Mati_Roy comments on Technical AGI safety research outside AI