Many articles about “AI deepfake of Congressman Porkbelly fucking a Goat” are clearly fake …
… because they’d couch it as “his affair with a goat.” Even the New York Times couldn’t use proper language like, “goatfucker” but, I digress…
Perhaps AI will save us from having to deal with puzzles like this one, which has been bothering me for some time. I confess I first saw this nearly a decade ago, and commented about 5 years ago, but have never been satisfied:
@4:08 in this clip appears civil rights icon Rosa Parks, rocking out with Neil Young.
Right? Because in the mid-realm where rock’n’roll is real and the power of an amplifier can change your soul that is what happens in that liminal space. Other data points: that is the VIP seating. Also: hell, yeah.
We are going to need to save all of the contemporaneous footage we can keep track of, to prevent AI-generated “retcons” of existing stuff. For example, my humble offering:
Winston Churchill in the mosh pit at a joy division concert, 1992 (stable diffusion Juggernaut model)
I actually had to add “surrounded by punks and goths” to the prompt because the first iteration was surreal:
You can tell that’s fake because he hasn’t got a bloody drink in his hand. The fact that the AI used the faces of a lot of the Kennedy clan as the background is decidedly unsettling.
I have noted this elsewhere but will reprise the theme: the availability of photoshop and 3D rendering will make “UFO pictures” obsolete. It’s not that UFO nuts or Churchill conspiracy theorists will get less stupid, it’s that our visual filtering will improve. It will have to. By the way, it has always been good, at the high end. In his book Photo Fakery Dino Brugioni (I have had this book a long time and it is priceless) describes a film grain-level analysis of the negative of the famous “oswald with newspaper” photo [amzn] spoiler: not fake. The evolution of fake detection will be an ongoing issue. Were I Hollywood persona I would have a prominent tattoo added to someplace on my unmentionables so that if someone claimed to have photos of them, I would offer to trade them a bullet for a picture, but only if the picture were accurate in content and location.
Meanwhile, here’s Neil Young on the grassy knoll:
It’s an interesting artifact of the model not ‘knowing’ things that while it can produce an image with an appearance of depth-of-field blurring effects, the fact that the vast majority of faces in its training sets would have been sharply in focus causes it to render them in crisp detail, even while it blurs their necks and bodies.
Still fun to play with, but I suspect that the degree of expert post-processing required to render one of these fakes convincing to an expert might make it just as difficult as doing it the old fashioned way (photshop), or the old-old fashioned way (cutting up negatives etc). After all, that expert would not only need to correct the various other hallucinations present in all of those images but would also need to create consistent focal blurring and even bokeh effects if the lighting is appropriate.
In the case of your grassy knoll, the expert/faker/artist would also have to correct the unsightly seam where it appears the model thinks a rear set of doors is beginning to emerge from that black coupe-turning-sedan. I particularly like the appearance of the door handle budding there in the chrome molding. I feel like if this was one of those disturbing ai-hallucination videos the car would, within about 2 seconds, have a second fully grown set of doors. What the nightmare horror in the driver’s seat would morph into, I’m less sure.
I think it’ll ultimately be easier to train an AI to detect fakes than to use it to create them. Just as AI will probably get fairly good at working at consistent tasks within defined problems sets like detecting anomalies on scans and stained path samples. But such mundane and possible applications aren’t suffifcently world-changing to fulfill the needs of the VCs and hype-mongers, so we’re going to have to live with another few years of “AI is going to fix/ruin/fundamentally change the world” until they find some other next-great-thing and leave it alone for the actual experts to find a real use for.
Which Idiot was it, way back in The Before Time, said, something like,
The Camera Does Not Lie
?
This raises questions. Such as: goatfucker: one word or two?
Other words which are not used enough, based on a search of Google News:
Gormless – a few times per year. The royal family seems to be leading in references.
Gobshite – also just a few per year. Makes this list of A countdown of 10 of the best Irish insults at the Irish Post.
And the specific combination “gormless gobshite” pulls zero references. Clearly that must change if our civilization is to survive.
Is it just me, or are the guy in olive drab and the guy in black on either side of punk Churchill the same guy or twins? And are there at least 3 clones of the kid directly behind him in black w/ a white under-tee?
I’m pretty sure it was in one or another edition of the Whole Earth Review/CoEvolution Quarterly that there was an article on the then-new Photoshop, which included a photo of a flying saucer partly obscured by a wrought-iron balcony railing, and used the aphorism “This is the end of photography as evidence for anything.”
@4:
That’s punks[*] for you. They all look the same. Unlike goths who are all beautiful unique individuals.
* Or Mancunians. Take your pick there.
Actually, I think I may have had a severe comprehension fail on this post. At the JD/Churchill point it kind of gets local and personal and I’m kind of tickled and amused by the whole thing. But above that I guess I just dont get whether there was an important point I am missing or not…
So this whole Rosa Parks/Neil Young thing: were those real things or not? Are you suggesting they are AI or not? Is this believable or not? Should I (UK) find any of this surprising or is this a local cultural thing that is actually more interesting than it looks from over here?