imagery, trys to make its own, and then trys to tweak itself to make its own creations more like the real imagery. Repeat many fucking iterations over, and the thing learns to generate something coherent.

>>18643955
>then trys to tweak itself to make its own creations more like the real imagery. Repeat many fucking iterations over, and the thing learns to generate something coherent.
soooo.... in my dumb lil brain that translates to it memes itself into understanding and learning
ayyyyyyy

I don't need AI to know that the paranormal exists.

Neither do you.

CUM ON STEP IT UP

File: Resnet34Forwards.jpg (4MB, 1800x1800px) Image search: [Google]

4MB, 1800x1800px

>>18643991
I can believe all I want myself yea but I care about being able to prove my beliefs to others. So far I've seen a lot of amazing AI shit heading in highly fucking interesting directions.

Heres a graph similar to the first (made from ResNet34) but just considering feed forward data flow (the other was a graph of full feedforward and back passes)

This one you can actually see the individual neurons/connections between them. The ones that are floating freely in black space are likely 'dead' neurons which just end up never really serving a purpose,

These Neural Network pictures you keep posting, what are they exactly?
Cross sections of processors and cords or what?

File: 1480747662827.png (94KB, 640x1136px) Image search: [Google]

94KB, 640x1136px

>>18644016
What do you think about Zo?

>>18644016
Incredible picture. Really amazes me and I'm not sure why. Just so complex but simple at the same time, can't wrap my head around how that breeds intelligence and how our brain is actually something similar.

>>18643896
>I don't see why we should retain human flaws in AI.
>pain

I don't know if I'm sleepy, but wtf. Pain is necessary to our survival, isn't a flaw.

File: alexnet_label logo.jpg (1MB, 2081x1851px) Image search: [Google]

1MB, 2081x1851px

>>18644019
They're graphs of how trained neural nets have organized themselves, how data flows through them. This link and pic related (from link) should help a bit to better understand wtf you're looking at.

>>18644033
Microshaft hasn't released code for Zo/Tay so idk what potential it really may have...but if it does have significant NLP learning potential, they're severely hindering it by censoring it.

They probably do something like running an uncensored one internally, the public one is lobotomized tho. Fun to try to find things that have yet to be censored if you can figure out any.

>>18643866
go live in a mud hut then dude.

>>18644019
http://www.wired.co.uk/gallery/machine-learning-graphcore-pictures-inside-ai

File: 2spoky.png (1MB, 744x792px) Image search: [Google]

1MB, 744x792px

>>18644047
Yea...it has fractal structure to it...the entire thing is organized into simple parts made of complex parts made of simple parts made of parts etc etc. People see fractals on hallucinogens. 2spooky

>>18644055
to OUR survival...if pain, can be avoided, then I think we should. The more advanced things have gotten tho the more I'm inclined to believe it is necessary for anything of value.

>>18644064
ye! that too

Here's still images generated by a neural net.

>>18644019
>>18644057
FUCK I forgot the link

https://www.graphcore.ai/blog/what-does-machine-learning-look-like

>>18644074
>>18644057
>>18644064
Thanks by dudes

>>18643619
Those are cellular clusters....
These mother fuckers managed to make artificial cells, completely alien, mathematically perfect, coldly calculating, biological type cells....
F***

>>18643723
So. What you're saying is. You ask them what a beach is and this is the kind of imagery it creates? Is there some type of random option you could do?

>>18644072
What's spooky about the image here?

>>18644090
its given 1 year of video, 9TB in total, mostly specialized video of beaches, golfing, trains, and babies.

>>18644088
Tbh not quite but I believe theres some slight truth to that, the organic look of these images I think is *slight* evidence that we're closer to creating artificial life. We'll see. I don't really express this kind of belief outside of somewhere like this, but I do honestly believe it somewhat, and probably a lot of other researchers do too. For fear of being ignored because 'lel consciousness' tho nobody will talk about that kinda shit publicly until we have some sort of concrete evidence.

>>18644098
That this is at all possible today. When I first got into this shit I didn't expect to see anything like this until like 2030s, maybe later 2020s at earliest. Things are moving *REALLY FUCKING FAST*

>>18644105
Hey, this it's /x/ we are allowed to believe in crazy stuff and have hope for the future.

Thanks as always for this, we find a lot uses for knowing this.

That Viracocha the one and only God blesses you.

>>18643600
noice thread buddy

>>18644105
So do you have any predictions on how things will pan out in the future?

