Skip to main content

Overview

Few-shot CoT requires curated examples. We can automate selection by clustering candidate questions via embeddings, sampling per cluster, and filtering using a simple criterion (e.g., ≤ 5 reasoning steps).

Example

References

  1. Automatic Chain of Thought Prompting in Large Language Models (https://arxiv.org/abs/2210.03493)
  2. The Prompt Report: A Systematic Survey of Prompting Techniques (https://arxiv.org/abs/2406.06608)