Yes, androids do dream of electric sheep

now find dogs in buch of sounds.

This reminds me of cyriak

google you are a bunch of stupid buttholes if you don't teach us how to make these

Google: The Future Of Goa Trance Album Covers Is Now

lol

http://www.twitch.tv/317070/mobile

Woops, not mobile specific (if it makes any diff)

http://www.twitch.tv/317070

i really want these full-size for a poster.

they are seriously IDM

im tripping just looking at them

just messaged the creators to see if we can get full-sized images to print.

they are totally bonkers

ive often hoodwinked into clicking isis beheading links in twitter that send me to archive.org, onl to discover archive is spilling over with electric sheep.

i don't really get it, but i like it. It's an odd presentation if you go to the actual source of the images (they are in fairly low res) so part of me thinks this might be some kind of vaporware theoretical thing that is more human aided than they claim. They need to release this software immediately

  On 6/19/2015 at 6:42 AM, mokz said:

Yeah, basically supports the idea that psychedelics put the human pattern recognition apparatus on overdrive. Seeing forms where should be just simple patterns.

 

Somebody should do this to audio.. creating speech from white noise, etc.

i think scrambledhackz is probably the closest thing i've seen to this in audio form. Or maybe that one guy who made a piano do speech synthesis, kind of similar but not really. audio version of this that sounds as good as this looks is probably quite a ways off. You would need something basically more powerful /adaptive than the Kyma spectral detection and sinusoidal decompisition algorithms

  On 6/25/2015 at 10:02 AM, John Ehrlichman said:

i don't really get it, but i like it. It's an odd presentation if you go to the actual source of the images (they are in fairly low res) so part of me thinks this might be some kind of vaporware theoretical thing that is more human aided than they claim. They need to release this software immediately

Well, it's neural networks who produced those images. it's not that you "release" a neural network. It has to be built and it has to be trained in order to properly classify what it's fed. Moreover, I bet it's no "little" network. I bet there are the details somewhere.

To make an example. Imagine that you want a neural network to always recognize the "c" character. You feed it with a large numbers of c images, in different shapes and sizes. In this way you tell it, this is what I want you to recognize. After some training (which involves feeding also a small set of characters different that "c", so that it doesn't "over recognize" a "c" character), the neural network will tell you if the image provided is a "c" character or else. A neural network trained to recognize a "c" character, will never recognize a "d" character as "d" but as a "non c" character.

So, basically, those neural networks were trained to recognize a category of things and to tell whether what it's fed to them belongs to that category or not. What they "see" is what they were taught to see. What we see in those images then, it's the training set they were trained with.

p.s. my last sentence is a guess, obviously, but I think I'm not so off from what really happens

im guessing the pareidolia thing is because they were trained to recognise earthy things, like people and animals?

  On 6/25/2015 at 10:26 AM, lala said:

im guessing the pareidolia thing is because they were trained to recognise earthy things, like people and animals?

Yep I'm guessing that too, which basically it's the same thing our brain do.

  On 6/25/2015 at 10:22 AM, pierlu said:

 

  On 6/25/2015 at 10:02 AM, John Ehrlichman said:

i don't really get it, but i like it. It's an odd presentation if you go to the actual source of the images (they are in fairly low res) so part of me thinks this might be some kind of vaporware theoretical thing that is more human aided than they claim. They need to release this software immediately

Well, it's neural networks who produced those images. it's not that you "release" a neural network. It has to be built and it has to be trained in order to properly classify what it's fed. Moreover, I bet it's no "little" network. I bet there are the details somewhere.

 

To make an example. Imagine that you want a neural network to always recognize the "c" character. You feed it with a large numbers of c images, in different shapes and sizes. In this way you tell it, this is what I want you to recognize. After some training (which involves feeding also a small set of characters different that "c", so that it doesn't "over recognize" a "c" character), the neural network will tell you if the image provided is a "c" character or else. A neural network trained to recognize a "c" character, will never recognize a "d" character as "d" but as a "non c" character.

 

So, basically, those neural networks were trained to recognize a category of things and to tell whether what it's fed to them belongs to that category or not. What they "see" is what they were taught to see. What we see in those images then, it's the training set they were trained with.

 

i see, so part of me wonders why these are the only images they are showing so far. Do you think these take a very long time to process? the way you described it is very similar to how scrambledhackz works but probably on a much smaller scale

i'll ask the developers if they get back to my email how it all works

  On 6/25/2015 at 10:34 AM, John Ehrlichman said:

i see, so part of me wonders why these are the only images they are showing so far. Do you think these take a very long time to process? the way you described it is very similar to how scrambledhackz works but probably on a much smaller scale

 

Well, probably they are showing only the meaningful ones, out of a meaningless bunch. As of time, I think the shorter time is achieved by having lot of RAM, more than processing power, but here I'm in the real guessing realm.

Scrambledhackz, from what I remember, seems similar, but it's quite different. Every piece of audio is mapped based on it's frequency rappresentation, and the ones that more resembles another is choosen. So it's more about metrics (i.e. defining a meaningful way of measuring "distance" based on frequency content), than recognition (which is more like "this resembles that because I tell so based on my experience").

Edit: see it this way. "does this image contain a dog?" -> neural network. "Does this image contain mostly the red color?" -> Metrics

yeah good point, the recognition aspect takes it to another level. not sure if this has already been posted but you can view the process in real-time and suggest inputs here : http://www.twitch.tv/317070

its still low res but definitely not some kind of vaporware, looks like a legit real time process

the sameness of the color scheme only seems to interrupt when someone suggests a colored object or animal like 'red panda' or 'green snake'. Most of the time its a very blue tinge. Animals or objects with very distinct textures (like armadillo or alligator) create really cool looking joined together texture maps. man if the resolution was higher this thing would be amazing for pulling images out of.

holy shit. into the matrix we go

That shit is way trippy guys

  On 6/25/2015 at 10:36 PM, John Ehrlichman said:

yeah good point, the recognition aspect takes it to another level. not sure if this has already been posted but you can view the process in real-time and suggest inputs here : http://www.twitch.tv/317070

 

its still low res but definitely not some kind of vaporware, looks like a legit real time process

 

the sameness of the color scheme only seems to interrupt when someone suggests a colored object or animal like 'red panda' or 'green snake'. Most of the time its a very blue tinge. Animals or objects with very distinct textures (like armadillo or alligator) create really cool looking joined together texture maps. man if the resolution was higher this thing would be amazing for pulling images out of.

i've had this tab open all day while i install laminate wood flooring.

sulfur butterfly into lip rouge was quite lush.

some are easy to make out what it's projecting like pelican & tarantula. but others, like space shuttle & cheeseburger, are just madness.

edit:wolf spider into pineapple into maillot back into wolf spider = lushest/mostidm

highly recommend oversteps as listening material to watch that.

dank

.

i had to re-post this.

insanity.