A deep dive with the MIT researcher behind Recursive Language Models on how models that call themselves over long contexts beat plain scaling on hard tasks.