Diverge to Induce Prompting: Multi-Rationale Induction for Zero-Shot Reasoning
The paper introduces a new framework called Diverge-to-Induce Prompting (DIP) aimed at improving zero-shot reasoning in large language models. By generating multiple diverse rationales for each question, DIP enhances the reasoning process beyond single-strategy approaches. Experimental results indicate that this method significantly outperforms traditional prompting techniques.
- ▪DIP generates multiple high-level rationales for each question.
- ▪Each rationale is elaborated into a detailed draft plan.
- ▪The framework enhances zero-shot reasoning accuracy without resource-intensive sampling.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Computation and Language arXiv:2602.08028 (cs) [Submitted on 8 Feb 2026] Title:Diverge to Induce Prompting: Multi-Rationale Induction for Zero-Shot Reasoning Authors:Po-Chun Chen, Hen-Hsen Huang, Hsin-Hsi Chen View a PDF of the paper titled Diverge to Induce Prompting: Multi-Rationale Induction for Zero-Shot Reasoning, by Po-Chun Chen and 2 other authors View PDF HTML (experimental) Abstract:To address the instability of unguided reasoning paths in standard Chain-of-Thought prompting, recent methods guide large language models (LLMs) by first eliciting a single reasoning strategy. However, relying on just one strategy for each question can still limit performance across diverse tasks.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
