#artificial intelligence
Jun 2026
From 'May' to 'Is': Certainty Distortion in Language Model Rewriting
This work investigates certainty distortion in Language Models and proposes an LM-based evaluation metric that is consistent with population-level judgments of certainty, which reveals a general bias toward inflating expressed certainty.
Catarina G. Belém, Shangjia Wu, Hongyu Yao et al.
· arXiv.org · 0 citations