An AI scientist can reportedly design an experiment → tell robots to perform it → analyze the results → decide what experiment to try next. That's much more consequential than another chatbot benchmark. It's basically an early version of an autonomous scientific researcher.
