I don’t understand why people dump such personal information into AI chats. None of it is protected. If they use chats for training data then it’s not impossible that at some point the AI might tell someone enough to be identifiable or the AI could be manipulated into dumping its training data.
I’ve overshared more than I should but I always keep in mind to remember that there’s always a risk of chats getting leaked.
There’s evidence that a lot of suicide hotlines can be just as bad. You hear awful stories all the time of overwhelmed or fed up operators taking it out on the caller. There’s some real evil people out there. And not everyone has access to a dedicated therapist who wants to help.
More so from corporate proprietary ones no? At least I hope that’s the only cases. The open source ones suggest really useful ways proprietary do not. Now I dont rely on open source AI but they are definitely better
The corporate models are actually much better at it due to having heavy filtering built in. The fact that a model generally encourages self arm is just a lie that you can prove right now by pretending to be suicidal on ChatGPT. You will see it will adamantly push you to seek help.
The filters and safety nets can be bypassed no matter how hard you make them, and it is the reason why we got some unfortunate news.
Real therapy isn’t always better. At least there you can get drugs. But neither are a guarantee to make life better—and for a lot of them, life isn’t going to get better anyway.
AI “therapy” can be very effective without the gaming, but the problem is most people want it to tell them what they want to hear. Real therapy is not “fun” because a therapist will challenge you on your bullshit and not let you shape the conversation.
I find it does a pretty good job with pro and con lists, listing out several options, and taking situations and reframing them. I have found it very useful, but I have learned not to manipulate it or its advice just becomes me convincing myself of a thing.
I agree, and to the comment above you, it’s not because it’s guaranteed to reduce symptoms. There are many ways that talking with another person is good for us.
Depends on how you do it. If you’re using a 3rd party service then the LLM provider might not know (but the 3rd party might, depends on ToS and the retention period + security measures).
Ofc we can all agree certain details shouldn’t be shared at all. There’s a difference between talking about your resume and leaking your email there and suicide stuff where you share the info that makes you really vulnerable
I’m on the “forward to a professional and don’t entertain side” but also “use at your own risk” camp. Doesn’t require monitoring, just some basic checks to not entertain these types of chats
A reminder that these chats are being monitored
I don’t understand why people dump such personal information into AI chats. None of it is protected. If they use chats for training data then it’s not impossible that at some point the AI might tell someone enough to be identifiable or the AI could be manipulated into dumping its training data.
I’ve overshared more than I should but I always keep in mind to remember that there’s always a risk of chats getting leaked.
Anything stored online can get leaked.
Still, what are they gonna do to a million suicidal people besides ignore them entirely
Well, AI therapy is more likely to harm their mental health, up to encouraging suicide (as certain cases have already shown).
There’s evidence that a lot of suicide hotlines can be just as bad. You hear awful stories all the time of overwhelmed or fed up operators taking it out on the caller. There’s some real evil people out there. And not everyone has access to a dedicated therapist who wants to help.
Suicide is big business. There’s infrastructure readily available to reap financial rewards from the activity, atleast in the US.
More so from corporate proprietary ones no? At least I hope that’s the only cases. The open source ones suggest really useful ways proprietary do not. Now I dont rely on open source AI but they are definitely better
The corporate models are actually much better at it due to having heavy filtering built in. The fact that a model generally encourages self arm is just a lie that you can prove right now by pretending to be suicidal on ChatGPT. You will see it will adamantly push you to seek help.
The filters and safety nets can be bypassed no matter how hard you make them, and it is the reason why we got some unfortunate news.
Real therapy isn’t always better. At least there you can get drugs. But neither are a guarantee to make life better—and for a lot of them, life isn’t going to get better anyway.
Are you comparing a professional to a text generator?
Have you ever had ineffective professional therapy?
Are you still trying to compare medical treatment with generating text?
Compare, as in equal? No. You can’t “game” a person (usually) like you can game an AI.
Now, answer my question
No, comparing as in comparing apples and oranges.
Real therapy is definitely better than an AI. That said, AIs will never encourage self harm without significant gaming.
AI “therapy” can be very effective without the gaming, but the problem is most people want it to tell them what they want to hear. Real therapy is not “fun” because a therapist will challenge you on your bullshit and not let you shape the conversation.
I find it does a pretty good job with pro and con lists, listing out several options, and taking situations and reframing them. I have found it very useful, but I have learned not to manipulate it or its advice just becomes me convincing myself of a thing.
I agree, and to the comment above you, it’s not because it’s guaranteed to reduce symptoms. There are many ways that talking with another person is good for us.
The keyword here is “person”.
Strap explosives to their chests and send them to thier competitors?
Convince each one that they alone are the chosen one to assassinate grok and that this mission is all that matters to give their lives meaning.
Take that Grok!!
Advertise drugs to them perhaps, or somd sort of taking advantage. If this sort of data is the hands of an ad network that is
No, no. They want repeat customers!
Unless they sell Lifetime deals. Probably cheap on the warranty/support side. If the drug doesn’t work 🤔
It’s never the drugs I want though :(
My pet theory: Radicalize the disenfranchised to incite domestic terrorism and further OpenAI’s political goals.
What are their political goals?
Tax breaks for tech bros
I think total control over the country might be the goal, and it’s a bit more than a tax break.
Absolutely blows my mind that people attach their real life identity to these things.
Depends on how you do it. If you’re using a 3rd party service then the LLM provider might not know (but the 3rd party might, depends on ToS and the retention period + security measures).
Ofc we can all agree certain details shouldn’t be shared at all. There’s a difference between talking about your resume and leaking your email there and suicide stuff where you share the info that makes you really vulnerable
But they tell you that idea you had is great and worth pursuing!
deleted by creator
But imagine the chances for your own business! Absolutely no one will steal your ideas before you can monetize them.
You have to decide, a few months ago everyone was blaming OpenAI for not doing anything
I’m on the “forward to a professional and don’t entertain side” but also “use at your own risk” camp. Doesn’t require monitoring, just some basic checks to not entertain these types of chats
Definitely a case where you can’t resolve conflicting interests to everyone’s satisfaction.