Sam Clarke comments on Some thoughts on risks from narrow, non-agentic AI