I have a bunch of spare s3 compatable storage from a cloud provider which I would really like to use to run Immich, but everytime i look into it it seems not possible.
Personally I think that even if the training data, is made of images of people who are over 18, the state of the adult industry is that there is a lot of harm and exploitation involved in that even. Do we have the evidence necessary to prove every over 18 person who's images are in the training data are okay with that use? So there is no somehow-okay anything. But if were dealing with models that can combine concepts .. then "adult" + "sex" = "sexualized adult" so "child" + "sex" = "sexualized child" .. this is just a fundamental capability of the technology. Without even getting into nudity where its easy to see how non CSAM, non-sexualize medical images etc, can fill in the gaps for the model.
This is why I think just calling it CSAM is confusing the arguments. I see the base harm as being the promotion of sexualization of children. And this is more a harm and risk to society and to people consuming these images. Of course if you extend this to "real childs image" + "sex" to get a sexualized deepfake then there is a kind of harm to that individual too, but its still a disctint thing from physical abuse..
This is just my instincts on this but I'd be really interested to see some real discussion between people with actual extensive understand and experience of all these areas.
I think the terminology here can confuse the issue a bit. And because its such a socially pernicious and stigmatized topic, it's hard to even talk about the phrasing without without raising doubts as to why you would get caught up in the weeds on such an issue. But, I would say there is use in making the distinction between something that is CSAM, where its a record of abuse that has happened to a real child, and sexualized depictions of children, or content promoting the sexualization of children. The social and personal harms are distinct, and if we are to firmly understand the arguments for/against guardrails on generative AI, then it's a distinction that needs to be made I think.
Most countries laws about health are based on freedoms and entitlements not obligations. So most people are free to choose to do things that are bad for their health.
Did you have to pretend a terminal was broken and call the domain admin or teacher over to log into it to get their credentials cached locally in that file in the first place?
In an age of affordable spellcheck, there was something heraldic about his lack of it.