I’m saying that it is possible to learn how to optimise for “better” or “worse” decisions without being trained on an opinionated dataset of every single event that has happened. An “intelligence” could exist without necessarily answering your nukes question.

You’re bringing alignment to a capability discussion.