Author

Théo Lasnier

Observed name: Lasnier, Théo

Institutions: Institut national de recherche en sciences et technologies du numérique, Sorbonne Université, Centre Inria de Paris

Observed institutions: Institut national de recherche en sciences et technologies du numérique, Sorbonne Université, Centre Inria de Paris

OpenAlex ID: A5125225847

TrustMI: Causally controlling how assistants trust their users

2026 · Lasnier, Théo, Froger, Romain, et al. · arXiv (Cornell University)

LLM Forensics: Where Do Backdoors Hide? Localizing and Controlling Trigger Mechanisms with Sparse Autoencoders

2026 · Antoun, Wissam, Kulumba, Francis, et al. · arXiv (Cornell University)

Backdoor Unlearning Generalization: A Path Toward the Removal of Unknown Triggers in LLMs

2026 · Bouger, Lisa, Lasnier, Théo, et al. · arXiv (Cornell University)

Language-Switching Triggers Take a Latent Detour Through Language Models

2026 · Kulumba, Francis, Antoun, Wissam, et al. · arXiv (Cornell University)

When Tables Go Crazy: Evaluating Multimodal Models on French Financial Documents

2026 · Virginie Mouilleron, Théo Lasnier, et al. · Language Resources and Evaluation Conference (LREC)

When Tables Go Crazy: Evaluating Multimodal Models on French Financial Documents

2026 · Mouilleron, Virginie, Lasnier, Théo, et al. · HAL (Le Centre pour la Communication Scientifique Directe)

Language Triggers Hijack Language Circuits: A Mechanistic Analysis of Backdoor Behaviors in Large Language Models

2026 · Lasnier, Théo, Antoun, Wissam, et al. · arXiv (Cornell University)

When Tables Go Crazy: Evaluating Multimodal Models on French Financial Documents

2026 · Mouilleron, Virginie, Lasnier, Théo, et al. · arXiv (Cornell University)

Translation Heads: Disentangling meaning from language in LLM-based machine translation

2026 · Lasnier, Théo, Zebaze, Armel, et al. · arXiv (Cornell University)

Works count: 9

Citation count:

H-index:

I10-index: