Daily Arxiv

This page organizes papers related to artificial intelligence published around the world.
This page is summarized using Google Gemini and is operated on a non-profit basis.
The copyright of the paper belongs to the author and the relevant institution. When sharing, simply cite the source.

Better by Comparison: Retrieval-Augmented Contrastive Reasoning for Automatic Prompt Optimization

Created by
  • Haebom

Author

Juhyeon Lee, Wonduk Seo, Hyunjin An, Seunghyun Lee, Yi Bu

Outline

This paper presents Contrastive Reasoning Prompt Optimization (CRPO), a novel framework for prompt optimization in large-scale language models (LLMs). CRPO utilizes the inherent reasoning capabilities of LLMs to optimize prompts through a retrieval-augmented reasoning process that learns from contrastive examples. Using the HelpSteer2 dataset, we compare high- and low-quality prompt-response pairs and allow the LLM to improve its prompts through tiered contrastive reasoning and multi-metric contrastive reasoning. Experimental results show that CRPO outperforms existing methods.

Takeaways, Limitations

•
Takeaways:
◦
A novel prompt optimization approach leveraging the inference capabilities of LLM is presented.
◦
Achieving more robust and interpretable optimization through contrastive example learning.
◦
Performance verification through experiments using the HelpSteer2 dataset.
◦
Effective application of retrieval-augmented reasoning to prompt optimization
•
Limitations:
◦
HelpSteer2 dataset dependency
◦
Possible lack of further explanation on the detailed implementation of tiered/multi-metric contrastive reasoning
◦
Generalizability to other LLM models and datasets needs to be verified.
👍
Made with Slashpage