Researchers at Anthropic asked roughly 1,000 Americans to write rules for their A.I. chatbot (New York Times)