THOUGHTSCULPT：基于中间修订与搜索的推理
THOUGHTSCULPT: Reasoning with Intermediate Revision and Search
AI总结：
本文提出 THOUGHTSCULPT，将 MCTS 搜索与中间修订动作结合，用于可分解输出任务，并在三类任务上显著优于现有推理方法。
摘要：
我们提出了 THOUGHTSCULPT，这是一种面向输出可分解为多个组成部分的任务的通用推理与搜索方法。THOUGHTSCULPT 使用 Monte Carlo Tree Search（MCTS，蒙特卡洛树搜索）探索潜在解决方案的搜索树，每次执行一个动作来逐步构建解决方案，并根据任意领域特定启发式方法进行评估；在实践中，该评估器通常只是一个 LLM 评估器。关键在于，我们的动作空间包含修订动作：THOUGHTSCULPT 可以选择修订其先前输出的某一部分，而不是继续构建输出的其余部分。实验表明，THOUGHTSCULPT 在三项具有挑战性的任务上优于最先进的推理方法：故事大纲改进（趣味性最高提升 30%）、迷你填字游戏求解（单词成功率最高提升 16%）以及受约束生成（概念覆盖率最高提升 10%）。
英文摘要：
We present THOUGHTSCULPT, a general reasoning and search method for tasks with outputs that can be decomposed into components. THOUGHTSCULPT explores a search tree of potential solutions using Monte Carlo Tree Search (MCTS), building solutions one action at a time and evaluating according to any domain-specific heuristic, which in practice is often simply an LLM evaluator. Critically, our action space includes revision actions: THOUGHTSCULPT may choose to revise part of its previous output rather than continuing to build the rest of its output. Empirically, THOUGHTSCULPT outperforms state-of-the-art reasoning methods across three challenging tasks: Story Outline Improvement (up to +30% interestingness), Mini-Crosswords Solving (up to +16% word success rate), and Constrained Generation (up to +10% concept coverage).
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
