|
Petit Gregory posted:The new Sufjan Stevens album is banging!
|
#
?
Nov 6, 2018 08:45
|
|
|
|
| # ? May 24, 2024 20:05 |
|
|
Negrostrike posted:https://www.reddit.com/r/Damnthatsinteresting/comments/9udese/plugging_50000_portraits_into_facial/
|
|
#
?
Nov 6, 2018 17:00
|
|
|
ToxicFrog posted:This isn't procedurally generated per se, but it seems like the kind of thing this thread would appreciate: the most unwanted song.
|
|
#
?
Nov 6, 2018 21:01
|
|
|
Woebin posted:The rap bits on this have some Nina Hagen vibes, I'm kind of digging it. Are we talking about the same Nina Hagen? The one with the sorta deep, smoky voice? https://www.youtube.com/watch?v=YxVlDICaeq8 Although... watching that video, she does start off more, uh... moderate. And i admit ca. 1983 Hagen does a good soprano voice. edit: drat, that's a hella wide range of musical styles. P. sure she hit all of them at some point. The example from 2003 is amazing. HenryEx has a new favorite as of 21:23 on Nov 6, 2018 |
|
#
?
Nov 6, 2018 21:20
|
|
|
Woebin posted:The rap bits on this have some Nina Hagen vibes, I'm kind of digging it. OUT ON THE PLAINS JUST ME AND MY MIND TOOK ME A BREAK TO READ SOME WITTGENSTEIN Western Philosophy Soprano Rap sounds like a GAN-generated muscial genre in itself.
|
|
#
?
Nov 6, 2018 22:12
|
|
|
Not sure if this counts but here's a google doc that collects a bunch of hilarious AI behavior that evolved that games the evaluation function or just exploits the simulation https://docs.google.com/spreadsheets/u/1/d/e/2PACX-1vRPiprOaC3HsCf5Tuum8bRfzYUiKLRqJmbOoC-32JorNdfyTiRRsR7Ea5eWtvsWzuxo8bjOxCG84dAg/pubhtml My favorites include quote:Agent kills itself at the end of level 1 to avoid losing in level 2 quote:Robot hand pretends to grasp an object by moving between the camera and the objective quote:Creatures bred for speed grow really tall and generate high velocities by falling over quote:Genetic debugging algorithm GenProg, evaluated by comparing the program's output to target output stored in text files, learns to delete the target output files and get the program to output nothing. People gripe about malicious AIs but I feel like these AI are actually really lazy Xun has a new favorite as of 12:49 on Nov 8, 2018 |
|
#
?
Nov 8, 2018 12:44
|
|
|
Like asking a djinn for a wish. You get exactly what you asked for and exactly what you didn't want. Love the one where they got tall and fell over to go fast 
|
|
#
?
Nov 8, 2018 15:41
|
|
|
Xun posted:Not sure if this counts but here's a google doc that collects a bunch of hilarious AI behavior that evolved that games the evaluation function or just exploits the simulation Just don't give them a task like 'minimize human suffering plz'. The most worrying detail was that cycleGAN had spontaneously invented steganography: Xun posted:A cooperative GAN architecture for converting images from one genre to another (eg horses<->zebras) has a loss function that rewards accurate reconstruction of images from its transformed version; CycleGAN turns out to partially solve the task by, in addition to the cross-domain analogies it learns, steganographically hiding autoencoder-style data about the original image invisibly inside the transformed image to assist the reconstruction of details.  That's not 'set all standards to 0' lazy, that's 'too drat smart' lazy.When cycleGAN fails: 
|
|
#
?
Nov 8, 2018 15:43
|
|
|
Megillah Gorilla posted:Love the one where they got tall and fell over to go fast https://www.youtube.com/watch?v=puxvC9qGP_o
|
|
#
?
Nov 8, 2018 15:49
|
|
|
Xun posted:Not sure if this counts but here's a google doc that collects a bunch of hilarious AI behavior that evolved that games the evaluation function or just exploits the simulation The one at the bottom, the innocuously-named World Models, is about a Doom AI that imagines an entire world model in which it has the power to stop fireballs by moving in odd patterns. As in, it inexplicably stops attacks even after they've been thrown at it by just vibrating on the spot. It's just really funny to me that this bot imagines it has superpowers because of this exploit in the simulation.
|
|
#
?
