Nick Bostrom spoke about his recent research on the interaction between AI control problems and governance strategy within AI risk, and the role of openness (slides/video). Bostrom and co-authors have recently published several technical reports on policy desiderata, implications of openness in AI development, and technological race dynamics in AI research.
Owain Evans presented work he and his collaborators have produced on safety in a reinforcement learning context, including a discussion of their recent paper at AAAI on inverse reinforcement learning with ignorant and inconsistent agents (slides). Owen Cotton-Barratt presented his forthcoming work on decision-relevant uncertainty in AI risk (slides). The full schedule for the event can be found here.
A number of AI researchers have signed their support of 23 principles for beneficial AI, which can be found here.