The advent of ChatGPT in November 2022, a Large Language Model (LLM) combined with a chatbot, represents a pivotal moment in the public recognition of these technologies, prompting companies to explore their multifaceted utility in various business domains. Concurrently, scientific studies are examining the advantages and constraints of artificial intelligence, encompassing LLMs and chatbots, even in the field of finance. The objective of this Bachelor’s thesis is to provide a comprehensive evaluation of the potential applications of LLMs in the field of finance. The integration of ChatGPT in the professional practice of the financial domain can potentially enhance efficiency, for example, by replacing time-consuming searches with fast and targeted queries to LLMs, thereby facilitating the identification of relevant legal and literature passages more expeditiously. In this Bachelor's thesis, the application is examined in an experiment using versions of ChatGPT that have not been specially trained further. The aim is to evaluate the reliability of ChatGPT's answers to accounting-relevant questions in accordance with the Austrian Commercial Code (UGB) and IFRS. In order to obtain the most optimal results, prompts are developed with the assistance of the optimization recommendation derived from the literature, prompt engineering. The results of the experiment demonstrate that ChatGPT is currently unsuitable for use as a reliable source for searches. Inaccuracies and superficialities are identified in the responses to accounting law questions in both the UGB and IFRS. Furthermore, the paragraphs of the UGB are incorrectly issued, whereas the standard clauses in the IFRS can at least serve as a rough guide to assist beginners in locating the relevant standard clauses for fundamental matters. Given that activities in the financial domain encompass not only the application of accounting regulations but also, for instance, communicative interactions with colleagues, it becomes pertinent to inquire about the further fields of application and challenges of LLMs in companies and to what extent financial employees recognise the potential of LLMs, such as ChatGPT, for support in everyday working life. A review of the literature reveals that the potential applications of LLMs in science are numerous. An online survey was conducted to empirically study actual user behaviour, and the results demonstrated that participants already utilise LLMs in a variety of ways. However, the challenges encountered were extensive and largely corresponded to those described in the literature. Moreover, the empirical study indicates that although the term "prompt engineering" is not widely known, this method is employed intuitively by the users. This thesis provides a comprehensive analysis of the fundamentals of LLMs, along with detailed insights into the optimisation approaches, potential applications, and challenges associated with the use of LLMs in the financial domain. It offers a theoretical and practical examination of these issues. Given the extensive scope of the topic and the rapid pace of developments in this field, the bachelor thesis can't provide a comprehensive overview of all aspects of the subject matter. However, it serves as a foundation for future research.
ChatGPT im Finanzbereich
Frenkenberger, S. (Author). 2024
Student thesis: Bachelor's Thesis