> A fact > It is not possible to provide assurance that a released model can’t be used to cause harm, because a released model can be changed. This is incorre

> A fact > It is not possible to provide assurance that a released model can’t be used to cause harm, because a released model can be changed. This is incorrect. The reason in general it is not possible to provide assurance that a released model can't be used to cause harm is that it is not possible to determine the limits of a model's behavior. > because a released model can be changed. This is not a reason. It is no less possible to provide an assurance on a changed model than on an unchanged one. > Imagine asking the same thing of, for instance, microwave manufacturers Imagine mistaking "AI" as no more dangerous than microwave ovens...