The word "research" is overloaded in a way that causes people to potentially miss how highly capable but frequently wrong models are useful for different kinds of research. Consider this take: 1/n