Raw LLM Responses

MAD is more or less a concept humans have decided on to feel better about wielding doomsday devices. It’s never really been tested, and the few times it’s been stressed as a principle (false alarms that seem valid in every way, undocumented test launches close to allies, etc.) humans at the end of the chain made a blatantly illogical choice to ignore all evidence and orders to deny a launch. There’s no way to hardcode MAD without also telling AI to ignore logic and evidence in its decision making process in spite of direct instructions from a human.