Gavin comments on Mediocre AI safety as existential risk