Eliciting Informative Text Evaluations with Large Language Models