AIs are more accurate at math if you ask them to respond as if they are a Star Trek character — and we're not sure why

Researchers asking a chatbot to optimize its own prompts found it was best at solving grade-school math when acting like it was on Star Trek.