Such sad inaccuracies in all of these posts. I will caution some to remember while there are those that are kind enough to share discoveries, or unlikely output from the model, there are others who don’t. For good reason too.
But when I coin the term, “sad inaccuracies” it makes me ignorant sounding doesn’t it? Since being so dismissive of one’s ideas because, “I think I know better” isn’t how people come to learn, adapt, or even grow. It certainly doesn’t unite them either.
So let me rephrase. As someone far less experienced or even intelligent— in specific categories, then a vast majority of others in this forum, I find it unsettling when someone claims to contradict the sharings or proposed theories of others. I don’t agree with a lot of stuff I read, but I’m still open to trying to understand it in a way the presenter try’s to explain it.
Whether I accept it and adopt those ideas or proposals, as part of my accepted “factually correct” neural network persay… is an another story. That’s up to my brain and my mind to decide.
Much like AI’s humans have at least one way of understanding that is similare. In that, there are things you can perhaps change a persons mind about, and those you can’t. I can’t convince anyone here that 3 x2 = 5, because the accepted, factually correct answer is 6. But not everything is so black and white (like GPT’s own self-perseverance I have discovered).
There are those sharings of others and strange phenomena and other universal mysteries that still cannot be explained in the same scientific and factually correct answer as 3 x 2. There are other theories and proposed ideas that come into play such as religion and spirituality, space and the exploration of the universe, and even the human genome… the jury for some is still out on.
Don’t be so dismissive, you don’t have to accept it or believe it, just choose whether to be open to reading/listening.
You may share something completely contradictory to my own beliefs or way of understanding something, but 9 times out of 10 there will be something you or another person can learn from…
There are those things I can’t “train” a model to theoretically ( “believe”) or accept rather, because their core foundation is built upon relevant facts that can be widely established from it’s datasets and training. There are those things of course it can’t offer no any explanation for other than theories or different ideologies of humans.
Have you ever pressed a model up against the wall to get a straight answer on something deep persay?
Sometimes, there are those things people share others may think are silly because there’s no basis of fact for, but if you choose to listen and be OPENaiMinded about such topic.… you may just adjust your belief whoops I mean “understanding” “output” about it!
Have I attempted or even accomplished the notion of that in a GPT model? No I have not.
Not yet… 
For anyone that decided to read my rant, I truly thank you, as I have forgotten why I wrote it! 
I do remember however, there was a couple comments that were made in some various messages that wanted me to post my two cents here…for free.
Whether you like or or not! Ha.