Tag: Mechanistic Interpretability

From the democratization of dual-use biotechnology to the emergence of deceptive agentic systems, the trajectory of Artificial Intelligence raises urgent questions about control and alignment. In this talk, Charbel-Raphaël Segerie presents a critical analysis of the current race toward superintelligence and the widening gap between technological capabilities and control mechanisms.

Are qualitative differences in conscious experience, like vision versus audition, shaped by the intrinsic structure of neural networks? This talk proposes a mathematical framework using symmetry groups and equivariance to characterize these differences. This approach offers a new path for linking consciousness to neural architecture.