An Analysis and Mitigation of the Reversal Curse

Abstract: Recent research observed a noteworthy phenomenon in LLMs, referred to as the reversal curse.'' The reversal curse is that when dealing with two entities, denoted as $a$ and $b$, connected by their relation $R$ and its inverse $R^{-1}$, LLMs excel in handling sequences in the form of$aRb$,'' but encounter challenges when processing $bR^{-1}a$,'' whether in generation or comprehension. For instance, GPT-4 can accurately respond to the queryTom Cruise's mother is?'' with Mary Lee Pfeiffer,'' but it struggles to provide a satisfactory answer when askedMary Lee Pfeiffer's son is?'' In this paper, we undertake the first-ever study of how the reversal curse happens in LLMs. Our investigations reveal that the reversal curse can stem from the specific training objectives, which become particularly evident in the widespread use of next-token prediction within most causal LLMs. We hope this initial investigation can draw more attention to the reversal curse, as well as other underlying limitations in current LLMs.