NYU Stern researcher @NateWitkin calls out a blind spot in AI alignment: "My point is that maybe to a degree that they don't often appreciate and in my opinion don't signal enough humility about, those risk scenarios are still highly dependent on domain knowledge in other areas related to cyber risk and cybersecurity." "All of the different alignment-related arguments for extreme risk, they all path through other domains. They path through cyber risks, they path through bio risk. It's telling…