一作ACL 2026Findings· Findings of ACL
D iff ER : Diffusion Entity-Relation Modeling for Reversal Curse in Diffusion Large Language Models
Shaokai He · Kaiwen Wei · Xinyi Zeng · Xiang Chen · Xue Yang · Zhenyang Li · Jiang Zhong · Yu Tian
摘要
The “reversal curse” refers to the phenomenon where large language models (LLMs) exhibit predominantly unidirectional behavior when processing logically bidirectional relationships. Prior work attributed this to autoregressive training—predicting the next token inherently favors left-to-right inf...