{"ID":2897280,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2507.06445","arxiv_id":"2507.06445","title":"Can Interpretation Predict Behavior on Unseen Data?","abstract":"Interpretability research often aims to predict how a model will respond to targeted interventions on specific mechanisms. However, it rarely predicts how a model will respond to unseen input data. This paper explores the promises and challenges of interpretability as a tool for predicting out-of-distribution (OOD) model behavior. Specifically, we investigate the correspondence between attention patterns and OOD generalization in hundreds of Transformer models independently trained on a synthetic classification task. These models exhibit several distinct systematic generalization rules OOD, forming a diverse population for correlational analysis. In this setting, we find that simple observational tools from interpretability can predict OOD performance. In particular, when in-distribution attention exhibits hierarchical patterns, the model is likely to generalize hierarchically on OOD data -- even when the rule's implementation does not rely on these hierarchical patterns, according to ablation tests. Our findings offer a proof-of-concept to motivate further interpretability work on predicting unseen model behavior.","short_abstract":"Interpretability research often aims to predict how a model will respond to targeted interventions on specific mechanisms. However, it rarely predicts how a model will respond to unseen input data. This paper explores the promises and challenges of interpretability as a tool for predicting out-of-distribution (OOD) mod...","url_abs":"https://arxiv.org/abs/2507.06445","url_pdf":"https://arxiv.org/pdf/2507.06445v1","authors":"[\"Victoria R. Li\",\"Jenny Kaufmann\",\"Martin Wattenberg\",\"David Alvarez-Melis\",\"Naomi Saphra\"]","published":"2025-07-08T23:07:33Z","proceeding":"cs.LG","tasks":"[\"cs.LG\",\"cs.AI\",\"cs.CL\"]","methods":"[\"Transformer\"]","has_code":false}