Specialization after generalization: towards understanding test-time training in foundation models - Best AI papers explained | Wave AI Podcast Notes