8 papers
Validating Causal Abstraction Metrics on Simulated Complex Systems
Maxime Méloux, Tiago Pimentel, François Portet +1
A central goal of science is to produce valid explanations of complex systems: high-level causal accounts that faithfully reflect the behavior of lower-level mechanisms. Yet no con…
What Makes an LLM a Good Optimizer? A Trajectory Analysis of LLM-Guided Evolutionary Search
Xinhao Zhang, Xi Chen, François Portet +1
Recent work has demonstrated the promise of orchestrating large language models (LLMs) within evolutionary and agentic optimization systems. However, the mechanisms driving these o…
What Matters to an LLM? Behavioral and Computational Evidences from Summarization
Yongxin Zhou, Changshun Wu, Philippe Mulhem +2
Large Language Models (LLMs) are now state-of-the-art at summarization, yet the internal notion of importance that drives their information selections remains hidden. We propose to…
Pantagruel: Unified Self-Supervised Encoders for French Text and Speech
Phuong-Hang Le, Valentin Pelloin, Arnault Chatelain +27
We release Pantagruel models, a new family of self-supervised encoder models for French text and speech. Instead of predicting modality-tailored targets such as textual tokens or s…
The Dead Salmons of AI Interpretability
Maxime Méloux, Giada Dirupo, François Portet +1
In a striking neuroscience study, the authors placed a dead salmon in an MRI scanner and showed it images of humans in social situations. Astonishingly, standard analyses of the ti…
MIST: Mutual Information Estimation Via Supervised Training
German Gritsai, Megan Richards, Maxime Méloux +2
We propose a fully data-driven approach to designing mutual information (MI) estimators. Since any MI estimator is a function of the observed sample from two random variables, we p…
Mechanistic Interpretability as Statistical Estimation: A Variance Analysis
Maxime Méloux, François Portet, Maxime Peyrard
Mechanistic Interpretability (MI) aims to reverse-engineer model behaviors by identifying functional sub-networks. Yet, the scientific validity of these findings depends on their s…
Everything, Everywhere, All at Once: Is Mechanistic Interpretability Identifiable?
Maxime Méloux, Silviu Maniu, François Portet +1
As AI systems are used in high-stakes applications, ensuring interpretability is crucial. Mechanistic Interpretability (MI) aims to reverse-engineer neural networks by extracting h…