habryka comments on an effective ai safety initiative