TY - GEN

T1 - Temporal Blind Spots in Large Language Models

AU - Wallat, Jonas

AU - Jatowt, Adam

AU - Anand, Avishek

N1 - Funding Information: This research was partially funded by the Federal Ministry of Education and Research (BMBF), Germany under the project LeibnizKILabor with grant No. 01DD20003 and Cubra with grant No. 13N16052

PY - 2024/3/4

Y1 - 2024/3/4

N2 - Large language models (LLMs) have recently gained significant attention due to their unparalleled zero-shot performance on various natural language processing tasks. However, the pre-Training data utilized in LLMs is often confined to a specific corpus, resulting in inherent freshness and temporal scope limitations. Consequently, this raises concerns regarding the effectiveness of LLMs for tasks involving temporal intents. In this study, we aim to investigate the underlying limitations of general-purpose LLMs when deployed for tasks that require a temporal understanding. We pay particular attention to handling factual temporal knowledge through three popular temporal QA datasets. Specifically, we observe low performance on detailed questions about the past and, surprisingly, for rather new information. In manual and automatic testing, we find multiple temporal errors and characterize the conditions under which QA performance deteriorates. Our analysis contributes to understanding LLM limitations and offers valuable insights into developing future models that can better cater to the demands of temporally-oriented tasks. The code is available https://github.com/jwallat/temporalblindspots.

AB - Large language models (LLMs) have recently gained significant attention due to their unparalleled zero-shot performance on various natural language processing tasks. However, the pre-Training data utilized in LLMs is often confined to a specific corpus, resulting in inherent freshness and temporal scope limitations. Consequently, this raises concerns regarding the effectiveness of LLMs for tasks involving temporal intents. In this study, we aim to investigate the underlying limitations of general-purpose LLMs when deployed for tasks that require a temporal understanding. We pay particular attention to handling factual temporal knowledge through three popular temporal QA datasets. Specifically, we observe low performance on detailed questions about the past and, surprisingly, for rather new information. In manual and automatic testing, we find multiple temporal errors and characterize the conditions under which QA performance deteriorates. Our analysis contributes to understanding LLM limitations and offers valuable insights into developing future models that can better cater to the demands of temporally-oriented tasks. The code is available https://github.com/jwallat/temporalblindspots.

KW - large language models

KW - question answering

KW - temporal information retrieval

KW - temporal query intents

UR - http://www.scopus.com/inward/record.url?scp=85191716137&partnerID=8YFLogxK

U2 - 10.48550/arXiv.2401.12078

DO - 10.48550/arXiv.2401.12078

M3 - Conference contribution

AN - SCOPUS:85191716137

SP - 683

EP - 692

BT - Proceedings of the 17th ACM International Conference on Web Search and Data Mining

T2 - 17th ACM International Conference on Web Search and Data Mining, WSDM 2024

Y2 - 4 March 2024 through 8 March 2024

ER -