AbsenceBench: Language Models Can't Tell What's Missing

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Fu, Harvey Yiyun, Shrivastava, Aryan, Moore, Jared, West, Peter, Tan, Chenhao, Holtzman, Ari
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866911003577417728
author Fu, Harvey Yiyun
Shrivastava, Aryan
Moore, Jared
West, Peter
Tan, Chenhao
Holtzman, Ari
author_facet Fu, Harvey Yiyun
Shrivastava, Aryan
Moore, Jared
West, Peter
Tan, Chenhao
Holtzman, Ari
contents Large language models (LLMs) are increasingly capable of processing long inputs and locating specific information within them, as evidenced by their performance on the Needle in a Haystack (NIAH) test. However, while models excel at recalling surprising information, they still struggle to identify clearly omitted information. We introduce AbsenceBench to assesses LLMs' capacity to detect missing information across three domains: numerical sequences, poetry, and GitHub pull requests. AbsenceBench asks models to identify which pieces of a document were deliberately removed, given access to both the original and edited contexts. Despite the apparent straightforwardness of these tasks, our experiments reveal that even state-of-the-art models like Claude-3.7-Sonnet achieve only 69.6% F1-score with a modest average context length of 5K tokens. Our analysis suggests this poor performance stems from a fundamental limitation: Transformer attention mechanisms cannot easily attend to "gaps" in documents since these absences don't correspond to any specific keys that can be attended to. Overall, our results and analysis provide a case study of the close proximity of tasks where models are already superhuman (NIAH) and tasks where models breakdown unexpectedly (AbsenceBench).
format Preprint
id arxiv_https___arxiv_org_abs_2506_11440
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle AbsenceBench: Language Models Can't Tell What's Missing
Fu, Harvey Yiyun
Shrivastava, Aryan
Moore, Jared
West, Peter
Tan, Chenhao
Holtzman, Ari
Computation and Language
Large language models (LLMs) are increasingly capable of processing long inputs and locating specific information within them, as evidenced by their performance on the Needle in a Haystack (NIAH) test. However, while models excel at recalling surprising information, they still struggle to identify clearly omitted information. We introduce AbsenceBench to assesses LLMs' capacity to detect missing information across three domains: numerical sequences, poetry, and GitHub pull requests. AbsenceBench asks models to identify which pieces of a document were deliberately removed, given access to both the original and edited contexts. Despite the apparent straightforwardness of these tasks, our experiments reveal that even state-of-the-art models like Claude-3.7-Sonnet achieve only 69.6% F1-score with a modest average context length of 5K tokens. Our analysis suggests this poor performance stems from a fundamental limitation: Transformer attention mechanisms cannot easily attend to "gaps" in documents since these absences don't correspond to any specific keys that can be attended to. Overall, our results and analysis provide a case study of the close proximity of tasks where models are already superhuman (NIAH) and tasks where models breakdown unexpectedly (AbsenceBench).
title AbsenceBench: Language Models Can't Tell What's Missing
topic Computation and Language
url https://arxiv.org/abs/2506.11440