Yarrow comments on ai safety question