No. 935 - A novel multi-step prompt approach for LLM-based Q&As on banking supervisory regulations
The study explores the use of Large Language Models (LLMs) to answer complex interpretative questions on banking regulations. In particular, it presents a method for identifying relevant regulatory texts with which to expand the information available to the model, beyond the information provided in the question itself. The study also introduces a system for automatically assessing the quality of the answers provided by the model.
The results indicate that the inclusion of additional regulatory references retrieved by applying the proposed method, enables LLMs to better understand and contextualize the questions, leading to the generation of more accurate and comprehensive answers than would otherwise be obtained. The approach we propose thus increases the utility of these models in supporting human experts.
Full text
-
23 April 2025
YouTube
X - Banca d'Italia
Linkedin
RSS
E-mail Alert
