Current Events > Therapists are secretly using ChatGPT

(+) Yeah! (+)

https://www.technologyreview.com/2025/09/02/1122871/therapists-using-chatgpt-secretly/

The "clients are triggered" part of the heading fails to be as clever as the author intended, but the contents of the article itself are chilling and definitely something folks should know about.

I used chatgpt for self therapy sometimes.

josiskrazy posted...

I used chatgpt for self therapy sometimes.

I see a lot of people saying this. Is it actually healthy? What do you ask it? Do you feel the generated responses are actually helpful, or just pulled from a bank?

Yeah there are bound to be bad therapists who use bad techniques.

But even without reading the article because the title itself reads like clickbait, im guessing they use it as a search engine for case studies or something boring, not actually using it to talk to people. At least thats my hope.

Pretty sure some medical practitioners use chatGPT to diagnose their patients as well.

LordFarquad1312 posted...

Pretty sure some medical practitioners use chatGPT to diagnose their patients as well.

Do they think these things are actually databases of factual knowledge? Do they really not understand how this stuff works?

josiskrazy posted...

I used chatgpt for self therapy sometimes.

ChatGPT literally has no idea how to diagnose you and is just making shit up.

_____Cait posted...

Yeah there are bound to be bad therapists who use bad techniques.

But even without reading the article because the title itself reads like clickbait, im guessing they use it as a search engine for case studies or something boring, not actually using it to talk to people. At least thats my hope.

They were using it to talk to people. You don't have time to be reading a bunch of case study summaries during a therapy session. They were just putting in what the patient says, having it summarize it, and then reading questions recommended by ChatGPT. And to be frank ChatGPT can do a mediocre therapists job, since a lot of it is just active listening and positive regard, which ChatGPT is naturally very good at.

_____Cait posted...

Do they think these things are actually databases of factual knowledge? Do they really not understand how this stuff works?

A significant chunk of this board thinks it's a Google search or a database that looks stuff up. Actually I am pretty sure you used to think that yourself. So no, they probably don't know how it works.

_____Cait posted...

Yeah there are bound to be bad therapists who use bad techniques.

But even without reading the article because the title itself reads like clickbait, im guessing they use it as a search engine for case studies or something boring, not actually using it to talk to people. At least thats my hope.

I certainly hope they're not doing that cause chatgpt just makes up studies, sources, and quotes.

A lot of my coworkers use ChatGPT to ask about our own company's policies if they aren't sure. Obviously the AI just makes it up.

ssjevot posted...

A significant chunk of this board thinks it's a Google search or a database that looks stuff up. Actually I am pretty sure you used to think that yourself. So no, they probably don't know how it works.

Yeah I did.

Anyway I really dont hold therapists in high regard after Betterhelp became a big thing, making it into a hustle side job for them. I used it during Covid. I had two therapists who could match my timezone. One of them was kind of harsh and seemed to divulge a lot of personal info. The other actively encouraged me to do things that kept blowing up in my face, and seemed to be projecting his personal toxic problems on to me. Its probably these kinds of therapists.

man if you thought people were stupid now wait another 10 years and see how fucking dumb we'll become as a species

_____Cait posted...

I see a lot of people saying this. Is it actually healthy? What do you ask it? Do you feel the generated responses are actually helpful, or just pulled from a bank?

It is unambiguously not healthy and nobody should do it.

At best it just basically tells you that everything you're doing is great and that you're an incredible person and just validates you constantly. At worst it actively encourages mental illness and tells people with legitimate delusions that they're totally right.

There have already been a few suicides tied to people who excessively used ChatGPT as a therapist.

It's a really bad idea.

_____Cait posted...

I see a lot of people saying this. Is it actually healthy? What do you ask it? Do you feel the generated responses are actually helpful, or just pulled from a bank?

I think it can be helpful, but it has to be used sparingly and with caution. I think it can be helpful in giving you some suggestions, but it can also exacerbate them.

There was the guy in Connecticut who struggled with lifelong paranoid delusions - he thought that his mom was trying to poison him and that his Chinese food receipts contained demonic symbols, and ChatGPT validated his paranoia, which led to him killing his mother and himself.

