@inproceedings{vladika-etal-2025-facts,
title = "Facts Fade Fast: Evaluating Memorization of Outdated Medical Knowledge in Large Language Models",
author = "Vladika, Juraj and
Dhaini, Mahdi and
Matthes, Florian",
editor = "Christodoulopoulos, Christos and
Chakraborty, Tanmoy and
Rose, Carolyn and
Peng, Violet",
booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2025",
month = nov,
year = "2025",
address = "Suzhou, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.findings-emnlp.487/",
doi = "10.18653/v1/2025.findings-emnlp.487",
pages = "9161--9174",
ISBN = "979-8-89176-335-7",
abstract = "The growing capabilities of Large Language Models (LLMs) can enhance healthcare by assisting medical researchers, physicians, and improving access to health services for patients. LLMs encode extensive knowledge within their parameters, including medical knowledge derived from many sources. However, the knowledge in LLMs can become outdated over time, posing challenges in keeping up with evolving medical recommendations and research. This can lead to LLMs providing outdated health advice or failures in medical reasoning tasks. To address this gap, our study introduces two novel biomedical question-answering (QA) datasets derived from medical systematic literature reviews: MedRevQA, a general dataset of 16,501 biomedical QA pairs, and MedChangeQA, a subset of 512 QA pairs whose verdict changed though time. By evaluating the performance of eight popular LLMs, we find that all models exhibit memorization of outdated knowledge to some extent. We provide deeper insights and analysis, paving the way for future research on this challenging aspect of LLMs."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="vladika-etal-2025-facts">
<titleInfo>
<title>Facts Fade Fast: Evaluating Memorization of Outdated Medical Knowledge in Large Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Juraj</namePart>
<namePart type="family">Vladika</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mahdi</namePart>
<namePart type="family">Dhaini</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Florian</namePart>
<namePart type="family">Matthes</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2025</title>
</titleInfo>
<name type="personal">
<namePart type="given">Christos</namePart>
<namePart type="family">Christodoulopoulos</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tanmoy</namePart>
<namePart type="family">Chakraborty</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Carolyn</namePart>
<namePart type="family">Rose</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Violet</namePart>
<namePart type="family">Peng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Suzhou, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-335-7</identifier>
</relatedItem>
<abstract>The growing capabilities of Large Language Models (LLMs) can enhance healthcare by assisting medical researchers, physicians, and improving access to health services for patients. LLMs encode extensive knowledge within their parameters, including medical knowledge derived from many sources. However, the knowledge in LLMs can become outdated over time, posing challenges in keeping up with evolving medical recommendations and research. This can lead to LLMs providing outdated health advice or failures in medical reasoning tasks. To address this gap, our study introduces two novel biomedical question-answering (QA) datasets derived from medical systematic literature reviews: MedRevQA, a general dataset of 16,501 biomedical QA pairs, and MedChangeQA, a subset of 512 QA pairs whose verdict changed though time. By evaluating the performance of eight popular LLMs, we find that all models exhibit memorization of outdated knowledge to some extent. We provide deeper insights and analysis, paving the way for future research on this challenging aspect of LLMs.</abstract>
<identifier type="citekey">vladika-etal-2025-facts</identifier>
<identifier type="doi">10.18653/v1/2025.findings-emnlp.487</identifier>
<location>
<url>https://aclanthology.org/2025.findings-emnlp.487/</url>
</location>
<part>
<date>2025-11</date>
<extent unit="page">
<start>9161</start>
<end>9174</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Facts Fade Fast: Evaluating Memorization of Outdated Medical Knowledge in Large Language Models
%A Vladika, Juraj
%A Dhaini, Mahdi
%A Matthes, Florian
%Y Christodoulopoulos, Christos
%Y Chakraborty, Tanmoy
%Y Rose, Carolyn
%Y Peng, Violet
%S Findings of the Association for Computational Linguistics: EMNLP 2025
%D 2025
%8 November
%I Association for Computational Linguistics
%C Suzhou, China
%@ 979-8-89176-335-7
%F vladika-etal-2025-facts
%X The growing capabilities of Large Language Models (LLMs) can enhance healthcare by assisting medical researchers, physicians, and improving access to health services for patients. LLMs encode extensive knowledge within their parameters, including medical knowledge derived from many sources. However, the knowledge in LLMs can become outdated over time, posing challenges in keeping up with evolving medical recommendations and research. This can lead to LLMs providing outdated health advice or failures in medical reasoning tasks. To address this gap, our study introduces two novel biomedical question-answering (QA) datasets derived from medical systematic literature reviews: MedRevQA, a general dataset of 16,501 biomedical QA pairs, and MedChangeQA, a subset of 512 QA pairs whose verdict changed though time. By evaluating the performance of eight popular LLMs, we find that all models exhibit memorization of outdated knowledge to some extent. We provide deeper insights and analysis, paving the way for future research on this challenging aspect of LLMs.
%R 10.18653/v1/2025.findings-emnlp.487
%U https://aclanthology.org/2025.findings-emnlp.487/
%U https://doi.org/10.18653/v1/2025.findings-emnlp.487
%P 9161-9174
Markdown (Informal)
[Facts Fade Fast: Evaluating Memorization of Outdated Medical Knowledge in Large Language Models](https://aclanthology.org/2025.findings-emnlp.487/) (Vladika et al., Findings 2025)
ACL