On the Making of Nonexistent Cats
Indeed. Life goes on. 😼
Not sure about GAN... I did watch almost half of the video without conceding whether I understood any of what I saw/heard.
A simple, very non-rigorous explanation:
You know the general class of nifty AI technologies that make reverse image search, facial recognition, the visual cortex for self-driving cars, etc., etc.?
Well, it turns out that if you put two different AI bots in a feedback loop, you can make one bot train the other to produce the images that the first bot has already been trained to recognize.
So: Train one bot to recognize cat photos. Make the other bot start producing more or less random photos; and force it to learn to fool the cat photo recognizer into believing that a nonexistent cat is a real cat.
When you run this little bot vs. bot game on a big pile of expensive GPUs for a week or two, the cat-recognizer bot teaches the nonexistent-cat bot to produce fake cat photos that can fool humans. 😸
It is such a powerful technology, it can even produce KYC dox photos of nullius!
Good one. Here are some images that I posted on Lauda’s goodbye thread; curiouscats humans should right-click and view the image on the left to view it full size, or even click through to the paper hereby linked on Arxiv:
I am curious* about which subjects fascinate you. Though I will understand if you don’t want to be too specific. Perhaps I may be too toxic. (* Sufficiently curious that I will spare you the “taken under advisement” macro about my alleged “bloviating”. ;-)
but surely life happens
Indeed. Life goes on. 😼
This nonexistent cat admired strength and despised weakness:
Free images of Lauda!
Protip: If nullius drops a Youtube link, it is probably not stupid.
https://www.youtube.com/watch?v=dCKbRCUyop8
Free images of Lauda!
Protip: If nullius drops a Youtube link, it is probably not stupid.
https://www.youtube.com/watch?v=dCKbRCUyop8
Not sure about GAN... I did watch almost half of the video without conceding whether I understood any of what I saw/heard.
A simple, very non-rigorous explanation:
You know the general class of nifty AI technologies that make reverse image search, facial recognition, the visual cortex for self-driving cars, etc., etc.?
Well, it turns out that if you put two different AI bots in a feedback loop, you can make one bot train the other to produce the images that the first bot has already been trained to recognize.
So: Train one bot to recognize cat photos. Make the other bot start producing more or less random photos; and force it to learn to fool the cat photo recognizer into believing that a nonexistent cat is a real cat.
When you run this little bot vs. bot game on a big pile of expensive GPUs for a week or two, the cat-recognizer bot teaches the nonexistent-cat bot to produce fake cat photos that can fool humans. 😸
It is such a powerful technology, it can even produce KYC dox photos of nullius!
realised or generative cats only require a warm place to nap... shrug_smile.py
“What I cannot create, I do not understand.” — Richard P. Feynman
“What I cannot create, I do not understand.” — Richard P. Feynman
Good one. Here are some images that I posted on Lauda’s goodbye thread; curious
Kitty loved science!
—And technology indistinguishable from magic!
—And technology indistinguishable from magic!
 |  |
* nullius loves the “generative model” that “produces cat-shaped objects more often”. 😼
Science and maths and technical intrigues are posted to the Wall Observer in memory of Kitty.
Nulli...on a more personal note I must admit some fascination with some of your posts...but by the gods man..bloviate much?
I am curious* about which subjects fascinate you. Though I will understand if you don’t want to be too specific. Perhaps I may be too toxic. (* Sufficiently curious that I will spare you the “taken under advisement” macro about my alleged “bloviating”. ;-)