Diverge to Induce Prompting: Multi-Rationale Induction for Zero-Shot Reasoning
A new paper on arXiv (2602.08028) introduces Diverge-to-Induce Prompting (DIP), a framework that addresses instability in standard Chain-of-Thought (CoT) prompting. Instead of relying on a single reasoning strategy, DIP first prompts an LLM to generate multiple diverse high-level rationales for each question. Each rationale is elaborated into a detailed step-by-step draft plan, and these plans are then induced into a final plan. This method enhances zero-shot reasoning accuracy without resource-intensive sampling. Experiments show DIP outperforms single-strategy prompting, demonstrating the effectiveness of multi-plan induction for prompt-based reasoning. The paper is categorized as cs.AI and was published on May 22, 2026.
Developers can improve LLM reasoning accuracy without extra computational cost.