THOUGHTSCULPT: Reasoning with Intermediate Revision and Search

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Chi, Yizhou, Yang, Kevin, Klein, Dan
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866917923801530368
author Chi, Yizhou
Yang, Kevin
Klein, Dan
author_facet Chi, Yizhou
Yang, Kevin
Klein, Dan
contents 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).
format Preprint
id arxiv_https___arxiv_org_abs_2404_05966
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle THOUGHTSCULPT: Reasoning with Intermediate Revision and Search
Chi, Yizhou
Yang, Kevin
Klein, Dan
Computation and Language
Artificial Intelligence
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).
title THOUGHTSCULPT: Reasoning with Intermediate Revision and Search
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2404.05966