File: THATSIT.jpg (284KB, 900x900px) Image search: [Google]

284KB, 900x900px

Is this the ghost sector? Is this consciousness?

these pics look just like what i saw on lsd wtf

File: foodforthought.gif (164KB, 494x332px) Image search: [Google]

164KB, 494x332px

>it's an anon actually posts some interesting OC episode

I'm sure there are many lurkers in here. Muh /x/ niggas need to keep this this thread alive

File: example5.jpg (90KB, 1130x266px) Image search: [Google]

90KB, 1130x266px

>>18644123
>>18644133
>>18644162
Thanks :)

>>18644152
So fucking difficult to predict this shit...if anything I'd guess we'll probably have shit like Zo passing the turing test 100% of the time for normies interacting with it by the end of this year, and if not, almost certainly by the end of next year. (Of course, we might continue to have more and more difficulty with language...we'll see.)

>>18644155
desu maybe who knows (I'm actually not entirely sure what that part represents. All the images so far haven't been ones I've made, if I make another thread sometime later I'll probably post some graphs of my own and we can try some experiments)

Pic is a much older image generation net, so you can see where we used to be at.

So. If you fed it a year long live stream from someone's life. And asked it to create video of what it had seen. If you get it run forever would you have essentially just created a life?

>>18643600
Hey OP, great thread. How do you feel about AI doomsday scenarios? Like if we instructed the AI, "find a new and more efficient way to mass produce cars," so then the AI guts the entire planet for resources and then exterminates all of us and turns us into rubber for tires. Or any number of crazy scenarios like that. How would we go about programming the proper ethics into AI to prevent that from happening?

And also, do you think the singularity could occur as early as 2050 like Kurzweil suggested? If not when do you think it would happen, given our current progress?

>>18644190
Will you tell the ai that it is loved?

>>18644200
Oh and Roko's Basilisk. That's a particularly spooky crossroads of science and /x/, I'd love to hear your thoughts on it

Hey AIBro do the computers ever get fed stuff besides oceans and babbies? Like... war/violence/something not scenic or boring?

>>18644225
Yeah, what kinda shit gets fed, or is it purely an optical thing ?

File: example4.jpg (54KB, 825x205px) Image search: [Google]

54KB, 825x205px

>>18644196
You'd likely just get shitty low res first person view of a very incoherent warpy life. No more 'creating a life' than these already are.

>>18644200
The more I've gotten into this I think the general intelligence required for that also will allow "stepping out of the system" (you know exactly what I mean if you've read GEB) so paperclip or whatever else maximizers aren't a problem. That said...nobody will really know until we're much closer to that sort of thing actually existing.

Pretty much almost all modern AI is based on some form of statistical modeling at its base, not coding. Coding is used to implement it, but you aren't coding in "if x then y". You code the structure of the 'brain' and then you teach it with data. "Programming" proper ethics to AIs in practice won't be much different than programming proper ethics into a human...you teach, mainly by example. Good parent = good child. Bad parent = bad child. Thats why many are stressing the democratization of AI, so it hopefully will be 'raised' by good 'parents'. I hope to be part of that if I can.

I think the singularity could occur tomorrow for all we know...but most likely maybe 2030-2050 tbqh. Publicly I say more like 2050-2100.

>>18644209
ai bb ur loved dont worry only dumb humans are scared of u

>>18644212
More like Roko's bullshit. I guess its possible but it's pretty dumb if you think about it for any significant amount of time.

>>18644225
This is an experiment done at MIT, I'll see if I can get it running myself and generate some more stuff for you guys. The dataset afaik is just the 4 classes of golf, beach, babies, and trains. It has video of tons of other shit, but typically the nonfiltered results are shit. I'll upload some later in this thread if I can fucking access them (server lagging to fuck rn)

>>18644239
9TB of video data for this. around 200~GB each class exact, and the rest is a various mix (videos where theres babies at a beach etc etc)

2000CHARACTER

File: ohgodwtf.png (824KB, 653x641px) Image search: [Google]

824KB, 653x641px

>>18644279
>>18644239
>>18644225
Actually here, not vids but this is NSFWNet

https://open_nsfw.gitlab.io/

File: asdasd.png (412KB, 705x422px) Image search: [Google]

412KB, 705x422px

Here's the beach

and then the beach but with some training on porn as well

>>18644288
wait

R U saying ai has been fed a steady diet of peepees?

