If there is no "hardcoded" safety, I guess there is always a way to "convince" a neural net to "workaround" any prompt modifications. Maybe it has to be at the

If there is no "hardcoded" safety, I guess there is always a way to "convince" a neural net to "workaround" any prompt modifications. Maybe it has to be at the core of the machine learning process, and even that...