Stefan_Schubert comments on Mediocre AI safety as existential risk