Towards Automated Interpretability for AI Safety — with Fazl Barez (Oxford)

about this room

BLISS — the Berlin Learning & Intelligent Systems Society — hosts Dr Fazl Barez, Senior Research Fellow at the University of Oxford, for a 45-minute talk on one of the central problems in modern AI: understanding how systems actually make decisions, and how to verify they behave safely. Barez's research spans interpretability, AI safety and governance, and the talk focuses on moving interpretability from a manual craft towards automated methods that can scale to frontier systems. The evening is pitched at ML students and PhD candidates, academic and industry researchers, and anyone seriously engaged with AI safety, and wraps up with networking over complimentary drinks and pizza at TU Berlin. A high-signal evening for the city's growing AI-research community.