Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Deliberative Searcher: Improving LLM Reliability via Reinforcement Learning with constraints

Created by
  • Haebom

Author

Zhenyun Yin, Shujie Wang, Xuhong Wang, Xingjun Ma, Yinchun Wang

Outline

In this paper, we propose Deliberative Searcher, the first framework that integrates certainty correction and retrieval-based question answering to improve the reliability of large-scale language models (LLMs). The agent performs multi-stage reflection and verification on Wikipedia data, and is trained with a reinforcement learning algorithm that optimizes accuracy under soft confidence constraints. Experimental results show that the proposed method improves the alignment between model confidence and accuracy, resulting in more reliable output. This paper will be continuously updated.

Takeaways, Limitations

•
Takeaways:
◦
A new framework (Deliberative Searcher) for improving the reliability of LLM is presented.
◦
Improving reliability by integrating certainty correction and search-based question answering
◦
Optimizing model accuracy and reliability with reinforcement learning-based training
◦
Generate more reliable output by improving the alignment between model confidence and accuracy.
•
Limitations:
◦
Limitations of relying on Wikipedia data
◦
There is a possibility that the current level of completion is lacking due to continuous updates.
◦
Further research is needed on the generalization performance of the proposed method.
◦
Need to evaluate performance on various datasets and question types
👍
Made with Slashpage