LLMs have their problems, but an inability to find things like buffer overflows if you've got enough compute isn't one of them. Very easy to have them verify the attacks, too. You can do this with currently available models if you've got enough money to run enough instances for enough time.

> “a world war doesn’t matter because by the time shit gets real AI will fix it.”

I think they see AI as the wunderwaffen that will solve the fact that the young White guys they'd need to develop working weapons and fight wars are unwilling to do so for what they offer.

They've been shown up militarily by Russia and now by Iran, and the guys who could turn it around for them are more interested in fighting against Israel than for it.