Theophile Gervet


I’m a research scientist at Mistral AI, working on accelerating AI open-source. I believe progress on language models as cultural technologies is currently our best lever on the rate of human progress. Come build with us!

Before that, I was doing an ML PhD at Carnegie Mellon University, working on computer vision and robotics with Katerina Fragkiadaki.

Even before that, I was working on reinforcement learning (RL) with Doina Precup and Joelle Pineau at McGill in Montreal. Early in my PhD, I worked on offline RL for education with Tom Mitchell. I took a couple of years of leave to build the ML product & team of Relyance AI.

I grew up in Paris, and my English accent has long been stuck at a local optimum. I like good food and wine, traveling, nature, and lifting weights while listening to cool science or philosophy.


Dec 11, 2023 Released Mixtral 8x7B, GPT3.5 or Llama 2 70B perf with 12B inference speed & cost
Nov 8, 2023 Joined Mistral to accelerate AI open-source!
Sep 13, 2023 Organizing the Workshop on Learning Paradigms Towards Generalist Robots at CoRL 2023
Aug 30, 2023 3 papers accepted at CoRL 2023: Act3D, ChainDiffuser, and HomeRobot
Jul 16, 2023 Organizing the Open Vocabulary Mobile Manipulation (OVMM) Challenge at NeurIPS 2023
Jul 13, 2023 Navigating to Objects Specified by Images accepted at ICCV 2023
Jun 28, 2023 Navigating to Objects in the Real World accepted in Science Robotics 2023
Feb 27, 2023 Habitat-Matterport 3D Semantics Dataset accepted at CVPR 2023