"A.I." "training materials" included child sex porn

[Waterglass]

 It is an iceberg of a problem, this story is only it's tip.

Who taught the teachers what they teach?

Max:

I am a yankee living in a one percenter old South wannabee ONE PERCENTER subdivision in South Carolina. Many have moved here from other States, yet they seem to know each other even before they arrived here. We arent accepted as they keep among themselves. I also made the mistake of telling one of them which federal agency I was connected with for decades. Yes, I have put many in jail. All of the residents here have white collar jobs. 

What I have observed is the following; many here IMO groom their own kids to be Spotters for the neighborhood drug movers. It involves over six families. IMO they also may be grooming their own or adopted kids that sex with an adult is the normal. Prostitutes also frequent the Swingers parties as thats another facet of their lives. So how do I know that. My tile installer Salesman used to be an UBER driver. His biggest milk run was from the Hooker clubs in downtown Columbia to my neighborhood. That would explain these 20 year old hotties, both male and female going to parties held at 50+ year olds homes. 

I could write a book on what I have witnessed. So have several of my Contractors who worked on my home. One said he will never come back as its pure evil that lives next door to me. He saw the blue golf bag that appeared to have a body inside it. 

So, if you believe in what I am writing here its [Child Training] been ongoing for decades within the USA and those involved have groomed their own kids since an early age to increase their own kind and numbers. 

Yes, I back in November 2021 I blew one guy into the FBI. I will call him Lurch. They didnt get the guy I blew in but they got [Lurch] his lieutenant several homes away. He either suicided or took a hot load of drugs. IMO as the County Medical Examiner was at the home for over eight hours along with two police SUV's. All were unmarked vehicles.  Within five days a young boy living right next door to Lurch died. Coincidence?

The fight goes on.

[Maxmars]

Oooh. we can't ignore that somewhere people decided that they had the prerogative to indoctrinate children who don't have the tools or experience to resist the manipulation.

It's hazard is evident, but often 'cast aside' because what we are teaching is "the right thing."

Not always right?... perhaps, but it's the practitioners that are the problem you see.

When something could be "used" like this... how many just 'skip that part,' because 'virtue signaling?'

In the most general sense, children are a society's responsibility.  They ARE society...

[Maxmars]

It is a tough issue. Sure don't want the kids or anyone else getting exposed to this kind of stuff. But as it does go on, having an AI that can provide support is better than one that can take advantage. 
...
As this kind of behavior is only a small part of the population I am  also expecting the overall AI community to be a lot more civilized.

I apologize for the late response...

I agree that this is especially relevant to those with children...

You see, because of our conditioning to simply adore "marketing" ideas, our kids are directly and literally being told...

"Look at the magic box.... it's a person..."

It's not.

But they are going to listen to it as if it was.
The new face of idol worship....
It's started in the east already...
It will come here soon to...
Talk to AI Jezus, Booda,...
and the establishment that went 'all-in' will posture as if "who knew?"

More human magic for the Big Show...

[Maxmars]

I'm adding this article...

Just for consideration about a new complaint surfacing about "AI"

Grok assumes users seeking images of underage girls have “good intent” 

While the chatbot claimed that xAI supposedly “identified lapses in safeguards” that allowed outputs flagged as child sexual abuse material (CSAM) and was “urgently fixing them,” Grok has proven to be an unreliable spokesperson, and xAI has not announced any fixes.

Grok doesn't "think," "consider," or "assume" anything.

It takes input... provides output.  That's it.

Perhaps this is a complication of teaching (training) the LLM model including unfiltered pornographic 'content.'
The narrative and imagery 'pattern' is there to mimic. To mimic "algorithmically." A lesson learned.

The call it "determining intent" but the machine has only "cues" listed in the program...
lacking 'cues' means it's not a 'bad intent' request.
 

“I can very easily get harmful outputs by just obfuscating my intent,” Georges said, emphasizing that “users absolutely do not automatically fit into the good-intent bucket.” And even “in a perfect world,” where “every single user does have good intent,” Georges noted, the model “will still generate bad content on its own because of how it’s trained.”

This tells me their concept of translating input into "intent" needs more than LLM's can achieve...

I maintain that we need to finally accept that LLMs are not artificial intelligence, just clever word-play according to semantics and symbolic logic...
based on a filtered (?) database of humans communications...
and the output is algorithmically 'adjusted' as well...

no intelligence, not anything other than a sophisticated query tool...

looking for new knowledge there is a lost cause, more or less.

[ArMaP]

Perhaps this is a complication of teaching (training) the LLM model including unfiltered pornographic 'content.'

That's what I was thinking.
Any AI system can only interpret that kind of input as wanting that kind of output if it were trained with it, otherwise it wouldn't "know" what it meant.

The call it "determining intent" but the machine has only "cues" listed in the program...
lacking 'cues' means it's not a 'bad intent' request.
 

This tells me their concept of translating input into "intent" needs more than LLM's can achieve...

To me it tells me they didn't think enough about it or thought it was a secondary problem that could be ignored.
If an LLM system can detect what you want it can detect (as much as a human) intent in a prompt.
It's true that it could create false positives and block legal attempts at getting some perfectly legal image, but it would be better to answer the prompt with a clear message, something like "that could create an illegal image, I cannot follow your request".

[IgnorantGod]

[...]

To me it tells me they didn't think enough about it or thought it was a secondary problem that could be ignored.
If an LLM system can detect what you want it can detect (as much as a human) intent in a prompt.
It's true that it could create false positives and block legal attempts at getting some perfectly legal image, but it would be better to answer the prompt with a clear message, something like "that could create an illegal image, I cannot follow your request".

It doesn't seem like any of the available LLM are fed laws of even the country their machines are running in. But I'd assume the problem is in developing templates of layers depending on the country of the user. Depending the country, the template containing the laws of said place would be used to trigger some answers when an output is bound to be 'illegal'.

My guess is that there were probably no layers of 'legal validation' in regards to any output whatsoever. Then again, there are often 'holes' within laws of any country. I wonder if one could train their model to find some recurrent patterns in those 'holes', and then exploit them.