Is Best-of-N the Best of Them? Coverage, Scaling, and Optimality in Inference-Time Alignment

Open in new window