Comment by ActorNightly

I mean, its literally just more efficient loop of model generating reasoning text, for it to be fed back as context. There is nothing groundbreaking here. At the end of the day, its all just search.