Fortune reposted this
People are always advising executives to think about AI like an intern, or PhD-level researcher, or a expert colleague who is occasionally fallible, or even a middle manager. But all these analogies to types of human workers are actually a kind of category error. It doesn't capture how truly bizarre AI models are. We no more understand how these models "perceive" the world than I truly understand how my dog perceives the world. Today's AI is more like working with an alien species than it is working with a human, be that an intern or experienced pro. Recent research from Stanford on the way in which multimodal AI models can experience "mirage reasoning" further highlights this. I cover that research and plenty more in my latest Fortune "Eye on AI" newsletter.