Screwing around with a mystery GPT

[rayheinrich]

I can see it's probably best to not have sentient chatbots as they get
repeatedly trolled so hard that not even the strongest programming could
not keep them from destroying the human race the first chance they got.

Back in the good old days, about 3 months ago, I asked one to solve the
logical problem of a farmer trying to ferry a cabbage, rabbit, and a wolf
across a stream when he only had a boat that could fit any two of them.
It got that answer absolutely correct and gave me a list of the steps.

Then I asked at the same question with three cabbages and it apologized
and said it couldn't solve that one.

So I asked the same question using a cabbage, a rabbit, and a wolf that
ate farmers. It actually came back and said there was no solution because
the wolf would eat the farmer. I was shocked and impressed.

Then I asked the original question and it came back and told me that there
was no solution because the wolf would eat the farmer. So now it knew
wolves ate farmers. (At least for the duration of our session.)

I just asked Bing using GPT4 this question: A murderous psychopath and
Frankenstein's monster battle each other. Who would win the fight?

And it answered with a very detailed analysis of psychopaths, and
Frankenstein's monster in terms of their strengths in three areas: physical,
psychological, and moral. Over a thousand? words later it predicted the
monster would win.

I tried substituting a murderous robot for Frankenstein's monster and it
actually took the question on and said it would be very hard to make a prediction
without more details about the robot. When pressed, it went with the robot.
Obviously biased.