FiSTECH: Financial Style Transfer to Enhance Creativity without Hallucinations in LLMs

TL;DR

This work proposes a novel two-stage fine-tuning strategy wherein in the first stage public domain financial reports are used to train for writing styles while allowing the LLM to hallucinate, which can be generalized to train creativity in LLMs by first allowing them to hallucinate.

Abstract

Recent trends in Generative AI have emerged towards fine-tuning foundational large language models (LLMs) to create domain-specific LLMs for automation and chatbot-like applications. Specialized applications for analytics-heavy domains such as Financial report generation require specific writing styles that comprise compound and creative sentences with minimized hallucinations. In this work, we explore the self-corrective auto-regressive qualities of LLMs to learn creativity in writing styles with minimal prompting. We propose a novel two-stage fine-tuning (FT) strategy wherein in the first stage public domain financial reports are used to train for writing styles while allowing the LLM to hallucinate. In the second stage the examples of hallucinations are manually corrected and further used to fine-tune the LLM. The finally trained LLM learns to generate specific financial report sections using minimal instructions and tabular data inputs while ensuring low fine-tuning costs. Our proposed two-stage fine-tuning boosts the accuracy of financial questions answering by two-folds while reducing hallucinations by over 50%. Also, the fine-tuned model has lower perplexity, improved ROUGE, TER and BLEU scores, higher creativity and knowledge density with lower uncertainty and cross entropy than base LLMs. Thus, the proposed framework can be generalized to train creativity in LLMs by first allowing them to hallucinate.

Figures (12)

• Figure 1: An example of KG generation from financial text. First, the named entities are detected in (a) followed by extraction of the complete KG that captures the relationships between the nodes in (b). The most pertinent KG to a user-query is retrieved for financial applications.
• Figure 2: Example of simple RAG-based approach using GPT4o to generate Financial Reporting text using input data as instructions and tabular sources.
• Figure 3: Basic prompt with instructions and tabular data as input.
• Figure 4: Examples of two-stage fine tuned financial report generation paragraphs. The two major considerations in style-transfer are the formation of compound sentences in (a) and creative logical sentences that are not hallucinations in (b).
• Figure 5: Impact of LLM augmentative approaches using LLM only, Web search$+$LLM and Web search$+$KG$+$LLM to improve quality of responses based on luna.