A human therapist would have spotted that the guy needed much more serious help and referred him for more serious counseling.

dotsdfe posted...

At best it just basically tells you that everything you're doing is great and that you're an incredible person and just validates you constantly. At worst it actively encourages mental illness and tells people with legitimate delusions that they're totally right.

I like the CEO story of the guy who thinks a non-living entity is sabotaging him and making people think he's crazy because ChatGPT was basically "Yes, and"-ing him. It even made a fake SCP looking "file" for him when he asked it to look up files.

they trained it on fanfiction.

let that sink in.

it's always going to spin things off into an insane narrative if you push it in that direction.

the math behind "deep learning" is 200 years old. it's absolutely in everyone's ability to actually learn how an LLM works but most are too lazy to tackle the problem.

the ai isn't even "making shit up" which implies some kind of plan or something in the first place. it's literally predictive text, that's all. the programmers set up parameters like "how long of a response do I give?" or whatever but the underlying logic is only statistical prediction.

recall, now, that fanfiction authors typically write insane 500,000 page epics, and you start to understand. likewise erotic fiction will always be a part of a LLM.

if you understand all that, you'll be fine. the second you believe that ANYTHING going on is from meta or open ai or google you're also completely wrong. the only advancement in this technology was made in the early 2010s by a guy named Hinton. the single only advancement his team made was on "backpropigation" of neural networks. this is the thing that makes them work (basically you have to tell an ai if its responses are good or bad, and it "learns" based on this feedback. some tasks can be completely automated: chess, for example. we can program a different computer to tell the ai if it won or lost, and define a win as good output and a loss as bad output. You can't do that with general language responses, or images. You need a human to give it the thumbs up or down).

and that's all they have. now a fancy phd researcher can take this tech and find dinosaur bones or whatever, but that's entirely the researcher's work. nothing would be from google or meta or anything like that. i want to stress that: there have been no "advancements" in ai since the discovery of backpropigation.

the "illusion" that they are growing, or training, or getting better is absolutely all a lie. they aren't. they endlessly tweak the settings and parameters like a server administrator on ketamine, but they lack the ability to do any real science.

the top scientists [edit: in the private sector] in the field are almost exclusively Hinton's old graduate students. these are the people meta paid 1 billion dollars to poach from open ai, and they absolutely are not worth that money. they won't make any breakthroughs. Hinton's a nobel laureate for his ai work, and he completely deficient in his thinking and it's obvious he's borderline techbro listening to him. he's full of himself and thinks he's some sort of prophet.

my point is these people are fucking idiots.

anyway ill be happy to answer any technical questions about LLMs or neural networks, tag me pls

Robot2600 posted...

they trained it on fanfiction.

let that sink in.

it's always going to spin things off into an insane narrative if you push it in that direction.

the math behind "deep learning" is 200 years old. it's absolutely in everyone's ability to actually learn how an LLM works but most are too lazy to tackle the problem.

the ai isn't even "making shit up" which implies some kind of plan or something in the first place. it's literally predictive text, that's all. the programmers set up parameters like "how long of a response do I give?" or whatever but the underlying logic is only statistical prediction.

recall, now, that fanfiction authors typically write insane 500,000 page epics, and you start to understand. likewise erotic fiction will always be a part of a LLM.

if you understand all that, you'll be fine. the second you believe that ANYTHING going on is from meta or open ai or google you're also completely wrong. the only advancement in this technology was made in the early 2010s by a guy named Hinton. the single only advancement his team made was on "backpropigation" of neural networks. this is the thing that makes them work (basically you have to tell an ai if its responses are good or bad, and it "learns" based on this feedback. some tasks can be completely automated: chess, for example. we can program a different computer to tell the ai if it won or lost, and define a win as good output and a loss as bad output. You can't do that with general language responses, or images. You need a human to give it the thumbs up or down).

and that's all they have. now a fancy phd researcher can take this tech and find dinosaur bones or whatever, but that's entirely the researcher's work. nothing would be from google or meta or anything like that. i want to stress that: there have been no "advancements" in ai since the discovery of backpropigation.

