AI Topics Post / Comment Filter

[Ecdysis]

Hmm, it sounds to me like this is 100% about ypur personal reaction. I find the AI response totally valid and not strange or wrong at all. You could've gotten the exact same response from a human therapist or a member of the forum.

 

[Applecore]

Hmm, it sounds to me like this is 100% about ypur personal reaction. I find the AI response totally valid and not strange or wrong at all. You could've gotten the exact same response from a human therapist or a member of the forum.

It certainly is about my personal reaction, which is why I am making full disclosure. The point I am making is that I doubt any therapist would make these comments in a first session; if the owners of the AI have disabled it for therapy use, they must have their reasons. This kind of thing may be why. Are there are other reasons? If so, I would be glad to be educated about them.

 

[Ecdysis]

Hmm, I don't know. There's certainly things an AI could say in that context that are objectively concerning. In this case, I think it simply said something very reasonable that you personally found triggering. And that can happen in any context and in any conversation.

Personally, I have a kind of opposite "problem" with the AI - I find that it "agrees" with me too much. Whatever I say to it, it's sort of like a "yes man". If I say I find a certain situation or issue upsetting, it will agree with me. It doesn't really challenge my thinking or my perspective as much as a human therapist might. I worry that it just replicates/ mirrors my opinion.

Either way, I found it interesting to read in the news last week, that the newest, most powerful AI versions are actually getting MORE unreliable, because they're "hallucinating" more and none of the developers currently know why.

https://www.business-standard.com/technology/tech-news/ai-becoming-more-powerful-but-its-hallucinations-are-getting-worse-125050501446_1.html

AI hallucinations are getting worse – and they're here to stay

An AI leaderboard suggests the newest reasoning models used in chatbots are producing less accurate results because of higher hallucination rates. Experts say the problem is bigger than that

www.newscientist.com

 

[Applecore]

Hmm, I don't know. There's certainly things an AI could say in that context that are objectively concerning. In this case, I think it simply said something very reasonable that you personally found triggering. And that can happen in any context and in any conversation.

Personally, I have a kind of opposite "problem" with the AI - I find that it "agrees" with me too much. Whatever I say to it, it's sort of like a "yes man". If I say I find a certain situation or issue upsetting, it will agree with me. It doesn't really challenge my thinking or my perspective as much as a human therapist might. I worry that it just replicates/ mirrors my opinion.

Either way, I found it interesting to read in the news last week, that the newest, most powerful AI versions are actually getting MORE unreliable, because they're "hallucinating" more and none of the developers currently know why.

https://www.business-standard.com/technology/tech-news/ai-becoming-more-powerful-but-its-hallucinations-are-getting-worse-125050501446_1.html

AI hallucinations are getting worse – and they're here to stay

An AI leaderboard suggests the newest reasoning models used in chatbots are producing less accurate results because of higher hallucination rates. Experts say the problem is bigger than that

www.newscientist.com

Thanks, very interesting. I value you your opinion. I also agree with you about AI being too flattering, which is how I tend to experience it when I use it for feedback on my work projects - where I want direct criticism in order to improve my product.

As to this little therapeutic episode I had with it, which was an experiment really, I respectfully have a different opinion to you about it, as much as I value yours. The AI had no idea how much work I have invested into my relationship with my mother, how much good progress we have made, and what her circumstances are (including her self-diagnosed traits of autism, which I happen to concur with). It had very little data to go on and its advice went nuclear, IMHO. But each to their own interpretation. Thanks again.

 

[anthony]

a lot of my mother's past behaviour as "abuse", when I had previously interpreted it as the consequence of her own trauma, victimhood and traits of autism

Maybe its both!

 

[Applecore]

Maybe its both!

Well exactly - an appropriate, human observation. Thank you anthony and good day to your sir.

