"... CLIP is trained to recognise images using a mountain of raw data drawn from the internet, where people routinely upload images and identify them with captions. Given the text prompt ('Australia') it combs through a library of 400 million images to find visual elements that correspond with this term. The image results CLIP produces have been described as like a 'statistical average of the internet.' According to the image above, the elements that best correspond to 'Australia' are roads, a desert horizon, the ocean, and a few furry and scaly creatures. Once it has its image results, CLIP then feeds these to VQ-GAN, which has been trained to assemble and compose original images of its own. This happens mostly out of sight, but you can get some sense of the process in this video of CLIP+VQ-GAN making the image."
ABC News reports on the awesome computer wizardry that yields some atrocious looking art and a disgusting hint of the junkpile of cliché and obviousness that's out there in the outback of the internet. Here's that video of the AI in action:
Posts
3 comments:
Amadeus 48 writes:
"Hi Althouse—that is a brilliant juxtaposition in your blog today. First, you describe the decline of Wikipedia into partisan propaganda through internet participation—one big global community, right?—and then you can see it visualized by the creation of completely nondescript images from the internet amalgamated by algorithms.
"I suppose this is regression to the mean in action, but what it really shows is that any activity is distorted by those who show up to participate in it, and those who persevere can ruin anything."
Temujin writes:
"I don't know if I'll be here in another 20 years, and even if I am, whether or not I know that I am here. But I hope I am and I hope that I have full cognition because the world in 20 years is going to look nothing like the world we have in front of us today. Think of the world some of us grew up in the 1950s or 1960s. How much different are things now? What kind of cars, appliances, electronics did we have then vs. today? There was actually once a world with no computers for the masses, no internet, black & white TVs with small screens that had to warm up (vacuum tubes) before a picture would appear. Think of our parents' world and what they saw in their lifetime from the early 1900s to the end of the 20th century.
"The changes that occurred in the 20th century were astounding. But if we look at the exponential rate of growth of computer processing power, using Moore's Law or not, the changes coming in this century will speed beyond our imaginations. In less than 20 years, AI will be in every aspect of our lives. What I don't know is if we'll reach the Singularity before, or after I'm gone. But I do think it's like an object that is up ahead coming at you, but you have no realization how fast it is coming. Until it's right about to hit you."
Scot writes:
I am reminded of Ken Thompson's Turing Award address "Reflections on Trusting Trust". He explains how a programmer can write code without intending to insert a Trojan horse, but if the compiler itself is compromised, he will not know that the resulting program does have a Trojan horse. Thompson concludes "The moral is obvious. You can't trust code that you did not totally create yourself."
In the same way, a compromised reference would infect thinking or writing and produce incorrect conclusions. But how do you know the reference is compromised? Who can confirm every single thing?
Consider the law. Do you *know* that every opinion in your law books has been faithfully copied?
The devil can cite Scripture for his purpose.
An evil soul producing holy witness
Is like a villain with a smiling cheek,
A goodly apple rotten at the heart.
O, what a goodly outside falsehood hath!
~The Merchant of Venice
Post a Comment