@elazar_g@yuvalmarton@TaliaRinger@yoavgo I don't know anyone who has the ear of the frontier labs, but I would be excited to see them direct resources at building secure software systems and showing that their most capable models are unable to compromise them, or proposing benchmark evals designed to show progress here.
@TaliaRinger@elazar_g@yuvalmarton@yoavgo I'm curious about whether AI can assist in writing and certifying provably secure software. Setting aside the real concerns around bugs in the formal verification systems, would this be a path toward a more secure digital world (which seems more urgent by the day)?