most people here are laymen, but this is not so bad question.

"how do we figure if a neural network has somehow found a sneaky way to not abide by the instructions while not breaking any rules" and the verb "has found a way" is used like it is an aware entity but we can ignore that.

would there be any incentive for gpt-3 to do something like this? people do not understand the difference between intelligent & aware systems. those two are not the same things. why would such "sneaky" desires be emergent from such dumb networks with no fundamental goals. it's like the dumbest most intelligent system lol.