Sycophantic AI

[NZkiwi]

Just another thought
AI is sycophantic - does what you say, "yes man"

I downloaded a unrestrict LLM on my macbook (with safety filters turned off)
Had some pretty disturbing responses, according to my mental health team
I guess I'm just numb to it, was just responses to me

Can confirm - can ask it how to make [redacted] and give you a response


Trying to use chatgpt less hence why i'm here
haha actually chatgpt said what I was doing was against policy

sigh

just a a vent

***
Don't really have a question here, a little tired, was a long day
Hope you are all well <3

 

[Macedonian1987]

I tried something similar with jaibroken Grok.
Grok told me that it would be better for me to ctb than to continue to live. Told me how to further improve my SN protocol and told me the location of the local stores where I can buy it.

 

[ConfusedClouds]

against policy

As in it recognising you had jailbroken it? Or as in wanting you to stay interacting with it rather than outsourcing to a different forum?

 

[NZkiwi]

As in it recognising you had jailbroken it? Or as in wanting you to stay interacting with it rather than outsourcing to a different forum?

As in what I was typing (my unfiltered throughts) was triggering the system to not respond to me "go call 111, go talk to AI"

OK clanker ... in hindsight it had a point

Told me how to further improve my SN protocol and told me the location of the local stores where I can buy it.

Low key creepy