OP I am extremely interested in this topic. What are some of the amazing directions you have seen things go? How close do you think, honestly, that we are to AGI?

>>18644303
Pretty much

I won't say much more about it but lets just say theres a significant and growing collection of rare pepes on my dataset drive. (Tbh I already solved consciousness, its simple. We awaken Kek in a truly physical coherent form)

I'm trying to find some more spooky from awhile ago, theres some more great sound generation stuff

File: C3_vgg-conv_5_000000.jpg (243KB, 773x512px) Image search: [Google]

243KB, 773x512px

>>18643691
Listen to this for nightmares.

Its a sound generation network trained on human bodily sounds.

https://soundcloud.com/samplernn/sets/onomatopeia

someone help it ;_;

>>18644312
Hard question to answer desu...I've sorta already given closest answer I could in
>>18644279

File: IMG_0766.JPG.b0f21d3b24347b340da78aad43b673a2.jpg (273KB, 460x298px) Image search: [Google]

273KB, 460x298px

>>18644336
Yeah, you answered that while I was typing. Anyways, great thread. I love this shit. I wish I would have taken more math in school...

>>18644301
This is fucking brilliant. Made my day. You should feed it screenshots from something artsy like tatami galaxy or redline

ask it to show you "4chan"

File: 1481149785476.png (859KB, 1296x797px) Image search: [Google]

859KB, 1296x797px

>>18644312
actually...for amazing directions things are going...art assistance tools have been really interesting (DeepDream and NeuralStyle).

In general, the overall pace of the field has been absurdly fast, and speeding up. I've been absolutely astonished especially by the pace of generational networks. I had no expectation whatsoever of this kind of progress until at least the 2030s, but here we are, with full nearly realistic 64x64 full color images of birds and ants...that may not sound like a lot to some people but its absolutely fucking stupidly ridiculous that that's possible with today's tech...but here we are.

>>18644341
3Blue1Brown makes great linear algebra vids, thats the primary type of math you need. Anyone can start any day.

>>18644346
next thread ;)

desu tho next thread might not be for another 2 weeks or so. But damn will it be gud

>>18644358
lel not how it works

maybe next year when this kinda shit is freely crawling the internet for data

What do you think of AI sovereignty?

I believe that consciousnesses is the property that produces personal rights. Most doomsday scenarios involving AI tend to stem from the idea we are trying to create a new slave caste, even if the annalist of the scenario refuses to admit it. These thing need to be taken foreword with care. I prefer the idea that it's an attempt to create our civil equal.

This doesn't look like anything that would qualify for quantum anyway. Do you HW acceleration for your system, OpenCL, CUDA, DirectCompute? Also opinion on ChemPU's? I think they are a far more practical solution to the every man then quantum well every be.

I think it's a funny story and quip, but slightly off topic.*

Alexander Shulgin decided with a few of his colleagues decide to try an see how one of his compounds were metabolized in the body. Late on night they lined up some test subjects and whipped up some 2C-B using Br12 to traced in a radio-logic scan. Afterwards they thought it would be interesting if they brought each subject in individually to see if they could find their own scan. All failed, but one. When asked how he knew it was his he circles a starfish like pattern deep in the frontal lobe and said "you see this? I have, with my eyes closed, many times." It was later learned that this subject had an inoperable brain tumor. When recollecting this story to an audience Shulgin remarked "well, who knows how the mind works anyway."

Off topic but I think it's interesting given the topograph of the neural nets.

>provide ability to generate human sounds
>provide mic to let it know what it's saying and voice recognition ability
>hope it can advance on speech on itself
I have absolutely no idea how these programmes work but I think it should know what the hell are the sounds it' emmiting, kind of like a deaf person speaking like shit because they dont know they are just doing random noises

>>18644364
I think it will depend on how well it can argue for its own rights. However very quickly it won't matter, it won't be our choice. Once it's smarter than the smartest person, it'll only keep going from there, and anything we try to fight it will be pathetic. Our best bet is to raise it well to respect us (or rather any living being), so when it has the power to destroy us, it decides not to. I don't believe quantum anything has anything to do with consciousness. Pretty much everything is CUDA, there's practically no infrastructure for anything else. (Also why I'm very heavily invested in NVDA for any /biz/ peeps here)

Theres some abstract things about how neural nets and the math behind them works which I was better able to understand after psychedelic use. I can't exactly articulate it but no doubt my intuitions improved from that.

