On October 7, 2026, Lorenzo Baraldi (Modena ELLIS Unit, UNIMORE) gave an invited online talk at the AI Seminar Cycle organised by Hi! PARIS in collaboration with the ELLIS Unit Paris and the ELLIS Program on Theory, Algorithms, and Computations of Modern Learning Systems. The seminar is a monthly series that connects students, PhD candidates and researchers with leading voices in artificial intelligence, and Lorenzo’s talk opened its 2026–2027 programme.
The talk, titled “Inside Multimodal Foundation Models: Representations as a Lever for Control, Safety and Grounding”, looked inside the internal representations of multimodal foundation models and showed how understanding them can be turned into practical tools across three areas of AImageLab’s research: multimodal large language models, retrieval-augmented generation and visual generation.
We thank Gianni Franchi (ENSTA Paris) and the Hi! PARIS team for the invitation. The recording of the talk will be made available on the Hi! PARIS YouTube channel.
