Lieberum

New research shows large language models cannot rely solely on a "correct answer feature" to answer multiple choice questions, as they perform equally well when options precede the question, a scenari…