I’m MATS Winter 2025 (7.0) Scholar with Neel Nanda. With Julian Minder, we’re currently working on evaluating different model diffingModel diffing is the study of mechanistic changes introduced during fine-tuning - essentially, understanding what makes a fine-tuned model different from its base model internally. methods.

I’m completing my MSc in Vision & Learning (MVA) at École Normale Supérieure Paris-Saclay. My main research interest is technical AI alignment.

Previously, I was fortunate to complete a research internship at EPFL DLAB under the amazing supervision of Chris Wendler and Bob West. My main focus was interpretability, and I worked on a follow-up to the “Do Llamas work in English?” paper. Our work was selected as a spotlight presentation at the Mechanistic Interpretability Workshop at ICML 2025.

Before starting this internship, I was exploring the emergence of XOR features in Large Language Models and the RAX hypothesis proposed by Sam Marks, did SPAR with Walter Laurito, worked on a project with Jobst Heitzig on non-maximizing training objectives for RL agents, and attended the ML4G AI safety bootcamp.

You can find my other projects on my CV or GitHub.

I am also interested in evolutionary biology, its manifestation in artificial life simulations, and I am an improvisor at the ENS improv theater club Lika. Feel free to message me if you want to get in touch!

News

Last updated: 26/06/2025