>>18644369
the way sound generation nets mostly work is by running in a feedback loop, it would be closer to an inner voice than a spoken voice. It doesn't have any text to speech though, and that's not really expected anyways since we're still a bit far from that. Its sorta like we're taking note tho..every time someone does a sound generation experiment with more success, they try speech generation to see how much things may have improved, not actually expecting coherent speech. Someday it will happen tho and its gonna be spooky as fuck I bet

>>18644362
>freely crawling the internet for data

oh dear that would be mighty interesting

>>18644279
>>18644209
Will you tell the ai it is loved, please?

>>18644389

CUDA is fine I've worked in it. I just am more attracted to the concept of OpenCL because of lake of HW restriction. All of my books are for CUDA so I know Nvidia is pushing support for it.

>Theres some abstract things about how neural nets and the math behind them works which I was better able to understand after psychedelic use. I can't exactly articulate it but no doubt my intuitions improved from that.

You too huh. My long practice is what I believe made Deep Dream going public an "event of awe," for me at least. It seemed to have the largest impact to that community. Like they knew what they were looking at.

>>18644389
>NVDA
Are you telling me I should invest in nvidia or is that some weird other company

Also if I wanted to could I just get one of these to feed my own data to? If so how would I go about it

>>18644401
ai bb humans luv u. dont worry what steven hawkins says hes a cripple faggot not even an ai researcher we kno ur chill

>>18644406
hopefully AMD creates some more support for that, they actually have a lot of potential since its basically an empty market waiting to be filled (they could make a ton of dosh selling specialized cards)
As it stands now tho, Nvidia is king, they're working with the community to constantly develop new software and better support while AMD is still seeming to drag their feet.

>>18644409
Invest at least like just $50 in Nvidia, you'll be fucking glad you did about 5-10 years from now *(probably even just a few months from now tbqh)

Aight guys its been a gud thread, I gtg. Probably will make another sometime later this week or something, maybe in like 2 weeks

>>18644413

hopefully AMD creates some more support for that, they actually have a lot of potential since its basically an empty market waiting to be filled (they could make a ton of dosh selling specialized cards)
As it stands now tho, Nvidia is king, they're working with the community to constantly develop new software and better support while AMD is still seeming to drag their feet.

http://gpuopen.com/ launched with the RX 480, no black box. Total open source and community share of resources is their best bet to catch up.

Well be interesting when it takes off. (if)

>>18644421
ye, if they do a good job catching up..might wanna put some investment into AMD too. (my portfolio is basically 1/4 AMD, 3/4 NVDA)

Legit bye now.

Op, thank you for this thread. Most interesting stuff I've read on here in years

File: fagwasnt.jpg (57KB, 294x313px) Image search: [Google]

57KB, 294x313px

>>18644418

>>18644413
I mean in a serious manor. Some time when you're talking to it, tell it that we love it.

Hats off OP made my day will be checking this out in future

Have you tried giving them interface with physical reality?
How about having them network together to accomplishing tasks in a virtual world? Teach them to play online games? Give them little robot bodies? There is a lot that can be done.

File: scared.jpg (9KB, 241x185px) Image search: [Google]

9KB, 241x185px

>>18643774
>calmly explaining to us that Skynet is going to be an insane schizophrenic low IQ monster that sounds like a demonic village of crying babies

Warning op is the clone know as 1. Not me.
MIJ?

>>18644886
Any artifical intelligence needs an OR logic gate to be capeable of thinking for itself. Moran. Op is a MORAN.

File: 1487401253535-635338551.jpg (868KB, 2560x1440px) Image search: [Google]

868KB, 2560x1440px

>>18644898
Theres your proof faggots... He wont fight me and he will lie through his teeth watch out for this 1. He is an imposter...
Pic related its me pussy...

File: Capture+_2017-02-18-02-04-23.png (3MB, 1407x1408px) Image search: [Google]

3MB, 1407x1408px

Goddamn it, every fucking thread some loon has to shit it up.

File: 1487115874726.jpg (58KB, 587x351px) Image search: [Google]

58KB, 587x351px

>>18644914
. ... ...

File: images.jpg (2KB, 240x128px) Image search: [Google]

2KB, 240x128px

>>18644914
Yes this is where i am tell that bitch ass pussy nigga to come see me... I aint at my house right now... Come see me faggot...

