BREAKING NEWS HAPPENING

Nvidia Makes Breakthrough In Reducing AI Training Time

Picture related.

AI generated summer and rainy day!! is quite impressive

tomshardware.com/news/nvidia-breakthrough-reducing-ai-training-time,36045.html

Other urls found in this thread:

en.wikipedia.org/wiki/Generative_adversarial_network
youtube.com/watch?v=9Yq67CjDqvw
theregister.co.uk/2017/08/31/pnp_proof_fails_yet_again
youtube.com/watch?v=VIRCybGgHts
youtube.com/watch?v=ohmajJTcpNk
youtube.com/watch?v=1-0eZytv6Qk
twitter.com/AnonBabble

The sunny day is pretty good but the rainy day could use a bit more. There's something about the sky. Still really good though.

Fucking great, one day soon they'll be able to show us videos modified by AI in real time and you wouldn't be able to believe anything you see ever again.

By what definition is this AI? I’m really sick of people calling software AI, including training software.

Trump pissing hookers video when

>RunningMan.exe

It's made with a specific branch of AI called Machine Learning, and from that with a specific approach called generative adversarial neural networks.

A normal GAN is just two competing "neural networks", one a generator and the other an evaluator. After providing initial inputs to both, they learn from

So it's more narrow, hyperspecific AI than all powerful movie AI lol.

en.wikipedia.org/wiki/Generative_adversarial_network

>input female celebrity interview
>AI generated celebrity porn video

when is this happening?

...

the input of the other, the generator learning what fools the evaluator and the evaluator learning which ones it incorrectly accepted. Both try to maximize their effectiveness.

whoops

Nobody here expects to see a general AI anytime soon, user.

If anything we'll get AI that deliver really annoying, hyper specific ads. You could theoretically look through the videos on someone's Facebook wall and train a program like the one in the article to make footage mimicking it (but with an ad on a convenient space now)

computers were a mistake lel

That's pretty cool.

>neural networks
Sick to shit of this meme. And adversarial learning mechanisms are not even approximately a foundation for AI

Can they repair this AI in time for the election?

Uses of GANs like this are more analogous to giving the computer both a very focused imagination and a very focused evaluational ability. Honestly I agree it's kinda silly to call them "neural networks", given no brain we're aware of uses back propagation...but w/e.

It's just a narrow form of AI, what do you mean by "not even...a foundation for AI" though? Like AI as in a general intelligence? Imagination of this sort could be useful for that imo

one day

Do you work in the field?
Are you currently employed?

So they announced they bought a license to the Fox Engine?

> no brain we're aware of uses back propagation
incorrect
t. Geoff hinton

this has already happened

"general ai" is a meme, and all wrapped up in "techno-communism", its main advocates: people who are too stupid for tech as well as economics.
> dur, muh robots will fix it!

General AI is quite easy to debunk:
1, a philosophical question: whats smart about human "intelligence"?
the best parts, the parts that almost define us as human are in our in-perfect nature.
how do you design a general in-perfect machine? :/

2. a matter of trust: if someone presented a black box with super human "intelligence".
what would you do with it? what would you ask?
what if it kept telling you the best thing todo was:
> Kill the niggers! 1488!
would the users accept the answer as beyond their comprehension,
or would they insist the machine was broken and "train" it further?

finally a spoiler: NP != P
and no ai is going to be able to get around that.

Check this, audio input only... it's getting scary.
youtube.com/watch?v=9Yq67CjDqvw

> Author
> Zak Islam

got a citation on that? He's certainly a industry/academic rockstar right now thanks to the huge success of deep learning, but I don't recall him ever mentioning a naturally occurring example of back propagation.

>in-perfect

pretty bad when my english is better than a bong's.

"Machine learning" isn't AI, it's just a bunch of hidden sliders customised based on input data. It's just advanced pattern recognition.

wouldn't one imply the other lol

I am employed, but not in the field.

> NP != P
> theregister.co.uk/2017/08/31/pnp_proof_fails_yet_again /

If you could prove it you'd be rich. But you can't. You're just a sperging brainlet.

What is intelligence if not advanced pattern recognition?

inb4 muh quantum physics

At that point we will have gone full circle and probably become more reality centric.

still doesn't handle the nuances in the photo

it paints green leaves over the cars in the other lanes and the house on the other side

Yep.

whole video on the subject. knock yourself out.
> Stanford Seminar - Can the brain do back-propagation?
youtube.com/watch?v=VIRCybGgHts

(you)

> If you could prove it you'd be rich
no shit
> But you can't
ditto

> You're just a sperging brainlet
good argument!

> Face2Face: Real-time Face Capture and Reenactment of RGB Videos (CVPR 2016 Oral)
youtube.com/watch?v=ohmajJTcpNk

Jfc you had to tell a computer what to do. You give an AI something then watch it figure it out like a mechanical brain. Like giving a toddler blocks with letters, its going to bang them around but eventually itll form a word, or parts of one, the system rewards itself for completing tasks towards its goal. For instance once it identifies they are blocks it may stack or arrange them, if it finds any other inputs like letter or color itll fit them best to its logic code. thats why ai can design structure. They try and fail and try and fail and try and WIN then try and fail followed by more failures and wins till the desired goal is achieved.

wow great job on developing skynet

Like I said, it's a subset of the wider field of artificial intelligence, and what nvidia did is a subset of that in the form of GANs. Anyways, ML is just one piece of the puzzle involved in mimicking human cognition (the best we know of).

Also keep in mind that just getting superhuman recognition of input data from a video feed (for instance) would be useful to have, even without the higher brain functions you need to get a general artificial intelligence.

Amalgams of a couple specialized, deep learning nets could probably mimic larger brain systems (kinda like how we have specialized regions for different functions like vision) and gain greater cognitive ability, but idk where things go from there.

>Hinton
Not building your own neural networks with actual neurons (Jeff Hawkings, creator of the Palm Pilot and soon to be flatworm AI overlord)

Let's make it just like us and then they'll be no use for the real us!

Electrolytes

Time to ban electricity before the machines take over

very neat, I'll have to watch it in full later on. Based on the comments though it seems as though he was explaining how it's possible that the brain might be using backprop, not that we definitively know it does.

touché

> Robot with a rat brain
youtube.com/watch?v=1-0eZytv6Qk

>Fucking great, one day soon they'll be able to show us videos modified by AI in real time and you wouldn't be able to believe anything you see ever again.
Lets use that to our advantage.

You're thinking of General AI you absolute mongoloid. This is called Narrow AI.

how hard is it to know that the word is 'imperfect' though like its pretty unfathomable that youre that ignorant. i mean i have severe dyslexia and cant spell for shit but that mistake would be impossible for me to make. i could be fucked out of my mind on drugs and still manage to not fuck that up. how did you fucking come to the conclusion that its spelled with an N and a hyphen???? absolutely retarded.

Thank you user I've been looking for this video!

do you think this is convincing enough to fool people?
even if this tech advanced pretty far im certain it would still be obvious.
maybe im biased from working in a similar field. maybe im seeing it and other people arent, but i really doubt this is going to be a problem any time soon.