PeopleofSuperintelligence

Researcher

Marie Pellat

Marie Pellat is an AI researcher whose work spans machine learning for online conversation, language models, reasoning, and AI safety. She studied mathematics and physics at École Polytechnique, earned an MS at Stanford, and worked on Nest's machine-learning team before moving to Google's Jigsaw. Her work includes Perspective, PaLM, instruction-tuning research, and papers on AI biorisk and open AI safety. By May 2024, she was an AI Scientist at Mistral; in June 2025 she coauthored Magistral, Mistral's first reasoning model, and a report on open AI safety. As of October 2026, she is a research engineer at Google DeepMind.

Timeline

  1. 2014–2018

    Wind research and Jigsaw

    In 2014, she coauthors a study of Ontario wind power, electricity prices, and greenhouse-gas emissions that finds congestion matters. By 2018, she holds a BS in mathematics and physics from École Polytechnique and an MS from Stanford. In 2018, she moves from Nest's machine-learning team to Google's Jigsaw, where she works on language models for online conversation.

  2. May 2019

    Perspective in French

    In May 2019, she and Daniel Borkan are Jigsaw software engineers as Perspective and Tune become available in French. Perspective uses machine learning to help publishers moderate comments and improve online conversation.

  3. 2022

    PaLM and instruction tuning

    In 2022, she coauthors PaLM, a 540-billion-parameter language model trained with Pathways. She also coauthors a study finding that instruction tuning across tasks, model sizes, and chain-of-thought data improves results across several benchmarks.

  4. May 2024–June 2024

    AI Scientist at Mistral

    By May 2024, she is an AI Scientist at Mistral. In June 2024, she points to France's large AI talent pool as a competitive advantage.

  5. December 2024

    AI and biorisk research

    In December 2024, she coauthors The Reality of AI and Biorisk, which reviews evidence on information access and planning through LLMs and the use of AI-enabled biological tools.

  6. June 2025

    Magistral and AI safety papers

    In June 2025, she coauthors Magistral, Mistral's first reasoning model and a scalable reinforcement-learning pipeline. The paper presents Magistral Medium, trained with reinforcement learning, and opens Magistral Small under Apache 2.0. She also coauthors A Different Approach to AI Safety, a proceedings paper on outcomes from the Columbia Convening and safety work for open foundation models.

  7. As of October 2026

    Research engineer at DeepMind

    As of October 2026, she is a research engineer at Google DeepMind and speaks on AI for health at Startup Day France.

Career

  1. Since 2026Google DeepMindResearch Engineer
  2. Until 2018Nest
  3. From 2018GoogleSoftware Engineer
  4. Mistral AIAI Scientist

Education

  1. 2018École PolytechniqueBS in mathematics and physics
  2. 2018Stanford UniversityMS

From the biography.

  1. Google had the most people shaping AI every year from 2014 to 2025Includes Google. Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.
  2. Google has the highest combined h-index among AI organizationsIncludes Google. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.
  3. 20% of people at companies who shape AI also hold a university role, led by Google and Stanford UniversityIncludes Google. We estimate that 539 of the 6,766 people with a known current organization (8%) hold both a university role and a company role, including 20% of people at companies and 13% of people at universities.
  4. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Google. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.

Colleagues

People with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Marie Pellat yet. Send us a free-licensed photo.

  1. Influence of wind power on hourly electricity prices and GHG emissions: Evidence that congestion matters from Ontario zonal data ideas.repec.org. Checked 20 October 2026.
  2. ISSIP Cognitive Systems Institute Group Speaker Series : October 4, 10:30am US Eastern Time issip.org. Checked 20 October 2026.
  3. Parlons des élections ! Perspective et Tune sont désormais disponibles en français france.googleblog.com. Checked 20 October 2026.
  4. PaLM: Scaling Language Modeling with Pathways arxiv.org. Checked 20 October 2026.
  5. Scaling Instruction-Finetuned Language Models arxiv.org. Checked 20 October 2026.
  6. L’école du numérique La Plateforme ouvre la réflexion sur l’intelligence artificielle gomet.net. Checked 20 October 2026.
  7. Business Solution : Spécial IA/Data laplateforme.io. Checked 20 October 2026.
  8. The Reality of AI and Biorisk arxiv.org. Checked 20 October 2026.
  9. Magistral arxiv.org. Checked 20 October 2026.
  10. A Different Approach to AI Safety: Proceedings from the Columbia Convening on Openness in Artificial Intelligence and AI Safety arxiv.org. Checked 20 October 2026.
  11. Startup Day France 2026 | Google Cloud cloud.google.com. Checked 20 October 2026.