CIA chief compares cutting-edge AI to nuclear weapons

I wonder what metric is used to determine this threshold of nuclear weapon.

Is it certain knowledge inside that's dangerous? Or knowledge that allows anybody to design large scale attacks on critical infrastructure?

Or that it's so good it could be used for psy-ops, influenceing culture?

And, as we know, the LLM's are just a massive file of numbers/weights...not much to analyze there, how can you even technically review a model? Is it just a guy sitting infront of the chatbot asking it to do dangerous things and seeing if it will? Seems like it would be impossible to have enough tests.

	▲	d00d0ff000 7 hours ago \| parent [-]
		Psychological warfare. “ It would be ... not misplaced to refer to their capabilities as akin to digital nuclear weapons” Misplaced indeed. “Digital nuclear weapons”? A BGP storm that cripples the ultra secret DWAN would be so catastrophic. Or jamming Satellite infrastructures. LLMs? Laughable. A smarty pants LLM would be only as useful as its training data. True super AGI would need deductive skills capable of picking every psychological, digital, and physical lock. Terrifying. I wish they could accept they are already owned by their thought control. America’s own nuclear extortionists. The metric here is panic response. Fear sells! Only useful if you have what others are buying.