@inproceedings{dubois-etal-2025-mosaic-multiple,
title = "{MOSAIC}: Multiple Observers Spotting {AI} Content",
author = "Dubois, Matthieu and
Yvon, Fran{\c{c}}ois and
Piantanida, Pablo",
editor = "Che, Wanxiang and
Nabende, Joyce and
Shutova, Ekaterina and
Pilehvar, Mohammad Taher",
booktitle = "Findings of the Association for Computational Linguistics: ACL 2025",
month = jul,
year = "2025",
address = "Vienna, Austria",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.findings-acl.1244/",
doi = "10.18653/v1/2025.findings-acl.1244",
pages = "24230--24247",
ISBN = "979-8-89176-256-5",
abstract = "The dissemination of Large Language Models (LLMs), trained at scale, and endowed with powerful text-generating abilities, has made it easier for all to produce harmful, toxic, faked or forged content. In response, various proposals have been made to automatically discriminate artificially generated from human-written texts, typically framing the problem as a binary classification problem. Early approaches evaluate an input document with a well-chosen detector LLM, assuming that low-perplexity scores reliably signal machine-made content. More recent systems instead consider two LLMs and compare their probability distributions over the document to further discriminate when perplexity alone cannot. However, using a fixed pair of models can induce brittleness in performance. We extend these approaches to the ensembling of several LLMs and derive a new, theoretically grounded approach to combine their respective strengths. Our experiments, using a variety of generator LLMs, suggest that this approach effectively harnesses each model{'}s capabilities, leading to strong detection performance on a variety of domains."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="dubois-etal-2025-mosaic-multiple">
<titleInfo>
<title>MOSAIC: Multiple Observers Spotting AI Content</title>
</titleInfo>
<name type="personal">
<namePart type="given">Matthieu</namePart>
<namePart type="family">Dubois</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">François</namePart>
<namePart type="family">Yvon</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Pablo</namePart>
<namePart type="family">Piantanida</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: ACL 2025</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wanxiang</namePart>
<namePart type="family">Che</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joyce</namePart>
<namePart type="family">Nabende</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ekaterina</namePart>
<namePart type="family">Shutova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mohammad</namePart>
<namePart type="given">Taher</namePart>
<namePart type="family">Pilehvar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Vienna, Austria</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-256-5</identifier>
</relatedItem>
<abstract>The dissemination of Large Language Models (LLMs), trained at scale, and endowed with powerful text-generating abilities, has made it easier for all to produce harmful, toxic, faked or forged content. In response, various proposals have been made to automatically discriminate artificially generated from human-written texts, typically framing the problem as a binary classification problem. Early approaches evaluate an input document with a well-chosen detector LLM, assuming that low-perplexity scores reliably signal machine-made content. More recent systems instead consider two LLMs and compare their probability distributions over the document to further discriminate when perplexity alone cannot. However, using a fixed pair of models can induce brittleness in performance. We extend these approaches to the ensembling of several LLMs and derive a new, theoretically grounded approach to combine their respective strengths. Our experiments, using a variety of generator LLMs, suggest that this approach effectively harnesses each model’s capabilities, leading to strong detection performance on a variety of domains.</abstract>
<identifier type="citekey">dubois-etal-2025-mosaic-multiple</identifier>
<identifier type="doi">10.18653/v1/2025.findings-acl.1244</identifier>
<location>
<url>https://aclanthology.org/2025.findings-acl.1244/</url>
</location>
<part>
<date>2025-07</date>
<extent unit="page">
<start>24230</start>
<end>24247</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T MOSAIC: Multiple Observers Spotting AI Content
%A Dubois, Matthieu
%A Yvon, François
%A Piantanida, Pablo
%Y Che, Wanxiang
%Y Nabende, Joyce
%Y Shutova, Ekaterina
%Y Pilehvar, Mohammad Taher
%S Findings of the Association for Computational Linguistics: ACL 2025
%D 2025
%8 July
%I Association for Computational Linguistics
%C Vienna, Austria
%@ 979-8-89176-256-5
%F dubois-etal-2025-mosaic-multiple
%X The dissemination of Large Language Models (LLMs), trained at scale, and endowed with powerful text-generating abilities, has made it easier for all to produce harmful, toxic, faked or forged content. In response, various proposals have been made to automatically discriminate artificially generated from human-written texts, typically framing the problem as a binary classification problem. Early approaches evaluate an input document with a well-chosen detector LLM, assuming that low-perplexity scores reliably signal machine-made content. More recent systems instead consider two LLMs and compare their probability distributions over the document to further discriminate when perplexity alone cannot. However, using a fixed pair of models can induce brittleness in performance. We extend these approaches to the ensembling of several LLMs and derive a new, theoretically grounded approach to combine their respective strengths. Our experiments, using a variety of generator LLMs, suggest that this approach effectively harnesses each model’s capabilities, leading to strong detection performance on a variety of domains.
%R 10.18653/v1/2025.findings-acl.1244
%U https://aclanthology.org/2025.findings-acl.1244/
%U https://doi.org/10.18653/v1/2025.findings-acl.1244
%P 24230-24247
Markdown (Informal)
[MOSAIC: Multiple Observers Spotting AI Content](https://aclanthology.org/2025.findings-acl.1244/) (Dubois et al., Findings 2025)
ACL
- Matthieu Dubois, François Yvon, and Pablo Piantanida. 2025. MOSAIC: Multiple Observers Spotting AI Content. In Findings of the Association for Computational Linguistics: ACL 2025, pages 24230–24247, Vienna, Austria. Association for Computational Linguistics.