The Illusion of Latent Generalization: Bi-directionality and the Reversal Curse
The reversal curse describes a failure of autoregressive language models to retrieve a fact in reverse order (e.g., training on `$A > B$'' but failing on $B < A$''). Recent work shows that objectives with bidirectional supervision (e.g., bidirectional attention or masking-based reconstruction for decoder-only models) can mitigate the reversal curse. We extend this evaluation to include a vanilla masked language modeling (MLM) objective and compare it to decoder-only masking-based training across four reversal benchmarks and then provide a minimal mechanistic study of \emph{how} these objectives succeed. We show that reversal accuracy requires training signal that explicitly makes the source entity a prediction target, and we find little evidence that success corresponds to a single direction-agnostic representation of a fact. Instead, representation distances and linear probes are consistent with storing forward and reverse directions as distinct entries, with different indexing geometry for MLM versus decoder-only masking-based training. Our results caution that objective-level `fixes'' can improve reversal behavior without necessarily inducing the kind of latent generalization one might expect from a unified concept.
Code (0)
등록된 구현이 없습니다.
Similar Papers 제목 키워드 기반
Is the Reversal Curse a Binding Problem? Uncovering Limitations of Transformers from a Basic Generalization Failure
Despite their impressive capabilities, LLMs exhibit a basic generalization failure known as the Reversal Curse, where they struggle to learn reversible factual associations. Understanding why this occurs could help ident…
Arithmetic ReasoningData AugmentationArrowGEV: Grounding Events in Video via Learning the Arrow of Time
Grounding events in videos serves as a fundamental capability in video analysis. While Vision Language Models (VLMs) are increasingly employed for this task, existing approaches predominantly train models to associate ev…
Reinforcement LearningThe Reversal Curse: LLMs trained on "A is B" fail to learn "B is A"
We expose a surprising failure of generalization in auto-regressive large language models (LLMs). If a model is trained on a sentence of the form "A is B", it will not automatically generalize to the reverse direction "B…
Data AugmentationSentenceAn Analysis and Mitigation of the Reversal Curse
Recent research observed a noteworthy phenomenon in large language models (LLMs), referred to as the ``reversal curse.'' The reversal curse is that when dealing with two entities, denoted as $a$ and $b$, connected by the…
DenoisingLanguage ModellingImproving Latent Generalization Using Test-time Compute
Language Models (LMs) exhibit two distinct mechanisms for knowledge acquisition: in-weights learning (i.e., encoding information within the model weights) and in-context learning (ICL). Although these two modes offer com…
Reinforcement LearningData Augmentation