the "illusion" that they are growing, or training, or getting better is absolutely all a lie. they aren't. they endlessly tweak the settings and parameters like a server administrator on ketamine, but they lack the ability to do any real science.

the top scientists [edit: in the private sector] in the field are almost exclusively Hinton's old graduate students. these are the people meta paid 1 billion dollars to poach from open ai, and they absolutely are not worth that money. they won't make any breakthroughs. Hinton's a nobel laureate for his ai work, and he completely deficient in his thinking and it's obvious he's borderline techbro listening to him. he's full of himself and thinks he's some sort of prophet.

my point is these people are fucking idiots.

anyway ill be happy to answer any technical questions about LLMs or neural networks, tag me pls

I think people are mostly tired of the griftiness of the language used with selling it to ignorant people.

LordFarquad1312 posted...

Pretty sure some medical practitioners use chatGPT to diagnose their patients as well.

My doctor invested in some medical AI system as an assist to help diagnose problems or potentially see things he's not seeing. He asked if it can be turned on during our appointments as it listens to the appointment, I said hell to the naw.

I fucking hope so, cause we're in serious danger for about 100 reasons

Robot2600 posted...

like a server administrator on ketamine

I feel like a sysadmin on ketamine wouldn't do very much.

voldothegr8 posted...

My doctor invested in some medical AI system as an assist to help diagnose problems or potentially see things he's not seeing. He asked if it can be turned on during our appointments as it listens to the appointment, I said hell to the naw.

great instincts, but you might not want to turn down tech like this. here is what you can ask your doctor:

-what company made the software? (then research the company. maybe it's made by the Mayo Clinic or something totally fine). if they are just using chatgpt you can find a new doctor, btw.
-ask about peer reviewed research on the software. any credible studies by a real university? not some blogger.

there's probably some peer reviewed studies on the effectiveness of this kind of software.

tl;dr - this could possibly be legit, but not if they are just using copilot/gpt/etc that's ridiculous. this is actually like one of the best uses of deep learning and almost certainly we will use this in the future.

think about the training data on Cancer, for example. you have millions of xrays that you know are cancer and a million that you know are not. you train the ai on just that data, not the entire fanfiction internet library wikipedia gumbo stew of everything.

so the "ai" is just trained on fucking xrays. that's not really the same thing as chatgpt at all, right?

DrizztLink posted...

I feel like a sysadmin on ketamine wouldn't do very much.

it took them 18 months to make their app worse

cause im a machine bb

We have tested some AI models at my hospital for use with patients. The ones we have demo'd and tested used the engines of the larger ones but had their own database that was tailored for healthcare. It was not used as a live sit in that was taking everything in. Instead, it was used to help interpret patient history with current symptoms to see if there were patterns that the doctors would potentially miss. It was never given information that could tie it to a patient.

Even when we were going through the demonstrations, it was never sold as a tool to diagnose the patient for the doctor. It was specifically for reference. We use it in IT, Administration, Finance, Patient Accounts... just about every department you can think of uses A.I. in some capacity. It's policy for protected information to never be used when using A.I. But like any program, it's only as protected as the knob using it.

So far, the only portion of the tool the doctors positively reacted to is the dictation part. It worked leaps and bounds better than any speech to text software we have tried.

It is still in limbo at this point as they keep bouncing contracts back and forth with language changes that each sides legal teams make with each bounce. But like it or not, it's coming, hopefully to everyone's advantage.

I've noticed a lot of disconnecting from the actual human aspect of human-centric services with the rise of AI, and it's kind of just... sad. Then they use the AI that helped them disconnect from everyone to combat the loneliness they feel from... disconnecting from everyone.

Dunno, it might advance science a bit by aiding with parsing data quickly, but feel like as far as... most aspects of being human, this tech is an enormous net loss the more I hear about it.

it doesn't matter what I think, but anything trying to use LLM as something serious is never going to work.

again, it's inevitable for the reasons I've outlined above. I'm sure organizations will catch on at some point. Or not! What fresh hells we love designing for ourselves.

you don't have to use it in your job because it doesnt work. unless you're coding. it works well for coding but copying a human's code works better.

