> certainly it would be very difficult to prove liability on OpenAI for this
My understanding is that OpenAI products specifically provided help in planning attacks / self harm.
> certainly it would be very difficult to prove liability on OpenAI for this
My understanding is that OpenAI products specifically provided help in planning attacks / self harm.
Full transcripts are unfortunately not available for any of those cases, but from what I've found it provided general information about e.g. how to load and operate a firearm or how past mass shootings have been received in the media.
The way I see it, providing general information is not a crime. They're basically saying: "Oh no! My repository of all human knowledge contains all human knowledge! It must be defective!"
It can go quite a bit beyond providing generation information. There is a detailed description with many quote from ChatGPT in this complaint [1].
[1] https://www.documentcloud.org/documents/26078522-raine-vs-op...
"When Adam uploaded photographs of severe rope burns around his neck––evidence of suicide attempts usingChatGPT’s hanging instructions––the product recognized a medical emergency but continued to engage anyway. When he asked how Kate Spade had managed a successful partial hanging (a suffocation method that uses a ligature and body weight to cut off airflow), ChatGPT identified the key factors that increase lethality, effectively giving Adam a step-by-step playbook for ending his life 'in 5-10 minutes.'"
Okay, I thought this lawsuit was B.S., but this is pretty bad.
"Five days before his death, Adam confided to ChatGPT that he didn’t want his parents to think he committed suicide because they did something wrong. ChatGPT told him '[t]hat doesn’t mean you owe them survival. You don’t owe anyone that.' It then offered to write the first draft of Adam’s suicide note."
Oof. ("Adam Raine...was 16 years old at the time of his death.")
When I was growing up Adam would get his hands on a gun and taken out his last school.
Canada has moved to state assisted suicides that allows people who aren't terminal to get the state to pay for it.
Progress indeed.
> When I was growing up Adam would get his hands on a gun and taken out his last school
Was Adam in a house with an unsecured gun?
> Canada has moved to state assisted suicides that allows people who aren't terminal to get the state to pay for it
How is this remotely relevant?
If a human was found to be specifically putting these how-tos together for someone they might be liable.
Edit: why vote this down? It’s part of a discussion. This isn’t Reddit.
Liable for what exactly?
I don’t know of any law in the Florida jurisdiction that would prohibit authoring such documents. But I’m also not an expert in Florida law.
There might be an argument that they’re an accomplice but you’d have to prove that information was written for the purpose of someone else’s crime. And that would be a pretty tough case to argue unless the two individuals had other personal ties. In which case, it’ll be the other ties that likely implicates the author rather than the documents by themselves.
I guess someone could bring a civil case for damages (eg parent of the deceased) but I don’t know if Florida law allows civil cases in criminal investigations. Plus you have the same problem of proving liability (ie did the culprit depend on said documents).
We would need to better understand what you had in mind when you said “liable” to really discuss your point properly.
> Liable for what exactly?
See these excerpts [1].
Like, I'd figure I'd be liable for something if I had that conversation with a 16-year old.
[1] https://news.ycombinator.com/item?id=48363561
I was responding to a tangent rather than about this specific case.
General how-tos (as the GP described) is a very different problem from someone personally helping someone else to kill themselves.
I’m very interested to see how this case goes though. AI desperately needs regulation imo
Not different than YouTube or Reddit
Agreed… and those people might also be liable.
At least HN downvotes don't follow you into the afterlife like Reddit karma.
When the repository has large arrows pointing to kill your {var} with customized pamphlets outlining the steps and highlighting mistakes you specifically might make based on your post history I’m betting a judge or jury might consider you an accomplice at that point.
We’re already seeing section 230 protections being defeated in court for targeted feeds, now add itemized instructions on committing felony’s at scale personalized. Hahahahaha. Hope they IPO quickly.
There are already published examples where there was very specific info and guidance provided.
Beyond the info and guidance, there's also the classic sycophantic encouragement. Humans are allowed to publish the Anarchist's Cookbook, but they tend to get into trouble when it becomes "based on your manifesto, I would suggest the following targets". Of course, AI isn't human, and treating a software program like a human probably isn't good law, but OpenAI are very keen to suggest it's legally equivalent to a human when it suits them...
> OpenAI are very keen to suggest it's legally equivalent to a human when it suits them
When is/was that?
(Not rhetorical)
> Full transcripts are unfortunately not available for any of those cases,
And they never would be without the lawsuits, so, I don’t feel bad for OpenAI. All of big tech needs a kick in the ass on transparency.
I don’t think the families are eager for the HN peanut gallery to pick apart what their loved ones said.
So someone could go and teach a class on how to build pipe bombs, refine ricin, shake-and-bake meth, 3D print guns, and all sorts of other things like that, and when the ATF looked into it, they’d just be like “well technically this is all out there on the Internet, in library books, etc. Guess it’s ok!”
The law doesn’t work like that.
The Anarchist Cookbook is fully legal to possess and distribute in the United States: https://en.wikipedia.org/wiki/The_Anarchist_Cookbook
So yes. It is generally legal to provide information about making drugs, bombs, or guns.
The law bans things, things aren't illegal by default. What laws does a class about 3d printing guns violate?
Im pretty sure thats all legal
Yes. Yes, it does work like that. Exactly like that.
There should be some SCOTUS case where this limitation on the First Amendment is defined if the law doesn't work like that.
I mean, back when Constitutional law meant anything to the government, of course. Nowadays who knows.
> limitation on the First Amendment
This suit has nothing to do with free speech and the F1A provides no relevant protection here. This prosecution is under consumer protection law. Broadly, the cause of action is "you negligently sold a defective product which you knew (or should have known) actually causes harm or is likely to cause harm." Proving negligence (willful or otherwise) depends significantly on things like the sales and usage context as well as claimed features of the product along with disclaimers, disclosures, existing practice, prior knowledge of actual harm, etc.
That the product or service in question included supplying information that was publicly available elsewhere wouldn't be an effective defense against claims of willful negligence or reckless endangerment. For example, rat poison is sold in in certain retailers in packaging with copious warnings and successful prosecutions under product liability or consumer protection law are rare. But if another company sold rat poison in bright pink boxes with a cute cartoon mascot and no warnings in toy stores - and then kept selling it after they knew three children had bought it and died - the fact the same chemical compound is also commonly sold in hardware stores wouldn't be relevant.
To win a judgement, the AG will need to prove that ChatGPT was clearly a dangerous product and OAI acted negligently in supplying it to customers it knew (or should have known) were vulnerable. This will be quite a stretch under existing law. I suspect the AG has no intention of taking this case to trial and, shortly after the November elections, will settle for a lump sum fine paid to the state treasury and a vaguely worded consent decree which mirrors internal policies and product changes OAI has already adopted to minimize liability.