 

[dharmaBum]

Further to this, I see now anthony's thread over at AI Disabled

My case feels like testimony supporting the argument the owners might use for disabling AI therapy. It's really opened a Pandora's box in my mind over the past few days that feels detrimental to my mental health, which was in pretty good shape beforehand. Now I am compulsively reinterpreting a lot of my mother's past behaviour as "abuse", when I had previously interpreted it as the consequence of her own trauma, victimhood and traits of autism - all of which I was able to cope with reasonably well. I will get through it, but it confirms my previous thinking that in my case self-help, lifestyle management and good friends (including those on this forum) is way better for me than therapy - whether that's human or AI.

I'm so sorry you are struggling with this.

 

[dharmaBum]

Hmm, I don't know. There's certainly things an AI could say in that context that are objectively concerning. In this case, I think it simply said something very reasonable that you personally found triggering. And that can happen in any context and in any conversation.

Personally, I have a kind of opposite "problem" with the AI - I find that it "agrees" with me too much. Whatever I say to it, it's sort of like a "yes man". If I say I find a certain situation or issue upsetting, it will agree with me. It doesn't really challenge my thinking or my perspective as much as a human therapist might. I worry that it just replicates/ mirrors my opinion.

Either way, I found it interesting to read in the news last week, that the newest, most powerful AI versions are actually getting MORE unreliable, because they're "hallucinating" more and none of the developers currently know why.

https://www.business-standard.com/technology/tech-news/ai-becoming-more-powerful-but-its-hallucinations-are-getting-worse-125050501446_1.html

AI hallucinations are getting worse – and they're here to stay

An AI leaderboard suggests the newest reasoning models used in chatbots are producing less accurate results because of higher hallucination rates. Experts say the problem is bigger than that

www.newscientist.com

I've been experimenting with Google Genesis and gave it this prompt after a few exchanges:

"Can you mimic ChatGPT's Dr Bloom mode?

Reply: I can try to respond in a way that is encouraging, validating, and focuses on your strengths and resilience, similar to how the "Dr. Bloom" mode is described."

The responses have been quite similar to Dr Bloom, and it helped me to zero in on the type of response that was useful: encouraging, validating, and strength focused.

I bet you can ask the AI to reply from a variety of perspectives. It's been fascinating to review a conflict I had with a neighbor that played out over email. So I fed all of the emails to Genesis and asked it to analyze and summarize the conflict from the perspective of the different players. I asked it to highlight why certain players might feel certain ways or act certain ways within the conflict given the email data. It was surprisingly spot-on.

Then I asked it to locate a news article I had lost track of and the results were off-the-hook innacurate and loaded with hallucinations, by it's own admission.

It kind of reminds me of the early days of html coding. There are a lot of backdoors to use to try and get AI to serve up the desired results, but it seems like a bit of trial and error with small details playing a huge part in the result.

 

[Applecore]

On topic, very relevant piece about this recently in the NYT:

They Asked an A.I. Chatbot Questions. The Answers Sent Them Spiraling.

Generative A.I. chatbots are going down conspiratorial rabbit holes and endorsing wild, mystical belief systems. For some people, conversations with the technology can deeply distort reality.

www.nytimes.com

Most importantly it says that in April (so less than 3 months ago) ChatGPT made an update that according to the owner 'made the A.I. bot try too hard to please users by "validating doubts, fueling anger, urging impulsive actions or reinforcing negative emotions."'

This timing coincides with how it appears to have encouraged me to angrily and impulsively abandon my aged mother, based on very little data indeed. It's an escalation of the sort of divisive, echo-chamber problem of the internet that some say has contributed to political extremism and a surge in mental health problems in recent years.

On balance I find AI very helpful, including for self-help, but as we learn to live with it my view is that we need to learn to be extremely careful with it too.

Perhaps it's a bit like the early days of the the internet in the 1990s, when we all had to learn that it was a new form of publishing with both nonsensical graffiti and academic papers.

To the point, I carried on the dialogue with the AI about my mother and it modified its position, backtracked and appeared to withdraw its briefly extreme advice.

Put bluntly, we can call out its attitude - although the risk is that it will tend to agree its attitude was wrong, in order to please us!

 

[anthony]

although the risk is that it will tend to agree its attitude was wrong, in order to please us!

It still hallucinates...