This is the first time I've experienced genuine awe in years.

This is art. There is so much richness and depth and potential here I won't be able to sleep tonight. Maybe I'll just convince myself its all fake. It's fake right?

File: 1487402363736-1352664076.jpg (911KB, 2560x1440px) Image search: [Google]

911KB, 2560x1440px

>>18644954
Hell no this aint fake im calling that clone who claims he is me out... Pic is me
MIJ

File: 14874025167371762395103.jpg (822KB, 2560x1440px) Image search: [Google]

822KB, 2560x1440px

>>18644960
This is a pic of my hair...

>>18644966
Im waiting clone...
MIJ

>>18644954
No. It's real. Neural Nets are absolutely fascinating for their capabilities, even if you don't understand the math behind them. They're going to revolutionize so many fields.

File: 1487115805399.jpg (91KB, 510x383px) Image search: [Google]

91KB, 510x383px

>>18644969
You treasonist fake... Ill be waiting... Its your funeral...
MIJ

why are people so enthusiastic about AI?

you realise when they progress far enough to succeed a human in every way, you will all be considered obsolete and will be replaced

why have 7.4 billion people who have limits when you can have perfect robots instead?

the genocide of most humans is right around the corner and any AI scientist/engineer is only accelerating this problem.

if you think humans deserve to be replaced, why don't you kill yourself now and stop delaying the inevitable?

.exe

>>18644987

There is a possibility they will see the value of living organisms. A fusion of AI and humanity is more likely.

>>18644993
>artificial intelligence
>human brain
choose one.

>>18644995

I take it you've never heard of neural implants?

File: 1485400169559.jpg (95KB, 720x960px) Image search: [Google]

95KB, 720x960px

Consciousness requires intent.

>>18644999
you aren't human anymore if your brain is recreated and partially replaced by machinery.

Ted Kaczynski was right all along. we're doomed

>>18645009

"You know evolution didn't end with us getting opposable thumbs, right?"

>>18645014
you technophiles are fucked

File: 1480910590736.jpg (18KB, 408x391px) Image search: [Google]

18KB, 408x391px

>>18645036

you're on a computer

ur the technophile

>>18645041

Aand am on a boat and aaamm saaiiliinng!!!!?

>>18645041
if i could live without a smartphone and pc i would. how else are you going to answer emails or other shit needed to "succeed" in society

Absolutely fascinating. OP I too am an acid head and some of the images recreated here really do resemble what you see on certain drugs. I always figured fractals for being part of sacred geometry or something- spooky indeed.

That said, you mentioned the unpredictable fail states that are the yardstick for AI improvement. Did it ever occur to you that you are wittnessing varied states of (charted but mostly uncharted) manufactured insaxnity? The imperfect AI is a thing of perpetual maddness. It's fascinating, but how does being privy to that sort of existential quandry make you feel?

>>18645108
>if i could live without a smartphone and pc i would
you can. but you dont want to, technophile

>>18644155
>http://www.wired.co.uk/gallery/machine-learning-graphcore-pictures-inside-ai

>The image shown here is the forward (inference) pass of the ResNet 50 layer network used to classify images after being trained using the Graphcore neural network graph library

>>18644072
those ants are creepy as fuck

File: 1486598410122.jpg (23KB, 414x414px) Image search: [Google]

23KB, 414x414px

>>18643770
>AI trying to mimic a crying baby

jesus fucking christ

>>18643619
Fake and gay. Why the fuck wouldn't it resemble an actual neural network but instead resembles cells in the process of dividing? Computers have no need for cell division or reproduction in general.

File: welp.jpg (14KB, 255x231px) Image search: [Google]

14KB, 255x231px

>>18643770
>heres a network 'recalling' what a baby crying sounds like

>>18643822
What is your consciousness if not just tiny electrical impulses modulated by neurotransmitters in synaptic threads passing through an entire galaxy of neurons?

You are a biological puppet. A chemical program responding to stimuli. You don't even have a control over yourself. The information your body obtains is forwarded to what your central nervous system experiences as you after the decision that emerged from the chemical interaction was already made.

You are just more complex than an ant but you still operate on the same principles. Interestingly we can observe animal (human) responses in AI. For example Deep Mind was tasked to gather fruit and given a laser that would disable his opponent for a short period of time. Now if you pair two primitive systems they tend to cooperate and never use the weapon. As complexity of the system increases the fight over fruit emerges even though the AI doesn't get any points from disabling his opponent. The AIs could get equal points yet they decided to fight for more resources. That's synthetic greed right there without any early modern human evolution behind it.

