New planning method improves language model reasoning on complex tasks
ReHoPER: Receding-Horizon Planning for Enhanced Reasoning
Computation and LanguageArtificial IntelligenceMachine Learning
Summary
Solving tricky reasoning problems often requires breaking them down into smaller questions. The authors proposed ReHoPER, a technique that plans multiple small questions ahead before answering the main question, helping large language models think more carefully. It works without needing special training or tailored instructions and performs better across different reasoning challenges. This method is especially good at handling problems that need several steps of thinking.
What this means in practice
- •For ai system developers: Improve language model-based assistants by enabling more reliable multi-step problem solving without extra training data.
- •For automated customer support teams: Enhance chatbot reasoning for complex user queries by planning intermediate clarifying questions to reach accurate responses.
Authors
Saeed Ahmadnia, Cornelia Caragea
Abstract
We propose ReHoPER, an inference-only, zero-shot method that improves large language models' reasoning by generating and answering intermediate questions along multiple paths before the final answer. It iteratively plans a horizon of candidate intermediate questions, selects one to answer, and replans from the updated history. ReHoPER is task-agnostic, using the same generic instructions across datasets and models without labeled data or task-specific prompt design. Across multiple datasets, including iLLC, a new controlled benchmark for compositional reasoning, ReHoPER outperforms strong baselines, with the largest gains in the most compositional settings. Our implementation and the iLLC generator are publicly available to support future work.