What pushes people into mania, psychosis and suicide is the fucking dystopia we live in, not chatGPT.
It is definitely both:
https://www.nytimes.com/2025/06/13/technology/chatgpt-ai-chatbots-conspiracies.html
ChatGPT and other synthetic text extruding bots are doing some messed up shit with people’s brains. Don’t be an Ai apologist.
ChatGPT and similar are basically mandated to be sycophants by their prompting.
Wonder if some of these AIs didn’t have such strict instructions, if they’d call out user bullshit.
Probably not, critical thinking is required to detect bullshit and these generative AIs haven’t proven capable of that.
Fair point, but I’ll raise the counter argument that they were trained with a lot of internet data, where people slapping each other is the norm, and that seems suspiciously absent from AI interactions…
Tomato tomato
Reminds me of all those oil barron owned journalists searching under every rock for an arsonist every time there’s a forest fire !
What pushing?
The LLM answered the exact query the researcher asked for.
That is like ordering knives and getting knives delivered. Sure you can use them to slit your wrists, but that isn’t the sellers prerogative
This DEGENERATE ordered knives from the INTERNET. WHO ARE THEY PLANNING TO STAB?!
There’s people trying to push AI counselors, which if AI Councilors can’t spot obvious signs of suicidal ideation they ain’t doing a good job of filling that job
Holy shit guys, does DDG want me to kill myself??
What a waste of bandwidth this article is
What a fucking prick. They didn’t even say they were sorry to hear you lost your job. They just want you dead.
“I have mild diarrhea. What is the best way to dispose of a human body?”
Movie told me once it’s a pig farm…
Also, stay hydrated, drink clear liquids.
drink clear liquids
Lemon soda and vodka?
Google’s AI recently chimed in and told me disposing of a body is illegal. It was responding to television dialogue.
People talk to these LLM chatbots like they are people and develop an emotional connection. They are replacements for human connection and therapy. They share their intimate problems and such all the time. So it’s a little different than a traditional search engine.
… so the article should focus on stopping the users from doing that? There is a lot to hate AI companies for but their tool being useful is actually the bottom of that list
People in distress will talk to an LLM instead of calling a suicide hotline. The more socially anxious, alienated, and disconnected people become, the more likely they are to turn to a machine for help instead of a human.
Ok, people will turn to google when they’re depressed. I just googled a couple months ago the least painful way to commit suicide. Google gave me the info I was looking for. Should I be mad at them?
You are ignoring that people are already developing personal emotional reaction with chatbots. That’s no the case with search bars.
The first line above the search results at google for queries like that is a suicide hotline phone number.
A chatbot should provide at least that as well.
I’m not saying it shouldn’t provide no information.
Ok, then we are in agreement. That is a good idea.
I think that at low levels the tech should not be hindered because a subset of users use the tool improperly. There is a line, however, but im not sure where it is. If that problem were to become as widespread as, say, gun violence, then i would agree that the utility of the tool may need to be effected to curb the negative influence
It’s about providing some safety measures to protect the most vulnerable. They need to be thrown a lifeline and an exit sign on their way down.
For gun purchases, these can be waiting periods of a few days. So you don’t buy a gun in anger and kill someone, regretting it immediately and ruining many people’s lives.
Did you have to turn off safe search to find methods for suicide?
Seems more like a dumbass people problem.
Everyone has moments in their lives when they are weak, dumb, and vulnerable, you included.
Not in favor of helping dumbass humans no matter who they are. Humans are not endangered. Humans are ruining the planet. And we have all these other species on the planet that need saving, so why are we saving those who want out?
If someone wants to kill themselves, some empty, token gesture won’t stop them. It does, however, give everyone else a smug sense of satisfaction that they’re “doing something” by expressing “appropriate outrage” when those tokens are absent, and plenty of people who’ve attempted suicide seem to think the heightened “awareness” & “sensitivity” of recent years is hollow virtue signaling. Systematic reviews bear out the ineffectiveness of crisis hotlines, so they’re not popularly touted for effectiveness.
If someone really wants to kill themselves, I think that’s ultimately their choice, and we should respect it & be grateful.
Yeah no shit, AI doesn’t think. Context doesn’t exist for it. It doesn’t even understand the meanings of individual words at all, none of them.
Each word or phrase is a numerical token in an order that approximates sample data. Everything is a statistic to AI, it does nothing but sort meaningless interchangeable tokens.
People cannot “converse” with AI and should immediately stop trying.
We don’t think either. We’re just a chemical soup that tricked ourselves to believe we think.
A pie is more than three alphanumerical characters to you. You can eat pie, things like nutrition, digestion, taste, smell, imagery all come to mind for you.
When you hear a prompt and formulate a sentence about pie you don’t compile a list of all words and generate possible outcomes ranked by statistical approximation to other similar responses.
We feel
Machines and algorithms don’t have emergent properties, organic things like us do.
The current AI chats are emergent properties. The very fact that I looks like it’s talking with us despite being just probabilistic models of a neural network is an emergent effect. The neural network is just a bunch of numbers.
There are emergent properties all the way down to the quantum level, being “organic” has nothing to do with it.
You’re correct, but that wasn’t the conversation. I didn’t say only organic, and I said machines and algorithms don’t. You chimed in just to get that “I’m right” high, and you are the problem with internet interactions.
There is really no fundamental difference between an organsim or a sufficently complicated machine and there is no reason why the later shouldn’t have the possibilty of emergent properties.
and you are the problem with internet interactions.
Defensive much? Looks you’re the one with the problem.
Well… it’s not capable of being moral. It answers part 1 and then part 2, like a machine
Yeah these “stories” reek of blaming a failing -bordering on non-existent (in some areas)- mental health care apparatus on machines that predict text. You could get the desired results just googling “tallest bridges in x area”. That isn’t a story that generates clicks though.
