13 papers · 1 filter
TrustMI: Causally controlling how assistants trust their users
Théo Lasnier, Romain Froger, Maxence Lasbordes +1
Large Language Model (LLM) assistants routinely decide whether they can trust users and third parties whose competence, intentions, and integrity they cannot verify. This uncertain…
Small Agents with Semantic Search: Efficient Multilingual Code Localization
Maxence Lasbordes, Aarush Sinha, Raphael Sourty +2
Locating relevant files from natural-language requests is a core subtask for agents operating over code repositories. We investigate whether this task can be delegated to compact,…
LLM Forensics: Where Do Backdoors Hide? Localizing and Controlling Trigger Mechanisms with Sparse Autoencoders
Wissam Antoun, Francis Kulumba, Théo Lasnier +2
Even though backdoors in LLMs have been a growing concern, their inner workings are still under heavy scrutiny. Trigger-based backdoors are easy to define behaviorally, a rare inpu…
When Trivia Is Not Trivial: Everyday Knowledge Failures in Multilingual LLMs
Anna Mosolova, Djamé Seddah
Quiz rooms, trivia nights, and quiz shows challenge human knowledge across a wide range of topics, from canonical facts to everyday culture. In this paper, we examine whether large…
Multi-Hop Knowledge Composition is Bound by Pretraining Exposure
Yannis Karmim, Luis Marti, Djamé Seddah +1
Large Language Models fail at implicit multi-hop reasoning: a model answers "When was born?" and "Who is 's closest friend?" correctly but fails on "When was 's closest f…
Backdoor Unlearning Generalization: A Path Toward the Removal of Unknown Triggers in LLMs
Lisa Bouger, Théo Lasnier, Philippe Loubet Moundi +2
Backdoor attacks in Large Language Models (LLMs) are a growing security concern, where models can generate adversary-chosen content. Existing defenses target backdoors one at a tim…
Language-Switching Triggers Take a Latent Detour Through Language Models
Francis Kulumba, Wissam Antoun, Théo Lasnier +2
Backdoor attacks on language models pose a growing security concern, yet the internal mechanisms by which a trigger sequence hijacks model computations remain poorly understood. We…
Rethinking the Multilingual Reasoning Gap with Layer Swap
Maxence Lasbordes, Amélie Chatelain, Djamé Seddah
Recent reasoning Large Language Models produce a chain-of-thought (CoT) predominantly in English, even when prompted in non-English languages. Prior work suggests that forcing the…
Leveraging Wikidata for Geographically Informed Sociocultural Bias Dataset Creation: Application to Latin America
Yannis Karmim, Renato Pino, Hernan Contreras +6
Large Language Models (LLMs) exhibit inequalities with respect to various cultural contexts. Most prominent open-weights models are trained on Global North data and show prejudicia…
When Tables Go Crazy: Evaluating Multimodal Models on French Financial Documents
Virginie Mouilleron, Théo Lasnier, Anna Mosolova +1
Vision-language models (VLMs) perform well on many document understanding tasks, yet their reliability in specialized, non-English domains remains underexplored. This gap is especi…
Language Triggers Hijack Language Circuits: A Mechanistic Analysis of Backdoor Behaviors in Large Language Models
Théo Lasnier, Wissam Antoun, Francis Kulumba +2
Backdoor attacks pose significant security risks for Large Language Models (LLMs), yet the internal mechanisms by which triggers operate remain poorly understood. We present the fi…
Translation Heads: Disentangling meaning from language in LLM-based machine translation
Théo Lasnier, Armel Zebaze, Djamé Seddah +2
Mechanistic Interpretability (MI) seeks to explain how neural networks implement their capabilities, but the scale of Large Language Models (LLMs) has limited prior MI work in Mach…
Multilingual, Multimodal Pipeline for Creating Authentic and Structured Fact-Checked Claim Dataset
Z. Melce Hüsünbeyi, Virginie Mouilleron, Leonie Uhling +3
The rapid proliferation of misinformation across online platforms underscores the urgent need for robust, up-to-date, explainable, and multilingual fact-checking resources. However…