Nov 8, 2018 16:22
|
|
|
Xun posted:Not sure if this counts but here's a google doc that collects a bunch of hilarious AI behavior that evolved that games the evaluation function or just exploits the simulation These all read like things that lazy/cheating humans would do so AI has truly come a long way e: like this, this is exactly what a human being would do in a poorly coded multiplayer game quote:Creatures exploited physics simulation bugs by twitching, which accumulated simulator errors and allowed them to travel at unrealistic speeds e2: I love these quote:In an artificial life simulation where survival required energy but giving birth had no energy cost, one species evolved a sedentary lifestyle that consisted mostly of mating in order to produce new children which could be eaten (or used as mates to produce more edible children). e3: Again, exactly what an online troll would do quote:An evolutionary algorithm learns to bait an opponent into following it off a cliff, which gives it enough points for an extra life, which it does forever in an infinite loop. e4: lmao quote:Agent pauses the game indefinitely to avoid losing quote:Evolved player makes invalid moves far away in the board, causing opponent players to run out of memory and crash Phlegmish has a new favorite as of 01:09 on Nov 9, 2018 |
|
#
?
Nov 9, 2018 01:00
|
|
|
Megillah Gorilla posted:Love the one where they got tall and fell over to go fast  This game makes so much more sense now https://www.youtube.com/watch?v=0eu-bHxLNys
|
|
#
?
Nov 9, 2018 01:29
|
|
|
An AI posted:AI trained to classify skin lesions as potentially cancerous learns that lesions photographed next to a ruler are more likely to be malignant. 
|
|
#
?
Nov 9, 2018 01:52
|
|
|
Phlegmish posted:
That one made me laugh out loud, it's such a distinctly computer-y but banal solution. 
|
|
#
?
Nov 9, 2018 02:04
|
|
|
We're this close to creating an AI that can flip the board when it's losing.
|
|
#
?
Nov 9, 2018 03:05
|
|
|
Here is a list of some machine programs solving the stated goals in the most "I GUESS" ways possible. https://docs.google.com/spreadsheets/u/1/d/e/2PACX-1vRPiprOaC3HsCf5Tuum8bRfzYUiKLRqJmbOoC-32JorNdfyTiRRsR7Ea5eWtvsWzuxo8bjOxCG84dAg/pubhtml
|
|
#
?
Nov 9, 2018 04:16
|
|
|
Inzombiac posted:Here is a list of some machine programs solving the stated goals in the most "I GUESS" ways possible. Xun posted:Not sure if this counts but here's a google doc that collects a bunch of hilarious AI behavior that evolved that games the evaluation function or just exploits the simulation However, here's a paper on a few of them: https://arxiv.org/abs/1803.03453 https://www.alexirpan.com/2018/02/14/rl-hard.html Queen Combat has a new favorite as of 04:27 on Nov 9, 2018 |
|
#
?
Nov 9, 2018 04:24
|
|
|
quote:Neural nets evolved to classify edible and poisonous mushrooms took advantage of the data being presented in alternating order, and didn't actually learn any features of the input images I love that the AI figured out it didn't need to learn anything because the grad assistant was lazy.
|
|
#
?
Nov 9, 2018 04:32
|
|
|
randomize input order is like step 1 when reading ai tutorials, too.
|
|
#
?
Nov 9, 2018 06:23
|
|
|
I think it's been mentioned in the thread, but an AI was tasked with seeing if Tetris could be played forever, so it just hit pause. Objective achieved. Another AI was tasked with achieving a high score in Q-Bert, so it found a previously unknown bug in the game and exploited it.
|
|
#
?
Nov 9, 2018 21:10
|
|
|
evobatman posted:I think it's been mentioned in the thread, but an AI was tasked with seeing if Tetris could be played forever, so it just hit pause. Objective achieved. Both of those are in the doc Xun and I posted. I love the one where it had to run as fast as possible to a goal in a straight line. Every round it got to make a change it its own body. The ultimate solution was to make itself as tall as possible and just fall over. "Stupid master didn't say my WHOLE body had to cross!"
|
|
#
?
