Science News HOME

Scientific News HOME, Scientific Results

A scientific breakthrough reveals why generative AI learns so well

A study conducted by Tony Bonnaire (IAS, CNRS & Université Paris-Saclay) with Raphaël Urfin and Giulio Biroli (ENS Paris, CNRS & PSL) and Marc Mézard (Bocconi University) sheds new light on a central question in generative AI research: how diffusion models—the key technology behind modern generative AI—manage to generalize effectively before memorizing the data on which they are trained.

Science News HOME

Léna Jossé, a New Doctor at the IAS

Léna Jossé, a member of the Astrochemistry and Origins team, successfully defended her dissertation on December 4, 2025, on the topic “3D Detection of Mineralogical Assemblies Using SR-μXCT in a Sample from the Asteroid Ryugu: A Multi-Analytical Study of Their Formation and Evolution.”

Scroll up