Hippolyte Pilchen
forename.lastname@gmail.com
Hello! I am a first-year PhD student in AI, pursuing a joint project between Kyutai—co-advised by Patrick Pérez and Edouard Grave—and the Université Grenoble Alpes - LIGLAB, co-advised by Maxime Peyrard and Eric Gaussier.
My research focuses on Natural Language Processing (NLP). I am particularly interested in methods that enhance LLM efficiency, such as context compression and auto-encoders. Additionally, my work investigates memorization dynamics and temporal alignment in language models. Overall, I aim to design language models that are both modular in their knowledge storage and highly efficient.
Previously, I graduated from the MVA (Mathematics, Vision, and Learning) Master’s program at ENS Paris-Saclay in September 2024. Prior to that, I earned an engineering degree from École Polytechnique, specializing in applied mathematics.
news
| May 27, 2026 | Our paper “Understanding Data Temporality Impact on Large Language Models Pre-training” has been accepted at ICML 2026! We introduce KairosQA, a dataset for evaluating temporal knowledge in LLMs, along with Sequential Helium, a 6B model trained to better capture temporal dynamics. The project page, code, and resources are available here. |
|---|---|
| Dec 01, 2025 | I am thrilled to announce that I have started my PhD between Kyutai and the LIG, Université Grenoble Alpes, supervised by Patrick Pérez, Edouard Grave, Eric Gaussier and Maxime Peyrard. The thesis title is “Mastering Information in Language Models: Modularity and Efficiency”. |
| Apr 20, 2025 | Kyutai has released a repository to which I actively contributed. It open-sources code for fine-tuning Moshi, our speech-to-speech foundation model, on any dialogue dataset. The project is available at moshi-finetune. |
| Sep 09, 2024 | I’m excited to announce that I’ll be starting a short-term contract with Kyutai. I’m truly looking forward to contributing to their efforts in advancing open science and more specially AI. |
| Apr 15, 2024 | I am starting a research internship at Gatsby Computational Neuroscience Unit, UCL, to work on diffusion models. |