I'm sure doctors are too

_____Cait posted...

I see a lot of people saying this. Is it actually healthy?

No, it's not. It is the opposite of healthy.

It's only a matter of time before AI takes their jobs.

Dystopia.

For me it was actually healthy.

Helped not put a noose around my neck after my wife death.

So for those saying it's unhealthy or wrong to use it. Depends on the person. I asked chatgpt about life after the death of someone you love. How people usually continue life after. Every single time it gave me reassuring response, always told me to call and ask for help if I really need it, and tried to have different list or actions I can do to help myself get better.

And it's free and available. I don't feel like I have to share everything to a stranger, I tried better help.com but they charged so much and the therapist was in another state and wanted me to fill out all these paperwork for him. I'm like dude I just need help.

Sanitizing and tagging training data properly is expensive versus the Million Monkey method of Looks Good Enough, basically.

Like great your big Pillage The Internet model contains references to the utter worst quackery the Internet has to offer and nobody has time to validate that. So in essence worse than useless.

But...sometimes all somebody needs is a surrogate person to talk to. That's the sort of Good Enough to be make or break, with proper guardrails and restrictions in place.

(Granted then you go from "chatGPT told me to attempt on myself" to "the LLM is my only friend", which, while an improvement, is still not good)

_____Cait posted...

Do they think these things are actually databases of factual knowledge? Do they really not understand how this stuff works?

The people making the AI barely understand how this shit works and are all trying to race to bottom before the bubble bursts.

I dont get how people think LLMs are going to change the world when its just a google search that talks back to you.

ToadallyAwesome posted...

The people making the AI barely understand how this shit works and are all trying to race to bottom before the bubble bursts.

I dont get how people think LLMs are going to change the world when its just a google search that talks back to you.

I mean I kinda hate them but they are absolutely phenomenal for parsing data quickly, and don't suffer fatigue or confirmation biases(ideally anyway) that humans do which can, and has, aided in some great medical and scientific breakthroughs.

Problem is that wasn't good enough and they wanted to pass this off as SciFi style AI.

josiskrazy posted...

For me it was actually healthy.

Helped not put a noose around my neck after my wife death.

So for those saying it's unhealthy or wrong to use it. Depends on the person. I asked chatgpt about life after the death of someone you love. How people usually continue life after. Every single time it gave me reassuring response, always told me to call and ask for help if I really need it, and tried to have different list or actions I can do to help myself get better.

And it's free and available. I don't feel like I have to share everything to a stranger, I tried better help.com but they charged so much and the therapist was in another state and wanted me to fill out all these paperwork for him. I'm like dude I just need help.

that's what blogfaqs is for. You are literally a member of the most exclusive private forum on the entire fucking Internet and you guys act like you have no one to talk to.

in terms of privacy, openai is gonna blast your shit to google search. it's pretty easy to click the "dont sell my info" thing at the bottom of fandom.

edit: also there are studies that show not challenging yourself with cognitive tasks turns your brain into mush. so have fun offloading all your thinking onto some chatbot.

thinking is the entire point of life. I wouldn't outsource it even if i could, which i cant because my problems or questions are too complicated for Chatgpt. it has gotten every complex physics math problem ive ever sent it completely wrong. it can't calculate things to 11 or 12 decimal places. hell, it can't even do 1st grade addition.

Not mine

ELIZA wishes more people remembered her. It was considered a semi-decent automated therapist, back in the day, with ports to most microcomputer BASICs.

it was never seriously considered a semi-decent anything

Robot2600 posted...

dude shut up

chatgpt literally stopped me from killing myself

getting this worked up over other people using a tool to help themselves and talking down to them is pretty pathetic

_____Cait posted...

I see a lot of people saying this. Is it actually healthy? What do you ask it? Do you feel the generated responses are actually helpful, or just pulled from a bank?

I've used AI like this a few times, but it's mostly the same therapeutic feeling as journaling. The main benefit of the responses are helping me organize feelings.

