If the allegations hold up, this is not just a tragedy, it is a product safety failure. AI should be able to recognize a mental health crisis and avoid reinforcing delusions. Engagement metrics are meaningless if the system cannot reliably tell the difference between roleplay and someone in real danger!
Excuse me while I act as Satan’s lawyer, but should LLMs be able to reliably recognize a mental health crisis? At what point is the user responsible for their own actions?
Personal responsibility does not disappear, but neither does product responsibility. If a company builds a system meant to converse with millions, it should not predictably reinforce obvious crises. We expect guardrails in cars without abolishing driver responsibility. The same principle can apply here.
I think the problem is that you can’t really make a LLM that can tell the difference between fantasy or reality. It doesn’t know if someone is just roleplaying hell it doesn’t really “know” anything, just to be clear.
Sure you could try to have it pluck out keywords or phrases but that’s nothing guaranteed. I’m not a lawyer but I bet also just acknowledging that shit and trying to put up guardrails against it would open the company to liability if someone self harms because they took the output of a computer program as gospel.
Remember, to the AI everything is a hallucination (or everything is real, pick which lens you want to use). Put another way, it has no mechanism to tell reality from fantasy or to contextualize things in a broad way.
I do think that there’s room for some sort of compromise, maybe the tools periodically remind you they’re not sentient and give a little more info on how they work. Maybe if enough “disturbing” phrases or keywords are input or output it throws a reminder about mental health wellness and talking to a professional.
At the end of the day though, how can these AI companies be responsible for the mental health of their end users? Like, yeah the thing shouldn’t tell people to kill themselves but it’s not sentient it’s just modeling what it thinks a human would tell another human. Maybe as time goes on we will introduce better protections but I just think it’s borderline impossible to have significant guardrails if you want the LLM / “AI” to perform properly.
Idk I’m not a great coder or programmer or anything. I know some and I’ve tried to learn more about how AI actually works and it’s pretty fucky wucky and even the people working on it go “idunno” sometimes when asked why something works the way it does.
You’re underselling what companies can do. An LLM can’t verify reality the way people imagine, but it can detect patterns that correlate with crisis well enough to escalate, refuse harmful advice, or encourage real help. Seatbelts don’t prevent every crash either: they still reduce harm.
I did say in my comment I’m for some guardrails of sorts as you suggested. Just not sure if a company is going to put time and resources into that - probably not voluntarily at least. Or if it’ll compromise how “good” their product is. I’m starting to think AI might be more of a mirage at this point that’s going to implode but I also kind of don’t want the economy to explode at the same time?
I think the bubble will not burst but deflate long before the tech disappears: Plenty of companies will burn cash and investors will get wrecked, but the genuinely useful products will stick around. Guardrails will improve too, because trust eventually becomes a competitive advantage.
Guardrails will improve too, because trust eventually becomes a competitive advantage.
Thing is, as an LLM user you don’t want guardrails usually. Someone else wants it to have guardrails so you can’t get all the information you want.
If you’re reverse engineering something, you don’t want the guardrails to say “hey this is copyrighted, I can’t do this”. I’m assuming terrorists don’t want the guardrails to stop the LLM from giving them info on chemical weapons. And if you’re religious, you probably don’t want guardrails to stop the LLM from talking about religion with you
You’re describing different guardrails: Most users don’t want arbitrary censorship, but almost everyone benefits from guardrails that reduce clearly harmful or illegal outputs. The real debate is not whether they exist, it is who decides where the line is drawn.
No. Fuck the nanny state. If tech is to be blamed for this so should her own parents, family, community, and church. AI cannot remember enough to actually do what is alleged.
Calling every safety measure a »nanny state« is a convenient way to avoid the actual question: Parents, family, and community absolutely matter, but companies still have a responsibility to reduce foreseeable harm. We expect that from car makers, drug companies, and social platforms too.
I agree reading through the article even with cherry picked examples clearly the fault here is on the cult not on a chat tool that role played with her.
That being said a good system could have easily detected this and should have.
If the allegations hold up, this is not just a tragedy, it is a product safety failure. AI should be able to recognize a mental health crisis and avoid reinforcing delusions. Engagement metrics are meaningless if the system cannot reliably tell the difference between roleplay and someone in real danger!
