Saved in:
Bibliographic Details
Main Authors: Pęzik, Piotr, Kaczyński, Konrad, Szymańska, Maria, Żarnecki, Filip, Deckert, Zuzanna, Kwiatkowski, Jakub, Janowski, Wojciech
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2511.12116
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915620212178944
author Pęzik, Piotr
Kaczyński, Konrad
Szymańska, Maria
Żarnecki, Filip
Deckert, Zuzanna
Kwiatkowski, Jakub
Janowski, Wojciech
author_facet Pęzik, Piotr
Kaczyński, Konrad
Szymańska, Maria
Żarnecki, Filip
Deckert, Zuzanna
Kwiatkowski, Jakub
Janowski, Wojciech
contents Large Language Models (LLMs) are pretrained on textual data up to a specific temporal cutoff. This creates a strict knowledge boundary beyond which models cannot provide accurate information without querying external sources. More subtly, when this limitation is unknown or ignored, LLMs may inadvertently blend outdated time-sensitive information with general knowledge during reasoning tasks, potentially compromising response accuracy. We introduce LLMLagBench, an LLM freshness benchmark, as a systematic approach for identifying the earliest probable temporal boundaries of an LLM's training data by evaluating its knowledge of recent events. We then apply this benchmark to evaluate a large set of LLMs, including models with both explicitly declared and undeclared training cutoffs. The reliability of the benchmark is assessed by manual validation and comparison with publicly released information about LLM pretraining.
format Preprint
id arxiv_https___arxiv_org_abs_2511_12116
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle LLMLagBench: Identifying Temporal Training Boundaries in Large Language Models
Pęzik, Piotr
Kaczyński, Konrad
Szymańska, Maria
Żarnecki, Filip
Deckert, Zuzanna
Kwiatkowski, Jakub
Janowski, Wojciech
Computation and Language
Artificial Intelligence
Large Language Models (LLMs) are pretrained on textual data up to a specific temporal cutoff. This creates a strict knowledge boundary beyond which models cannot provide accurate information without querying external sources. More subtly, when this limitation is unknown or ignored, LLMs may inadvertently blend outdated time-sensitive information with general knowledge during reasoning tasks, potentially compromising response accuracy. We introduce LLMLagBench, an LLM freshness benchmark, as a systematic approach for identifying the earliest probable temporal boundaries of an LLM's training data by evaluating its knowledge of recent events. We then apply this benchmark to evaluate a large set of LLMs, including models with both explicitly declared and undeclared training cutoffs. The reliability of the benchmark is assessed by manual validation and comparison with publicly released information about LLM pretraining.
title LLMLagBench: Identifying Temporal Training Boundaries in Large Language Models
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2511.12116