And I think there is fertile ground for discussion inside and outside labs on ~"OK under what circumstances should we want our models to detect someone engaged in delusional thinking and intervene, and should we have same bar if they legit work at WHO or White House, etc."