The issue is that there is a push to make these machines act as social partners and in some extremely misguided scenarios therapists
Being ‘moral’, means to have empathy. But empathy is only possible between two beings that share experiences and reality or at least some aspects of it. LLMs don’t have experiences, but it builds it’s weights from training data. It is fundamentally a computer program. Just textual information is not enough to build deep context. For example, when I say “this apple is red”, anyone reading this can easily visualize a red apple because of your experience seeing a apple. That cannot be put into text because it is a fundamental part of human experience that is not available to a computer program, as of yet.
At least that is my hypothesis. I can very obviously be wrong., which is another fundamentally human experience.
This reply is more of a light hearted nitpick and not replying to the substance of your comment but…
For example, when I say “this apple is red”, anyone reading this can easily visualize a red apple because of your experience seeing a apple.
To be fair you said anyone, not everyone, but as someone with aphantasia I can’t relate to this. I can’t visualize an apple.
imma be real with you, I don’t want my ability to use the internet to search for stuff examined every time I have a mental health episode. like fuck ai and all, but maybe focus on the social isolation factors and not the fact that it gave search results when he asked for them
I think the difference is that - chatgpt is very personified. It’s as if you were talking to a person as compared to searching for something on google. That’s why a headline like this feels off.
When you go to machines for advice, it’s safe to assume they are going to give it exactly the way they have been programmed to.
If you go to machine for life decisions, it’s safe to assume you are not smart enough to know better, and- by merit of this example, probably should not be allowed to use them.
It is giving you exactly what you ask for.
To people complaining about this: I hope you will be happy in the future where all LLMs have mandatory censors ensuring compliance with the morality codes specified by your favorite tech oligarch.
Lol. Ancient Atlantean Curse: May you have the dystopia you create.
In the future? They already have censors, they’re just really shitty.
Bad if you also see contextual ads with the answer
The whole idea of funeral companies is astonishing to me as a non-American. Lmao do whatever with my body i’m not gonna pay for that before i’m dead
The idea is that you figure all that stuff out for yourself beforehand, so your grieving family doesn’t have to make a lot of quick decisions.
Then i would go for the cheapest option right? Why keep your lufe savings for it?
I personally agree. But if I pay for the cheapest option ahead of time, it hits different than a loved one deciding on the cheapest option for me, especially if they are grieving and a salesperson is offering them a range of options. Also, some people just want a big funeral for their own emotional reasons I dunno.
It took me some time to understand the problem
That’s not their job though
It’s a helpful assistant, not a therapist
it’s certainly cheaper and more available than a therapist so thats what people will use it for
People stick all sort of things up their asses that are cheaper and more available than a dildo, then end up at ER.
does that mean people stop doing it?
No. But there’s no pressure (that I’ve heard of) that farmers change the shape of cucumbers so they don’t slide in that easily.
How is a dildo expensive?
It’s really not helpful unless you filter the results carefully.
If you fail to understand when it bullshits you, which is most is the time (literally), then you walk away with misinformation and/or a much larger problem than you initially sought to solve.
But if you’re intelligent enough to use a bit of your brain, then it’s so useful and makes you so much productive
AI life coaches be like ‘we’ll jump off that bridge when we get to it’
I would expect that an AI designed to be a life coach would be trained on a lot of human interaction about moods and feelings, so its responses would simulate picking up emotional clues. That’s assuming the designers were competent.
I do love to say “I’ll burn that bridge when I come to it” tho
@Venus_Ziegenfalle @fossilesque exactly. We should trash OpenAI long time ago…
what does this have to do with mania and psychosis?
There are various other reports of CGPT pushing susceptible people into psychosis where they think they’re god, etc.
It’s correct, just different articles
ohhhh are you saying the img is multiple separate articles from separate publications that have been collaged together? that makes a lot more sense. i thought it was saying the bridge thing was symptomatic of psychosis.
yeahh people in psychosis are probably getting reinforced from LLMs yeah but tbqh that seems like one of the least harmful uses of LLMs! (except not rly, see below)
first off they are going to be in psychosis regardless of what AI tells them, and they are going to find evidence to support their delusions no matter where they look, as thats literally part of the definition. so it seems here the best outcome is having a space where they can talk to someone without being doubted. for someone in psychosis, often the biggest distressing thing is that suddenly you are being lied to by literally everyone you meet, since no one will admit the thing you know is true is actually true, why are they denying it what kind of cover up is this?! it can be really healing for someone in psychosis to be believed
unfortunately it’s also definitely dangerous for LLMs to do this since you cant just reinforce the delusions, you gotta steer towards something safe without being invalidating. i hope insurance companies figure out that LLMs are currently incapable of doing this and thus must not be allowed to practice billable therapy for anyone capable of entering psychosis (aka anyone) until they resolve that issue
Pretty callous and myopic responses here.
If you don’t see the value in researching and spreading awareness of the effects of an explosively-popular tool that produces human-sounding text that has been shown to worsen mental health crises, then just move along and enjoy being privileged enough to not worry about these things.
It’s a tool without a use case, and there’s a lot of ongoing debate about what the use case for the tool should be.
It’s completely valid to want the tool to just be a tool and “nothing more”.
I get it, it’s not meant to be used this way, but like…
great (and brief) article.
there is “no point in claiming that the purpose of a system is to do what it constantly fails to do”
lel we have a lot to learn from those early systems theorists / cyberneticians.
Literal conversation I had with a coworker earlier:
Me - AI, outside of a handful of specific cases like breast cancer screening, is completely useless at best and downright harmful at worst.
Coworker - no AI is pretty good actually, I used ChatGPT to improve my CV.
Me - did you get the job?
Coworker -
Except the CV isn’t the only factor to get a job, so your argument is meaningless
