what have we become!
after that horror of such a barbaric inhuman attack, we also hide the reality and push for disinformation so this could happen again and again.
and these nightmares will happen in more and more places and it will be harder and harder to understand what is real and what is fake.
I wouldn't panic much yet. Trust, chains of evidence, and authentication aren't new mechanisms. These systems deliberately work independently of the quantity or quality of manufactured bullshit.
I don't mean to be dismissive. Recent technological breakthroughs will disrupt the status quo. Problems will arise; the (anti-)spam arms race will continue, etc. But hope is defensible. The sky isn't falling, and I think we'll be okay.
agreed. Photoshop has existed for quite a long time, and often just made earlier retouching techniques easier and faster, so it's not like this is new. Read Disclosure, by Michael Crichton, for example.
Being suspicious of media emerging during wartime is not "living in an alternate reality." Fuzzy images get passed around on Telegram, make their way to Twitter/X, and then 2-months later it's determined they're from an entirely different conflict despite the media and official outlets uncritically running with the narrative.
Are you talking about the time an alt-right group tricked a news outlet in to sharing images as a gotcha?
Of course the images coming from officials in Israel are going to be propaganda for their side, and if you are consuming those images, you need to keep in mind that of course officials from Israel are going to be pushing images that make the hamas look worse.
Does this mean that of course you should believe that the images are fake? I don’t think so. That’s not what “skepticism” means. Not yet anyway — ai images aren’t anywhere close to good enough for this.
When the Internet circulated unverified reports of a legendary pilot that was taking out ridiculous numbers of Russian aircraft…
Turned out it was 100% fabricated.
Remember Snake Island?
When we were told 90 Ukrainian forces heroically refused to surrender and they were slaughtered by Russian MIL…
Turned out Russia took all of them alive and they all surrendered willingly.
Remember when the media were using old photos and video from Syria and footage from video game simulators?
My point is, that war propaganda is real, and Western media have been been producing it in Ukraine and for decades before that.
As it pertains to Israel/Palestine; we don’t know what reports are real, what is being exaggerated, who is behind what. But history would suggest that everything the official narrative says now will end up being vastly different later on.
I wish we didn’t have to think like this, but the media have proven themselves to be not just inaccurate, but intentionally distorting and manipulating the truth to control public perception and advance their agenda, especially in the early stages of a new war when narratives are being established.
Needless to say, there is a lot of disinformation out there.
> They're black boxes that come up with an answer. And at their best they operate at 90 percent accuracy.
Well at least the truth is right there in the article. Every single “detector” that’s been going on is a heuristic and has false positives. OpenAI quietly walked back on theirs (for text). I mean, nobody should rely on these for students taking tests, let alone making judgments about war crimes.
Either typical Twitter noise amplification is at fault here, and AI tooling just happens to be involved, or we have a serious education problem where people think these detectors actually work for anything. I’m leaning both.
I'm not so sold on the specific example they use here. The tool specifically says it is a proof of concept and only predicted 52% chance of AI generated. Using those signals to conclude that something is "likely AI generated" is a bigger problem from the evaluator, not the tool.
While this is abundantly clear with ML, it is far more general: metrics are guides, not answers. Metrics are models, and like all models are wrong (wrong != useless && wrong != incorrect). And no model/metric perfectly aligns with intent. This needs to be something drilled into every single person's head. There is always more than needs to be done after a machine spits out a number, a lot more. If all we do is take a number at face value then we're going to fuck up a lot of stuff (which we're doing a lot of)
There are plenty of other examples of ML creating noise in the war environment, even around citizen interpretations of those images. For example, I've seen accounts whos profiles are of people that don't exist and I could confidently tell you the style generator that made them without an ML classifier (if you work in generation and haven't looked at millions of ML generated photos go to the top of this comment). (this is not saying I'm not a noisy detector, just I know naturally some key mistakes that get made that are non-realistic and these photos weren't cleaned up in post)
I’ll care more when I can visit a website that does not blast me with a popover requiring me to press something. This is getting out of hand. Some sites have up to 3 separate popovers that demand my attention.
16 comments
[ 0.85 ms ] story [ 521 ms ] threadI don't mean to be dismissive. Recent technological breakthroughs will disrupt the status quo. Problems will arise; the (anti-)spam arms race will continue, etc. But hope is defensible. The sky isn't falling, and I think we'll be okay.
A lot of people use “be skeptical” as a means to reject what’s right in front of them.
Skepticism here is quite rational.
Of course the images coming from officials in Israel are going to be propaganda for their side, and if you are consuming those images, you need to keep in mind that of course officials from Israel are going to be pushing images that make the hamas look worse.
Does this mean that of course you should believe that the images are fake? I don’t think so. That’s not what “skepticism” means. Not yet anyway — ai images aren’t anywhere close to good enough for this.
When the Internet circulated unverified reports of a legendary pilot that was taking out ridiculous numbers of Russian aircraft…
Turned out it was 100% fabricated.
Remember Snake Island?
When we were told 90 Ukrainian forces heroically refused to surrender and they were slaughtered by Russian MIL…
Turned out Russia took all of them alive and they all surrendered willingly.
Remember when the media were using old photos and video from Syria and footage from video game simulators?
My point is, that war propaganda is real, and Western media have been been producing it in Ukraine and for decades before that.
As it pertains to Israel/Palestine; we don’t know what reports are real, what is being exaggerated, who is behind what. But history would suggest that everything the official narrative says now will end up being vastly different later on.
I wish we didn’t have to think like this, but the media have proven themselves to be not just inaccurate, but intentionally distorting and manipulating the truth to control public perception and advance their agenda, especially in the early stages of a new war when narratives are being established.
Needless to say, there is a lot of disinformation out there.
Well at least the truth is right there in the article. Every single “detector” that’s been going on is a heuristic and has false positives. OpenAI quietly walked back on theirs (for text). I mean, nobody should rely on these for students taking tests, let alone making judgments about war crimes.
Either typical Twitter noise amplification is at fault here, and AI tooling just happens to be involved, or we have a serious education problem where people think these detectors actually work for anything. I’m leaning both.
While this is abundantly clear with ML, it is far more general: metrics are guides, not answers. Metrics are models, and like all models are wrong (wrong != useless && wrong != incorrect). And no model/metric perfectly aligns with intent. This needs to be something drilled into every single person's head. There is always more than needs to be done after a machine spits out a number, a lot more. If all we do is take a number at face value then we're going to fuck up a lot of stuff (which we're doing a lot of)
There are plenty of other examples of ML creating noise in the war environment, even around citizen interpretations of those images. For example, I've seen accounts whos profiles are of people that don't exist and I could confidently tell you the style generator that made them without an ML classifier (if you work in generation and haven't looked at millions of ML generated photos go to the top of this comment). (this is not saying I'm not a noisy detector, just I know naturally some key mistakes that get made that are non-realistic and these photos weren't cleaned up in post)
AI is good but not that good (yet). It’s clearly an actual photo, but it’s not clear what exactly the image is of.
It is an image of children: https://www.reuters.com/world/nato-ministers-shown-horrific-...
The claim is not that the whole photo is AI generated. The claim is they used AI to remove a living puppy and then edited in the burned thing.
https://files.catbox.moe/e3uba2.jpg
I have not inspected the pixels on this. I do not know which is the original image, or if that can be determined from the small size images we have.
I am pointing out the story denies the wrong thing, so it is using a straw man argument.