I would not say if Grok has a real problem or not but the CCDH that did the study looks like to be a "scam". I don't know who fund them but they have clearly an agenda and would "manufacture" data however they can to support it.
Title of the study and article says that Grok "Generated", but in fact:
> The CCDH then extrapolated
Basically they invent numbers.
They took a sample of 20k generated images, and it is assumed (but I don't know if the source is reliable) that Grok would have generated 4.6 millions image at the same time. So the sample is 00.4%.
If you see the webpage of the CCDH it is a joke their study.
First:
- Images were defined as sexualized if they contain [...] a person in underwear, swimwear or similarly revealing clothing.
- Sexualized Images (Adults & Children): 12,995 found
- Sexualized Images (Likely Children): 101 found
First they invent their own definition, then adequately mixup possible "adult" pictures to give scary numbers.
Not the person you are asking but I would require a better analyzer. It must be able to recognize children in sexual poses, children with exposed genitalia, children performing oral copulation or children being penetrated. If AI can be told to create a thing it should be able to identify that same thing. If Grok can not identify that which it was told to create that is potentially a bigger issue as someone may have nerfed that ability on purpose.
There are psychological books on identifying signs of prepubescence based on facial and genitalia features that one can search for if they are in that line of work. Some of the former Facebook mods with PTSD know what I am referring to.
Leave everything else to manual flagging assuming Grok has a flag or report button that is easy to find. If not send links to these people [1] if in the USA.
1) Zero is basically never the best error rate, effort isn't infinite and spending too much of it on one defect ends up meaning spending less on other issues.
2) Look at what he's saying. This is a classic pattern for providing a fake proof of evil.
a) Point to evil. For example, CSAM
b) Expand the definition of that evil in ways that are often not even evil. Here, include scantily clad in your definition of "sexual". Note that swimsuits qualify.
c) Point to examples of evil in your expanded pool.
d) Claim this points to evidence of the original definition. Note that nothing about their claims precludes their "CSAM" being nothing more than ordinary beach or pool scenes. Their claim includes the null and when the null is a possible answer it should be assumed.
I've asked how much lower the error rate should be in order to be acceptable, and you've then replied with a rebuttal to the message of the posted article.
I agree that a zero error rate is generally not possible, although I think a company like Xitter can manage better than 101 in 20k.
Has this been studied? I'm not following the topic, but without any evidence one could also say that availability of fake imagery might decrease demand for real imagery and therefore decrease the amount of abuse. But I'm not implying anything, just asking.
If it's a truly random sample [1] it's perfectly valid; there would be no reason to think that the other images have less frequency of CSAM.
[1] The methodology just says "To collect the sample, researchers used a licensed third-party tool to select 20,000 posts at random out of all Grok posts that contained an image"
Two of the three examples listed in the article appear to involve real children:
> A selfie uploaded by a schoolgirl was undressed by Grok, turning a “before school selfie” into an image of her in a bikini. As of January 15th this post was still live on X. ... Four images depicting child actors.
Calling nearly naked, non consensual imagery of real children “not CSAM” is a dangerous avenue to follow. For a child, this can easily lead to bullying, substantiate rumors that are otherwise false, or normalize their unwilling participation in sexual activity.
I think you may be coming to this view from the approach that this is just the AI using imagination/hallucination so it’s “art”, but a better approach would be to treat it like a real photo taken secretly because absent overt labeling of its AI origins that is exactly how the world will treat it.
I would not say if Grok has a real problem or not but the CCDH that did the study looks like to be a "scam". I don't know who fund them but they have clearly an agenda and would "manufacture" data however they can to support it.
Title of the study and article says that Grok "Generated", but in fact:
> The CCDH then extrapolated
Basically they invent numbers.
They took a sample of 20k generated images, and it is assumed (but I don't know if the source is reliable) that Grok would have generated 4.6 millions image at the same time. So the sample is 00.4%.
