How Language Models Process Negation
arXiv:2605.03052v1 Announce Type: new
Abstract: We study how Large Language Models (LLMs) process negation mechanistically. First, we establish that even though open-weight models often provide wrong answers to questions involving negation, they do po…