So Kuroki

Adaptive Branching MCTS, a novel inference-time framework for LLMs, generalizes repeated sampling with multi-turn exploration and exploitation.