Let’s not anthropomorphize silicon now. It’s wrong, not hallucinating.

This is very bad. She talks about "...needing to accept potentially imperfect compute....", which of course underpins the entire "AI" phenomenon. We are going to have to make very hard and sharp distinctions between areas where correctness does and does not matter. In order for anything to be correct, cross-contamination will have to be absolutely prevented.