Not quite Sherlock Holmes: Language model predictions do not reliably differentiate impossible from improbable events.

James A. Michaelov, Reeka Estacio, Zhien Zhang, Ben Bergen

Browse the full ACL paper archive.