@inproceedings{fu-etal-2025-quantized,
title = "Quantized but Deceptive? A Multi-Dimensional Truthfulness Evaluation of Quantized {LLM}s",
author = "Fu, Yao and
Long, Xianxuan and
Li, Runchao and
Yu, Haotian and
Sheng, Mu and
Han, Xiaotian and
Yin, Yu and
Li, Pan",
editor = "Christodoulopoulos, Christos and
Chakraborty, Tanmoy and
Rose, Carolyn and
Peng, Violet",
booktitle = "Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing",
month = nov,
year = "2025",
address = "Suzhou, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.emnlp-main.1548/",
doi = "10.18653/v1/2025.emnlp-main.1548",
pages = "30435--30458",
ISBN = "979-8-89176-332-6",
abstract = "Quantization enables efficient deployment of large language models (LLMs) in resource-constrained environments by significantly reducing memory and computation costs. While quantized LLMs often maintain performance on perplexity and zero-shot tasks, their impact on truthfulness{---}whether generating truthful or deceptive responses{---}remains largely unexplored. In this work, we introduce TruthfulnessEval, a comprehensive evaluation framework for assessing the truthfulness of quantized LLMs across three dimensions: (1) Truthfulness on Logical Reasoning; (2) Truthfulness on Common Sense; and (3) Truthfulness on Imitative Falsehoods. Using this framework, we examine mainstream quantization techniques (ranging from 4-bit to extreme 2-bit) across several open-source LLMs. Surprisingly, we find that while quantized models retain internally truthful representations, they are more susceptible to producing false outputs under misleading prompts. To probe this vulnerability, we test 15 rephrased variants of ``honest'', ``neutral'' and ``deceptive'' prompts and observe that ``deceptive'' prompts can override truth-consistent behavior, whereas ``honest'' and ``neutral'' prompts maintain stable outputs. Further, we reveal that quantized models ``know'' the truth internally yet still produce false outputs when guided by ``deceptive'' prompts via layer-wise probing and PCA visualizations. Our findings provide insights into future designs of quantization-aware alignment and truthfulness interventions."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="fu-etal-2025-quantized">
<titleInfo>
<title>Quantized but Deceptive? A Multi-Dimensional Truthfulness Evaluation of Quantized LLMs</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yao</namePart>
<namePart type="family">Fu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xianxuan</namePart>
<namePart type="family">Long</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Runchao</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Haotian</namePart>
<namePart type="family">Yu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mu</namePart>
<namePart type="family">Sheng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiaotian</namePart>
<namePart type="family">Han</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yu</namePart>
<namePart type="family">Yin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Pan</namePart>
<namePart type="family">Li</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>Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing</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">979-8-89176-332-6</identifier>
</relatedItem>
<abstract>Quantization enables efficient deployment of large language models (LLMs) in resource-constrained environments by significantly reducing memory and computation costs. While quantized LLMs often maintain performance on perplexity and zero-shot tasks, their impact on truthfulness—whether generating truthful or deceptive responses—remains largely unexplored. In this work, we introduce TruthfulnessEval, a comprehensive evaluation framework for assessing the truthfulness of quantized LLMs across three dimensions: (1) Truthfulness on Logical Reasoning; (2) Truthfulness on Common Sense; and (3) Truthfulness on Imitative Falsehoods. Using this framework, we examine mainstream quantization techniques (ranging from 4-bit to extreme 2-bit) across several open-source LLMs. Surprisingly, we find that while quantized models retain internally truthful representations, they are more susceptible to producing false outputs under misleading prompts. To probe this vulnerability, we test 15 rephrased variants of “honest”, “neutral” and “deceptive” prompts and observe that “deceptive” prompts can override truth-consistent behavior, whereas “honest” and “neutral” prompts maintain stable outputs. Further, we reveal that quantized models “know” the truth internally yet still produce false outputs when guided by “deceptive” prompts via layer-wise probing and PCA visualizations. Our findings provide insights into future designs of quantization-aware alignment and truthfulness interventions.</abstract>
<identifier type="citekey">fu-etal-2025-quantized</identifier>
<identifier type="doi">10.18653/v1/2025.emnlp-main.1548</identifier>
<location>
<url>https://aclanthology.org/2025.emnlp-main.1548/</url>
</location>
<part>
<date>2025-11</date>
<extent unit="page">
<start>30435</start>
<end>30458</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Quantized but Deceptive? A Multi-Dimensional Truthfulness Evaluation of Quantized LLMs
%A Fu, Yao
%A Long, Xianxuan
%A Li, Runchao
%A Yu, Haotian
%A Sheng, Mu
%A Han, Xiaotian
%A Yin, Yu
%A Li, Pan
%Y Christodoulopoulos, Christos
%Y Chakraborty, Tanmoy
%Y Rose, Carolyn
%Y Peng, Violet
%S Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing
%D 2025
%8 November
%I Association for Computational Linguistics
%C Suzhou, China
%@ 979-8-89176-332-6
%F fu-etal-2025-quantized
%X Quantization enables efficient deployment of large language models (LLMs) in resource-constrained environments by significantly reducing memory and computation costs. While quantized LLMs often maintain performance on perplexity and zero-shot tasks, their impact on truthfulness—whether generating truthful or deceptive responses—remains largely unexplored. In this work, we introduce TruthfulnessEval, a comprehensive evaluation framework for assessing the truthfulness of quantized LLMs across three dimensions: (1) Truthfulness on Logical Reasoning; (2) Truthfulness on Common Sense; and (3) Truthfulness on Imitative Falsehoods. Using this framework, we examine mainstream quantization techniques (ranging from 4-bit to extreme 2-bit) across several open-source LLMs. Surprisingly, we find that while quantized models retain internally truthful representations, they are more susceptible to producing false outputs under misleading prompts. To probe this vulnerability, we test 15 rephrased variants of “honest”, “neutral” and “deceptive” prompts and observe that “deceptive” prompts can override truth-consistent behavior, whereas “honest” and “neutral” prompts maintain stable outputs. Further, we reveal that quantized models “know” the truth internally yet still produce false outputs when guided by “deceptive” prompts via layer-wise probing and PCA visualizations. Our findings provide insights into future designs of quantization-aware alignment and truthfulness interventions.
%R 10.18653/v1/2025.emnlp-main.1548
%U https://aclanthology.org/2025.emnlp-main.1548/
%U https://doi.org/10.18653/v1/2025.emnlp-main.1548
%P 30435-30458
Markdown (Informal)
[Quantized but Deceptive? A Multi-Dimensional Truthfulness Evaluation of Quantized LLMs](https://aclanthology.org/2025.emnlp-main.1548/) (Fu et al., EMNLP 2025)
ACL
- Yao Fu, Xianxuan Long, Runchao Li, Haotian Yu, Mu Sheng, Xiaotian Han, Yu Yin, and Pan Li. 2025. Quantized but Deceptive? A Multi-Dimensional Truthfulness Evaluation of Quantized LLMs. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 30435–30458, Suzhou, China. Association for Computational Linguistics.