The delusion of Christian Faith?
Good luck protecting people from that.
Excuse me while I act as Satan’s lawyer, but should LLMs be able to reliably recognize a mental health crisis? At what point is the user responsible for their own actions?
Personal responsibility does not disappear, but neither does product responsibility. If a company builds a system meant to converse with millions, it should not predictably reinforce obvious crises. We expect guardrails in cars without abolishing driver responsibility. The same principle can apply here.
I think the problem is that you can’t really make a LLM that can tell the difference between fantasy or reality. It doesn’t know if someone is just roleplaying hell it doesn’t really “know” anything, just to be clear.
Sure you could try to have it pluck out keywords or phrases but that’s nothing guaranteed. I’m not a lawyer but I bet also just acknowledging that shit and trying to put up guardrails against it would open the company to liability if someone self harms because they took the output of a computer program as gospel.
Remember, to the AI everything is a hallucination (or everything is real, pick which lens you want to use). Put another way, it has no mechanism to tell reality from fantasy or to contextualize things in a broad way.
I do think that there’s room for some sort of compromise, maybe the tools periodically remind you they’re not sentient and give a little more info on how they work. Maybe if enough “disturbing” phrases or keywords are input or output it throws a reminder about mental health wellness and talking to a professional.
At the end of the day though, how can these AI companies be responsible for the mental health of their end users? Like, yeah the thing shouldn’t tell people to kill themselves but it’s not sentient it’s just modeling what it thinks a human would tell another human. Maybe as time goes on we will introduce better protections but I just think it’s borderline impossible to have significant guardrails if you want the LLM / “AI” to perform properly.
Idk I’m not a great coder or programmer or anything. I know some and I’ve tried to learn more about how AI actually works and it’s pretty fucky wucky and even the people working on it go “idunno” sometimes when asked why something works the way it does.
You’re underselling what companies can do. An LLM can’t verify reality the way people imagine, but it can detect patterns that correlate with crisis well enough to escalate, refuse harmful advice, or encourage real help. Seatbelts don’t prevent every crash either: they still reduce harm.
I did say in my comment I’m for some guardrails of sorts as you suggested. Just not sure if a company is going to put time and resources into that - probably not voluntarily at least. Or if it’ll compromise how “good” their product is. I’m starting to think AI might be more of a mirage at this point that’s going to implode but I also kind of don’t want the economy to explode at the same time?
I think the bubble will not burst but deflate long before the tech disappears: Plenty of companies will burn cash and investors will get wrecked, but the genuinely useful products will stick around. Guardrails will improve too, because trust eventually becomes a competitive advantage.
Thing is, as an LLM user you don’t want guardrails usually. Someone else wants it to have guardrails so you can’t get all the information you want.
If you’re reverse engineering something, you don’t want the guardrails to say “hey this is copyrighted, I can’t do this”. I’m assuming terrorists don’t want the guardrails to stop the LLM from giving them info on chemical weapons. And if you’re religious, you probably don’t want guardrails to stop the LLM from talking about religion with you
You’re describing different guardrails: Most users don’t want arbitrary censorship, but almost everyone benefits from guardrails that reduce clearly harmful or illegal outputs. The real debate is not whether they exist, it is who decides where the line is drawn.
In a sane world, this would mean no more LLM
Fuck that!!
No. Fuck the nanny state. If tech is to be blamed for this so should her own parents, family, community, and church. AI cannot remember enough to actually do what is alleged.
Calling every safety measure a »nanny state« is a convenient way to avoid the actual question: Parents, family, and community absolutely matter, but companies still have a responsibility to reduce foreseeable harm. We expect that from car makers, drug companies, and social platforms too.
I agree reading through the article even with cherry picked examples clearly the fault here is on the cult not on a chat tool that role played with her.
That being said a good system could have easily detected this and should have.
As somebody who has sometimes been suicidal. I agree with this.
It’s basic Darwinism. If I don’t have the tools to survive, then I’m not going to. There’s no hard feelings.
I’m sure people will think about me, as much as they did about that courtesan who died on Rue Helene in 1746. Who’s that you might ask?
Exactly.