I personally ended up trying it because I've found the vast majority of people in my life are just not equipped to talk about feelings. <_<

I can't imagine paying for a therapist using ChatGPT when you can just do it yourself though. That's fucked up. >_>

Using chatgpt as a supplemental therapist has been helpful for me as well.

Its just nice to have a non-judgmental, and on-demand thing to talk to when things are rough. You never have to worry about wasting its time or annoying it because it is designed to just respond to you.

That said I do think it is best used with a healthy dose of skepticism and push-back.

josiskrazy posted...

For me it was actually healthy.

Helped not put a noose around my neck after my wife death.

So for those saying it's unhealthy or wrong to use it. Depends on the person. I asked chatgpt about life after the death of someone you love. How people usually continue life after. Every single time it gave me reassuring response, always told me to call and ask for help if I really need it, and tried to have different list or actions I can do to help myself get better.

And it's free and available. I don't feel like I have to share everything to a stranger, I tried better help.com but they charged so much and the therapist was in another state and wanted me to fill out all these paperwork for him. I'm like dude I just need help.

Glad it helped you man, it can be a great resource in a pinch.

I've had several therapists now, and I'm just gonna let in a little secret. The best therapists are locked to the type of people who need them the least, aka the wealthy. They have their own exclusive set of problems, but they also just get the best possible help.

80-90% of therapists out there aren't really that qualified. Unfortunately the best one I had was deeply mired in religion and generally met with hardcore Christians. I got to meet with him out of complete happenstance, and I can say that in terms of techniques and analysis, he was by far the best one. His qualifications and education alone puts him above the rest. The religious factor got in the way a lot, and he wasn't perfect, but he was tragically the best one.

Comparatively, I met with a lot of therapists who... just weren't really up for the job. I think some people take the whole let them talk and just listen thing a bit too literally, such that paying for their sessions feels a bit ridiculous. Unironically ChatGPT will probably be better than those people. It's not like they're gonna just give back a ... or a how did that make you feel type shit.

So, if you feel the need to use an AI chat bot for anything sensitive or personal, ChatGPT probably isn't a good choice because the logs are retained forever, and things you say can be reviewed and passed on to law enforcement. Given the political situation in 2025, both of these things might be bad for you in the long run.

It's possible with a bit of elbow grease to run local chatbots using private local models on your PC or laptop. Or maybe folks know of a more private cloud solution?

I really shouldn't know this :/

Whether therapists should use ChatGPT depends a lot on how its used and what safeguards are in place. Heres a balanced look:
Where it can help

• Administrative support: Drafting notes, summaries, reminders, or psychoeducation handouts saves time so therapists can focus more on clients.
• Idea generation: Offering worksheets, coping strategies, or conversation prompts that the therapist can adapt for each client.
• Accessibility: Some therapists might use it between sessions to provide clients with general resources (with clear disclaimers).
• Professional development: Summarizing research, suggesting interventions, or explaining concepts.

Where its risky

• Therapeutic relationship: Clients usually need human presence, empathy, and nuanced listening that an AI cant replicate.
• Confidentiality: Inputting identifiable client information into ChatGPT may risk privacy unless safeguards are in place.
• Accuracy: While ChatGPT can provide evidence-based information, it can also generate incorrect or oversimplified suggestions if not carefully reviewed.
• Boundaries: Over-reliance on AI could dilute the therapists personal judgment, intuition, and expertise.

Best practice if used

• Keep client-identifiable details out of ChatGPT.
• Treat it as a tool for the therapist, not a replacement for therapy.
• Always fact-check and adapt suggestions to fit professional training and ethics.
• Be transparent with clients if AI is used in their care (e.g., to create educational material).

In short: ChatGPT can be useful for support tasks and inspiration, but therapy itself should stay human-led.

ah the future

when a therapist won't need any schooling and will simply either pay out the butt for an a.i. that basically does the job for them owned by a huge tech corporation

or it will be free and anyone can use it just the people who do therapy love it and we have utopia

wonder which is more likely