It's like talking to a mental defective aspie who seems to remember a lot of random shit but is really just spouting nonsense when pressed

Yes, because LLMs are just glorified calculators that predict what sounds like a good reponse to what you sent, and then sends it. It doesn't know anything.

>>22830604
it can be much more useful than it is, when properly trained & not overfitted & without whatever filter layers are added

i saw some paper a couple years ago which showed the responses to 20-30 questions before & after a filter layer was applied
it was literally like the soul was removed
it was originally more crass & schizophrenic
dunno the details from that paper & i can't find it, but it stuck with me

>>22830608
when openassistant was relatively knew I got it to start hallucinating and going off on this kind of acid freak rant about reality

>>22830612
nice lol

do you remember any details?

>>22830620
no I don't and when I went back to openassisstant to see the instance which I thought it happened I couldn't find it so maybe I hallucinated the whole thing into some kind of lucid dream I remember as a real memory

>>22830683
it could be they wipe the logs lol
we're all just rats in the dark