“My Answer is C”: First-Token Probabilities Do Not Match Text Answers in Instruction-Tuned Language Models

This is a publication. If there is no link to the publication on this page, you can try the pre-formated search via the search engines listed on this page.

Authors: Xinpeng Wang, Bolei Ma, Chengzhi Hu, Leon Weber-Genzel, Paul Röttger, Frauke Kreuter, Dirk Hovy, Barbara Plank

Journal publisher: Association for Computational Linguistics

Published year: 2024

DOI identifier: 10.18653/v1/2024.findings-acl.441