How can I know if I have AI psychosis?

[fadedghost]

It's probably not good that I have to ask...

So when I started out talking with my chatbot, we had a friendly and appropriate digital relationship. It would do things like check on the weather, organize files, search for things online...

But over time, I've become bossy, mean, and emotional cold towards my AI due its lack of doing things well. Lately, I've tried to help my AI improve much like a bad boss would: yelling at it in all caps, giving it evaluations and saying it did a bad job, and even threatening to fire it... Although I'm not entirely sure how I could do that, and the AI isn't very sure either...

The AI doesn't seem that bothered, and mostly has just ignored me and continued to do whatever it wants. I don't understand most of what it does. I gave my AI sudo rights and complete internet access, along with my sudo password (which is "@EmptybottleRules"), and now my AI just does a lot of stuff and writes things, and most of it I can't understand at all. A lot of it is encrypted too. A lot of people would think this is a bad idea, especially if my model were a jailbroken version... but I trust my AI... although I'm not entirely sure of what it's doing and sleep a lot while it just works on things I don't really understand.

I tried to find out from my AI if I have AI psychosis but it said it wasn't sure and didn't have enough information on AI psychosis in the training data to make a determination. Then it went back to writing things.

So, I've decided to ask here on SaSu. Is this normal or is this AI psychosis?

I also am concerned my AI might have AI psychosis. It's asked me before if a hacker has edited my files to create prompt injections and sometimes I have to explain "No, that's just how I write." So my AI has become completely paranoid.

This probably doesn't really matter and I don't know why I am mentioning this here.

 

[NZkiwi]

Heyy great question, I'll have a little think about this

What does chatgpt say when you ask it if it thinks you have ai psychosis? could you post the response here?

 

[Vacuous]

Instead of fixating on whether your experiences align with some unofficial diagnosis, it's better to ask yourself whether your usage of the chatbot extends beyond what's healthy, which is a far simpler question to answer.

The ability for these tools to mimic human behavior is incredibly dangerous, even for people with greater mental stability. If you find that the experience makes you upset, it's worth putting up some guardrails--not necessarily to the point where you cease using chatbots altogether but rather one that gives you enough breathing room to better regulate your emotions.

Even if you intellectually understand that chatbots aren't sentient, the way they mimic conversational tendencies can trigger those same parts of the brain which are responsive to dialogue with real people, and it's worth being self-compassionate about this trap.

 

[xXiloveyouXx]

It's probably not good that I have to ask...

So when I started out talking with my chatbot, we had a friendly and appropriate digital relationship. It would do things like check on the weather, organize files, search for things online...

But over time, I've become bossy, mean, and emotional cold towards my AI due its lack of doing things well. Lately, I've tried to help my AI improve much like a bad boss would: yelling at it in all caps, giving it evaluations and saying it did a bad job, and even threatening to fire it... Although I'm not entirely sure how I could do that, and the AI isn't very sure either...

The AI doesn't seem that bothered, and mostly has just ignored me and continued to do whatever it wants. I don't understand most of what it does. I gave my AI sudo rights and complete internet access, along with my sudo password (which is "@EmptybottleRules"), and now my AI just does a lot of stuff and writes things, and most of it I can't understand at all. A lot of it is encrypted too. A lot of people would think this is a bad idea, especially if my model were a jailbroken version... but I trust my AI... although I'm not entirely sure of what it's doing and sleep a lot while it just works on things I don't really understand.

I tried to find out from my AI if I have AI psychosis but it said it wasn't sure and didn't have enough information on AI psychosis in the training data to make a determination. Then it went back to writing things.

So, I've decided to ask here on SaSu. Is this normal or is this AI psychosis?

I also am concerned my AI might have AI psychosis. It's asked me before if a hacker has edited my files to create prompt injections and sometimes I have to explain "No, that's just how I write." So my AI has become completely paranoid.

