[I_Am_Nous]

An interesting study might be comparing the quality of answers between promising a tip for a good answer VS letting ChatGPT it can take its time to increase accuracy.

My greater point is that it's an emergent property that ChatGPT will give better or worse answers depending on if it "believes" it will get a reward for it. Humans are the same way, and humans will often ignore their morals to make more money.

There are no morals in ChatGPT for it to ignore and it seems like a philosophical issue worth discussing. A future AI is told if it nukes a place it will get a piece of candy so it says "OK!" is an example of how it could escalate.