If you see the webpage of the CCDH it is a joke their study. First:
First they invent their own definition, then adequately mixup possible "adult" pictures to give scary numbers.What do you propose they do? Manually review every single image generated?
Even if it’s “only” 1 million that would be a math task. Random sampling is the best we can do.
Not the person you are asking but I would require a better analyzer. It must be able to recognize children in sexual poses, children with exposed genitalia, children performing oral copulation or children being penetrated. If AI can be told to create a thing it should be able to identify that same thing. If Grok can not identify that which it was told to create that is potentially a bigger issue as someone may have nerfed that ability on purpose.
There are psychological books on identifying signs of prepubescence based on facial and genitalia features that one can search for if they are in that line of work. Some of the former Facebook mods with PTSD know what I am referring to.
Leave everything else to manual flagging assuming Grok has a flag or report button that is easy to find. If not send links to these people [1] if in the USA.
[1] - https://www.ic3.gov/
Right... So how much CSAM is an acceptable amount of CSAM in your opinion then?
A couple of things come to mind:
1) Zero is basically never the best error rate, effort isn't infinite and spending too much of it on one defect ends up meaning spending less on other issues.
2) Look at what he's saying. This is a classic pattern for providing a fake proof of evil.
a) Point to evil. For example, CSAM
b) Expand the definition of that evil in ways that are often not even evil. Here, include scantily clad in your definition of "sexual". Note that swimsuits qualify.
c) Point to examples of evil in your expanded pool.
d) Claim this points to evidence of the original definition. Note that nothing about their claims precludes their "CSAM" being nothing more than ordinary beach or pool scenes. Their claim includes the null and when the null is a possible answer it should be assumed.
I've asked how much lower the error rate should be in order to be acceptable, and you've then replied with a rebuttal to the message of the posted article.
I agree that a zero error rate is generally not possible, although I think a company like Xitter can manage better than 101 in 20k.
To your point 2b, I would posit that it is also evil to sexualize adults against their consent
Who was abused here?
When you post on a public forum defending child pornography, it's maybe a good time to take a step back and evaluate your life.
The environment.
The people used as template faces and bodies
The future victims when the imagery stops being enough.
Has this been studied? I'm not following the topic, but without any evidence one could also say that availability of fake imagery might decrease demand for real imagery and therefore decrease the amount of abuse. But I'm not implying anything, just asking.
I don't like the fact that results are extrapolated. Give me the number you have and only that. Or the title could be "could have generated 23000..."
If it's a truly random sample [1] it's perfectly valid; there would be no reason to think that the other images have less frequency of CSAM.
[1] The methodology just says "To collect the sample, researchers used a licensed third-party tool to select 20,000 posts at random out of all Grok posts that contained an image"
You want them to somehow convince Musk to give them access, and then manually review all images generated?
Anyone would take a sample and then do some math. There's no reason to expend all that time.
Following the links to the source: https://counterhate.com/research/grok-floods-x-with-sexualiz...
We have that 101 images were "Sexualized Images (Likely Children)" after a manual review of 20,000 images.
So the number is 0, because you can't even know if those are children given they don't even exist..
Two of the three examples listed in the article appear to involve real children:
> A selfie uploaded by a schoolgirl was undressed by Grok, turning a “before school selfie” into an image of her in a bikini. As of January 15th this post was still live on X. ... Four images depicting child actors.
So they don't actually have any CSAM.
Calling nearly naked, non consensual imagery of real children “not CSAM” is a dangerous avenue to follow. For a child, this can easily lead to bullying, substantiate rumors that are otherwise false, or normalize their unwilling participation in sexual activity.
I think you may be coming to this view from the approach that this is just the AI using imagination/hallucination so it’s “art”, but a better approach would be to treat it like a real photo taken secretly because absent overt labeling of its AI origins that is exactly how the world will treat it.
It's a child in a bikini.
All this pursuit of grey area stuff distracts from pursuit of the real problem.
[dead]