[wluk]

well that's one possibility. The key (unproven) idea here is that if you use Anthropic to edit o1's responses, it's less likely to hallucinate than if you use o1 to edit o1's responses (which is what o1 actually does).