does it take two (models) to think? does having two models talk to each other beat one model thinking hard? tl;dr "It takes two to think" doesn't replicate for LLMs. It takes ITERATION to think. A single well-prompted model does that just as well as two models in dialogue.