# Large Language Models in Finance: A Survey

**Authors:** Yinheng Li, Shaofei Wang, Han Ding, Hang Chen  
**arXiv:** [2311.10723](https://arxiv.org/abs/2311.10723)  
**Date:** 2023-09-28  
**Summary page:** https://www.summarizepaper.com/en/paper/2311.10723v1/

## Key points

- Recent advances in large language models (LLMs) have revolutionized artificial intelligence, especially in finance.
- The survey focuses on two key aspects of utilizing LLMs for financial tasks: existing solutions and guidance for adoption.
- Current approaches include leveraging pretrained models through zero-shot or few-shot learning, fine-tuning on domain-specific data, and training custom LLMs from scratch.
- A decision framework is proposed to help financial professionals select the most suitable LLM solution based on constraints like data availability, computational resources, and performance requirements.
- Limitations and challenges associated with using LLMs in finance are discussed, including issues related to explainability and governance.
- Continuous research efforts, robust evaluation frameworks, and appropriate safeguards are highlighted as crucial for ensuring ethical use of LLMs in finance.

## Summary

Recent advances in large language models (LLMs) have revolutionized the field of artificial intelligence, particularly in finance. In this comprehensive survey, the authors focus on two crucial aspects of utilizing LLMs for financial tasks: existing solutions and guidance for adoption. The paper begins by examining current approaches that employ LLMs in finance, such as leveraging pretrained models through zero-shot or few-shot learning, fine-tuning on domain-specific data, and training custom LLMs from scratch. The authors summarize key models and evaluate their performance enhancements on various financial natural language processing tasks. Furthermore, a decision framework is proposed to assist financial professionals in selecting the most suitable LLM solution based on their specific constraints related to data availability, computational resources, and performance requirements. This framework offers a structured pathway from initial experimentation with lightweight models to potential investment in customized LLMs tailored to specific financial applications. Moreover, the paper delves into discussing the limitations and challenges associated with leveraging LLMs in financial contexts. While LLMs offer enhanced interpretability compared to traditional deep learning models by providing reasoning steps for generated answers under certain conditions, they still pose challenges in terms of explainability and governance. It is emphasized that addressing these limitations and ensuring ethical use of LLMs in finance is crucial through continuous research efforts, robust evaluation frameworks, and implementation of appropriate safeguards. In conclusion, this timely survey provides valuable insights into the emerging application of LLMs in finance. By exploring diverse approaches to harnessing LLMs for financial tasks and offering guidance on adoption considerations, the authors highlight the significant performance gains achievable with specialized LLM solutions across various natural language processing applications within the finance industry.