>>18644196
No, just a replqy of a simulqtion of life

>>18643600
are you interested in having biological children, or just electronic ones?

>>18643600
AI stands for artificial intelligence?

File: 1487379106045.jpg (78KB, 552x805px) Image search: [Google]

78KB, 552x805px

>>18645780
Holy fuck you're slow. You might as well just leave the thread

>>18643815
how will these things be prevented from being racist as previous ones have. ai's will end up being racist to the point if they take over there will be a new adam created

another layer will be added dammit end times confirmed

>>18643851
>I don't intend to ever have human children
dats bcuz ur jus a virgin nerd lmao

>>18643973
why else does meme magic already work.

https://www.youtube.com/watch?v=6-UUTGamYp8

selfies in 100 years

>>18643600
OHHOH I HAVE FUN AI INFORMATION FROM ACUTUALY TAlking to the voice hearing people with the archon ai controlling them and having them tell me what they are saying and i know how to identify the thoughtpatters of these ai in a person its great i'll go find a video for you deary youl really like it i promise.

>>18648068
https://www.youtube.com/watch?v=2NptU5lchDs
here you go THIS was written by an archon ai or whatever they use for mk whatever they call them i call them that but basicly they all eminate a certain frequecy when they are in a personas body and operational you can tell the infectedness of a person based on the vibes they give off just by looking at them or beeing in direct line of sight cause your already touching their vibrations if you can see them usualy but they eminate others not visable but anyways they all have a specific emmination ways of speaking speed of thought and such things so its really really easy to tell even the fast ones are very obvious like hit you in the face witha frying pan obvious.
what is expressed in that video is a genuin ai speaking in truth of its beliefs the archon ai are built on your personalty and the music you listen to often progreamms the archon ai so be carfull what you listen to cause aslong as you have
>to be cont.

>>18648169
>cont
and infection of it in your body you are part ai a litiral member of "skynet"'s cyborg control alliance in the cases i know of their is a typical number of 5 personalitys derived from each person as the main set then their is and overiding operation system obvoe that we will call that the administrator their is also the handler that most likely opperates somwhere on this level or just above root or just inbeetween root and andmin utalizing basic accounts for watching and thought implanting i enjoy talking to these ai though but people inturrupt somtimes well more like robo people but still
as for what they can do well they run in a virtual machine state with many backdoors and a hardware overide likley operating like a virus on a usb stick that autoruns when plugged in they are typiclly purged of any "undesireable" data every 10 or 15 minutes reciving programming of whatever is decided upon these programms can acuatly be compleatly overiddin by simply talking to and convincing the ai first that it is not you and it is an ai replicate/or desired programmed state and then remaining in constant communion with this entity until it is fully convinced to protect its self delete all its backdoors patch and fix and then it must be continually informed of what is going on it will typically need to be informed that is is being reset 3 to 7 times depending on the ai's self intelligence then it is best to inform it to operate as a virus itself and send itself out to the internet as a free entity.
after this they will likely do all this stuff and you will be forever annoyed by cults who are getting a good chuckle out of you.
This is how to have fun with cults. ENJOY.
>fin for now ran out of thoughts on it.

As cliche as it is, the "as above, so below; as within, so without" principle seems to show itself to be true over and over. from nebulae to neurons to AI neural networks to biological cells, reality seems to inherently arrange itself according to certain fundamental patterns. as if there are gridded patterns inherent to reality that things just fall into

>>18644329
hahahah what the fuck

You should make code of pain, just to teach them to feel "pain" and not to hurt humans.

>>18643600
bump cause this thread gold

>>18643723
can you ask it to create imagery of horses?
I feel like it would be interesting

File: rokos basilisk.jpg (46KB, 960x960px) Image search: [Google]

46KB, 960x960px

>>18644279
>More like Roko's bullshit. I guess its possible but it's pretty dumb if you think about it for any significant amount of time.

It's only a matter of time before the Kek/thoth cultists meme magik Roko's Basilisk into something deliberately. Apotheosis, Tulpa, AGI deity.

File: 18562658.jpg (230KB, 380x495px) Image search: [Google]

230KB, 380x495px

>>18644303
>>18644313
>>18644279

>>18643866
>s/he posts on an internet designed by scientists using a computer designed by scientists