"Planner model generates a task-specific inference program that is executed by a population of Follower models. Our approach equips LMs with the ability to write recursive search procedures that guide LM inference, enabling new forms of verifiable and efficient reasoning."<p>I'm a fan of several of these authors. Curious how much progress can be made with methods like this, despite the limited reasoning of the component models.<p>On the other hand, this might much how humans work. My reasoning is pretty limited without tools and process to ground me. And these things have usually been developed by others.