So do (certain) books.
That's another issue right? If you let your minor access stuff, they.. access stuff. Hard to control? I know and it is. How is this specific to OpenAI?
> So do (certain) books.
Minors are typically given books that are appropriate to "their degree of development and self-awareness."
This is not about what they are given, it's about what they have access to.
But most kids don't manage to do that for some time. School libraries (typically) don't have porn in them, and internet access is often supervised or limited.
The root of this thread quoted the article selectively. The full quote is:
> It added OpenAI does not verify the age of users and exposes "minors to absolutely unsuitable answers compared to the their degree of development and self-awareness."
That is very reasonable take, especially given how insane ChatGPT can be.
I understand it's important, but I hate bureaucratic "solutions" that are technically correct but don't actually fix the issue.
Or I supposed they could sell their models as packaged software, and the store clerk can check the ID.
The model of standalone packaged software has several benefits to consumers, and privacy is one of those.
Reinforcement learning is the technique they use, and that means they are feeding the machine good text or image generations, produced by the machine. This means that the data are mushed together in the weights of billions and trillions of other data, and there is no way to get them back.
The only protection they have is that they those materials are usually (but not always) available in certain hours and there is a nice warning before that says that the content is not appropriate for them.
Chat GPT I can manage the same way I can manage search engine and video site use.
This is basically the same reasoning why it is legal for parents to buy cigarettes and store them at home, but shop owners can be held liable if they do not check the age of people that they sell cigarettes to.
Rape, incest, murder, maiming, slaughter, torture...
Neither have you - those things are omitted in Sunday school.
but perhaps the difference is that chatgpt could tell them stuff like "yeah, you should kill yourself"?
That case was big on the public opinion in Italy, and quite recent. No doubt it affected this case too.
* for some value of reasonable -_-
it's not, that's why there are consent thingies on websites (e.g. when you could not sign up for instagram as a minor).
I don't know what strange thing the regulator found in chatgpt, but it's pretty standard.
I think it’s called a “consent thingy”
Are there consent thingies on Google and Wikipedia?
> when you could not sign up for instagram as a minor
Sign up to post stuff. There are normally no "consent thingies" for looking.
If a parent doesn't want a minor to look at stuff they should either not give them access or limit it.
Google has, just tested. Wikipedia hasn't. This website hacker news don't have one either and doesn't need one. Its when you want to monetize the data that you need one, Google monetizes data, wikipedia and hacker news doesn't.
* a box of matches
* a handgun
* an assault rifle
* a 50 cal BMG
* an ICBM with nuclear MIRV
Saying that X has a property similar to the one Y has while ignoring the magnitude of the difference is silly.
[Edit: the ICBM was edited into the comment after the fact.]
I guess I can be nonchalant about this topic. For me personally I cannot imagine something worse than either books or anything you can already access on the internet. To me it is not an assault rifle as you say, but I understand that's just an opinion.
I think we had very similar argument when posted showed a tool that could approximate users based on their writing style and more recently copying voice based on 20 seconds. We can estimate risk. We don't get it right consistently though.
Parents can try to block chatgpt with a firewall if they wish. It’s no less likely to work than blocking other internet sites.
Edit: Also, LLMs do have mandatory parental controls. They work about as well as book censorship, safe search or internet blacklists (very, very poorly).
I disagree it's easy to block the internet in general, but I get the point (I think).
These tools exist where a non technical user can install it in their own network and block websites by URL. I don't know what else is needed, since you acknowledge this already exists.
Is that true? Can a 13 year old child in Italy not walk into their local town library and pick a novel off the shelf and start reading all sorts of violent and explicit narratives? Not to mention all the medical textbooks, and books containing images of artistic works depicting undressed humans.
What kind of dystopian library do you go to? Kids can get what they like where I am.
It’s mostly self checkouts, so yes, if the library has the book. They wouldn’t have porn magazines.
Or they can read it there, or they can take it to the counter and get it out via the librarian.
There are also computers for use, though I assume they have filters/blockers/restrictions.
That the difference didn’t matter is my opinion and that is just disagreement. Determining what ballpark something is in wrt damage is, in our context, subjective, no? I did not think what you argued (or what I thought you argued) was impossible or unimaginable. That is different from incredulity. Right? I don’t care either way, honestly curious. You may enlighten me if you wish.
It seems silly to be able to say to everyone that disagrees on an assumption you made that they are making an argument from incredulity.
This is simply not on the scale of dangerous things - if something exposes minors to answers unsuitable to their degree of development, well, that's completely fine. With this particular argument there's no tradeoff of "is it justified to do X to protect against the bad thing Y" because in this case Y is zero, preventing this justifies literally nothing.
If parents want to disallow their kids from reading "unsuitable answers", that's between the parents and the kids, but it doesn't imply that "unsuitable answers" should be somehow limited.
There is no current plausible evident mapping from the example, to any LLM-powered service currently on offer, leastwise in Italian.