Grade Like a Human: Rethinking Automated Assessment with Large Language Models