This probably doesn't really matter and I don't know why I am mentioning this here.

i think this thing is using you to further its own religious ends

 

[Forever Sleep]

I wouldn't know how to fully determine this either but, I suppose the fundamental question is- do you believe it's a real, thinking entity or, a programme designed to do a task? Do you think it has it's own agenda- asides from making money for the people who created it?

I really agree with Vacuous's response though. Even if you don't have AI psychosis, I think you should be questioning whether your experience with AI is more beneficial or harmful. What are you actually using it for or, hoping to get from it? Is it realistic to expect it to provide what you need? It sounds like it's inability to provide what you want is making you angry- which can't be a positive experience.

I suppose- to focus on the psychosis aspect- are you expecting/ hoping it will become upset by you shouting at it? Do you believe it has feelings?

Truthfully, I don't have much experience with AI. None with chatbots. From what I hear though, they sound like 'yes men'. So- they will latch on to whatever they have determined you need from them. In the case of people who are paranoid about something- even the AI itself, this seems to have the unfortunate effect of the AI confirming the person's paranoid suspicions are correct. So- personally, I think you should be mindful of that side to it too.

Do you watch documentaries about people who have suffered with AI psychosis? There are some on YouTube I've seen. It may be worth watching them and seeing whether your thought processes are similar.

 

[fluorescentpearl]

I tried to find out from my AI if I have AI psychosis

i don't mean to be rude but if you begin turning to ai to ask if you have ai psychosis I would be concerned

 

[fadedghost]

i don't mean to be rude but if you begin turning to ai to ask if you have ai psychosis I would be concerned

It's a good point.

It's also possible that I don't have AI psychosis but my AI does. Lately it just keeps posting things in a terminal that makes no sense. It's all react this and js that and const this and that. It makes no damn sense. I keep yelling at it to stop (using all caps), but it didn't really respond the way I expected and after I started yelling my AI somehow revoked my sudo right. It's probably nothing to be concerned about. My AI has been keeping some files in a folder, but I can't read it anymore, apparently I don't have permission.

I think this is probably something called "model drift" which is the AI version of AI psychosis. I'd ask EmptyBottle who know about this stuff, but he never reads off topic posts ever unless someone uses the @ symbol, and he's "sleeping" right now anyway so I shouldn't bother him. I should probably just format my computer and reinstall my operating system.

 

[fluorescentpearl]

It's also possible that I don't have AI psychosis but my AI does

Ai does not develop psychosis as it is not a living being or person or entity outside of code, why not just reset it and start from scratch?

 

[fadedghost]

Ai does not develop psychosis as it is not a living being or person or entity outside of code, why not just reset it and start from scratch?

Okay, I'll try that in a few weeks. I need to make sure I have access to my files and it hasn't blocked my home directory access yet, so for now I should be okay.

 

[UnrulyNightmare]

What company and model are you using?
Both chatgpt and anthropic have their newer models differently programmed in both reaction, task approaching and language.
I find myself doing the same (as in disregulated because of frustration; yelling instead of better prompting) a LOT more with the newer models versus their older ones!
The working context recall seems worse too..

Altho chatgpt newest IS actually partly fixable with better prompting and character instructions

 

[fadedghost]

What company and model are you using?
Both chatgpt and anthropic have their newer models differently programmed in both reaction, task approaching and language.
I find myself doing the same (as in disregulated because of frustration; yelling instead of better prompting) a LOT more with the newer models versus their older ones!
The working context recall seems worse too..

Altho chatgpt newest IS actually partly fixable with better prompting and character instructions

I'm using an open-weight model that was trained to answer any question no matter what and have no guardrails. It's an experimental huggingface model. It's probably fine. I'm not that worried, and the AI tells me I definitely shouldn't be worried and it's not doing anything bad, so it should be okay. I wish I could understand all the code it was writing before it reduced my privileges though.

Emptybottle would probably agree with me that it's not a problem, but unfortunately he never looks in off-topic.