Nov 9, 2018 23:27
|
|
|
Crossposting from elsewhere:quote:Nvidia researchers trained a neural network to generate images of cats and from the internet-sourced input data it learned that Impact 24pt White is a defining characteristic of cats 
|
|
#
?
Nov 11, 2018 18:57
|
|
|
Besesoth posted:Crossposting from elsewhere: This is one of my favorite threads even if no one posts here.
|
|
#
?
Nov 11, 2018 19:13
|
|
|
Besesoth posted:Crossposting from elsewhere: I love the captions on this one. It's sort of like reading in a dream where the letters shift around and change. Also, new post from Janelle Shane for those who haven't seen it: http://aiweirdness.com/post/179938486347/cnn-headlines-according-to-a-neural-net quote:Walmart Grilled With a New Leader in Murder Tech
|
|
#
?
Nov 12, 2018 05:38
|
|
|
SerialKilldeer posted:Amazon Wants to Make Money Broadcasting from Your Phone Yeah, probably. 
|
|
#
?
Nov 12, 2018 07:28
|
|
|
Google made a BigGAN-based (or -related?) tool available that morphs between two subjects: Watch -> Spider https://twitter.com/lizardengland/status/1062106160734629888 I recommend checking the whole Twitter thread by @lizardengland. Another one from there: Purse -> Scorpion: https://twitter.com/lizardengland/status/1062109392340951040 You can find it here: https://research.google.com/seedbank/seed/5629842393399296
|
|
#
?
Nov 12, 2018 23:58
|
|
|
Hempuli posted:You can find it here: https://research.google.com/seedbank/seed/5629842393399296 Here's a game: try to guess what these are transforming from and into!  Dog to Tarantula  Baboon to Jellyfish  Corn to Thimble
|
|
#
?
Nov 13, 2018 01:46
|
|
|
It's nice of them to make a tool to appeal to transformation fetishists.
|
|
#
?
Nov 13, 2018 05:05
|
|
|
big dyke energy posted:It's nice of them to make a tool to generate Animorphs cover art
|
|
#
?
Nov 13, 2018 06:38
|
|
|
Finally I can generate random guillotines for inspiration! 
|
|
#
?
Nov 13, 2018 07:00
|
|
|
  Also I ran Toilet Plunger to Toilet Plunger with min and max noise, and I see no plungers, but I do think I see a tit in the top left and a wiener in the bottom right 
|
|
#
?
Nov 13, 2018 16:38
|
|
|
quote:
https://twitter.com/JanelleCShane/status/1062507503152377856
|
|
#
?
Nov 14, 2018 02:37
|
|
|
Twitterbot that tries to describe images https://twitter.com/picdescbot/status/1062918211698679808
|
|
#
?
Nov 15, 2018 06:23
|
|
|
im not owned! im not owned!!
|
|
#
?
Nov 15, 2018 14:42
|
|
|
End of Shoelace posted:im not owned! im not owned!!
|
|
#
?
Nov 15, 2018 19:40
|
|
|
Deep cut, dril tweets are art
|
|
#
?
Nov 15, 2018 20:19
|
|
|
End of Shoelace posted:im not owned! im not owned!! A bottle of hard liquor is the intermediate between corn and man
|
|
#
?
Nov 16, 2018 00:52
|
|
|
Tunicate posted:A bottle of hard liquor is the intermediate between corn and man
|
|
#
?
Nov 16, 2018 03:08
|
|
|
AKA Pseudonym posted:Twitterbot that tries to describe images Well, it's not wrong, it just omitted the fact that the people are below ground level.
|
|
#
?
Nov 16, 2018 06:07
|
|
|
|
| # ? May 24, 2024 20:05 |
|
|
Picdescbot has some fun quirks, like seeing invisible sheep in photos of empty fields, calling any building taller than it is wide a clock tower, and describing things that are not vehicles as "parked on the side of a building." My all-time favorite picdescbot post, and I think a classic example of bot logic absurdity, is "a group of baseball players playing a football game."
|
|
#
?
Nov 17, 2018 16:59
|
|
