Skip to main content Skip to navigation

Gillmore Centre Webinar Symposium

04.07.2023 Revolutionising Finance: The Rise of Large Language Models and NLP

Revolutionising Finance: The Rise of Large Language Models and NLP

Join us online to examine the research that has underpinned Large Language Models (LLMs) and their likely use and impact on the world of Finance and investment management. Explore the advancements in natural language processing (NLP) that has led to the development of LLMs and gain valuable insights into the potential and limitations of these powerful language models. Engage with leading AI researchers and industry experts as they discuss the implications of LLMs on NLP, share real-world applications and use cases, and examine the challenges and opportunities in integrating LLMs into investment management processes. Don't miss this opportunity to stay ahead of the curve and understand how LLMs are reshaping the future of finance.

Hosted by

Dr Dan PhilpsLink opens in a new window is a founding head of Rothko Investment Strategies and is a widely published artificial intelligence (AI) and Finance researcher. With over 20 years of quantitative investment experience, managing numerous top-quartile performing strategies across both equities and fixed-income asset classes. Dan holds a PhD in AI and Computer Science from City, University of London, a BSc (Hons) from King’s College London, is a CFA charter holder, a member of CFA Society of the UK, co-leads AI research strategy at the Gillmore Center for FinTech at Warwick Business School, and is an Honorary Research Fellow at the University of Warwick.

Dr Tim LawLink opens in a new window - Experienced Data Scientist and Quantitative Researcher with strong Machine Learning, Statistics, Quantitative Finance and Software Development background. 10+ years’ experience in data science responsibilities in commercial environments of which 7+ years in the financial sector. 5+ years of experience in leading teams of professionals and researchers from different backgrounds. Doctoral research specialized in Machine Learning, Artificial Intelligence and Financial Computing. Appointed as Honorary Research Fellow at both UCL and Warwick Business School

Overview - Unleashing the Power of Large Language Models in Finance: A Deep Dive 

In the rapidly evolving world of fintech, the potential of artificial intelligence (AI) is being harnessed more than ever. A recent webinar hosted by the Gillmore Centre offered a fascinating exploration of how large language models (LLMs) like GPT-3 and GPT-4 revolutionise financial analysis. Here's a detailed look at the key takeaways from this insightful session. 
Demystifying Large Language Models 
LLMs are AI models trained on vast amounts of text data, capable of generating human-like text. They've been making waves across various sectors, including finance, for their ability to perform tasks like sentiment analysis, price prediction, and business lead generation. 
However, it's not all smooth sailing. While LLMs are fluent in language generation, their reliability and reasoning can sometimes be questionable. This makes them less suitable for tasks requiring high accuracy, such as mortgage approvals or writing critical code with real-world implications. 
The Role of LLMs in Financial Analysis 
The webinar highlighted several exciting applications of LLMs in financial analysis. For instance, they can be used to analyze financial news, predict stock prices, and even generate financial reports. AI tools for these applications can be identified using the "there's a for that" plugin, a handy tool that finds AI tools for specific use cases. 
Moreover, LLMs can be used to analyze scientific literature relevant to financial analysis. The "scholarai" plugin can be used to find relevant papers based on keywords, retrieve the full text of a paper, and even save citations to a reference manager. 
Overcoming Challenges with Innovative Solutions 
Despite their potential, LLMs face significant challenges in financial analysis. One of the main hurdles is their inability to handle long texts, such as ESG reports, due to their length restrictions. 
To tackle this issue, Xinyu Wang presented her innovative method, "Orange," based on reading order and font size. This method divides the document based on a tree structure rather than pages, allowing for a more global understanding. In her experiments, Wang's method outperformed other non-large language model-based methods. For instance, her method correctly identified and organized the table of contents of an ESG report, while GPT-4 failed to do so. 
Looking Ahead 
The webinar concluded with a discussion on the future of LLMs in financial analysis. Wang mentioned that she plans to integrate large language models into her method to enable them to understand long documents. She also plans to release her source code on GitHub, which could be a valuable resource for students and researchers. 
The Gillmore Centre webinar offered valuable insights into the potential and challenges of using large language models in financial analysis. While these models have significant potential, they face challenges, particularly in handling long documents. Innovative approaches like Wang's "Orange" method offer promising solutions to these challenges, paving the way for more effective use of LLMs in financial analysis. The future of LLMs in this field looks promising, with ongoing research and development to overcome their current limitations. Stay tuned for more exciting developments in this space! 


Professor Yulan HeLink opens in a new window - is an Honorary Professor at the University of Warwick and a Professor in Natural Language Processing at King's College London. She has published over 210 papers in the areas of natural language understanding, sentiment analysis and opinion mining, question-answering, topic/event extraction from text, biomedical text mining, and social media analytics. She currently holds a five-year Turing AI Fellowship, funded by the UK Research and Innovation (UKRI). Yulan obtained her PhD degree in spoken language understanding from the University of Cambridge and her MEng and BASc (First Class Honours) degrees in Computer Engineering from Nanyang Technological University, Singapore.

Dr Adriano Koshiyama co-founded Holistic AI, an AI Governance, Risk and Compliance (GRC) software solution. Holistic AI services many large and medium-size organizations on their journey to adopting AI, ensuring due risk-management, and compliance with the changing regulatory & standards environment. Previously, he was a Post-doctoral Research Fellow in Computer Science at University College London, and academically, he has published more than 50 papers in international conferences and journals.

Dr Tillman Weyde Link opens in a new windowIs a researcher and academic in Deep Machine Learning, Signal Processing, and Music Informatics. Working on - analysing and processing signals (audio, images) and symbolic data (language, music)- understanding and improving deep (and non-deep) neural networks.Specialities: Neural-Symbolic Integration, Neural Network Models and Theory, Audio Analysis and Signal Enhancement, Music and Language Models, Multimodal Learning and Processing, Time Series.


Dr Dan Philps and Dr Tim Law: Contextualizing the impact of LLMs on FinTech and investment management and the potential disruption to come.

Professor Yulan He: Have LLMs Solved NLP?

Abstract: Natural Language Processing (NLP) has witnessed significant advancements recently, with Large-scale Language Models (LLMs) such as ChatGPT, GPT-4 and LLaMA pushing the boundaries of machine reading comprehension and language generation. In this talk, I will delve into the question of whether these LLMs have successfully overcome the challenges of NLP by examining their capabilities in a range of NLP tasks. I will conclude my talk with the exciting future of AI-driven language understanding.

Dr Adriano Koshiyama: On Providing Risk Management Guardrails to LLMs

Abstract: the speaker will provide a quick summary of the tools and solutions that can be adopted by medium to large organizations to risk manage the use of LLMs by non-technical users. He will discuss the main safety concerns and how they can be translated into technological solutions for risk prevention, detection and correction.

Dr Dan Philps: LLMs in investment management: risk and opportunity

Abstract: For most investment managers, ChatGPT represents the starting whistle in a tech arms race many had hoped to avoid. How can quant and fundamental analysts apply LLMs like ChatGPT? How effective a “copilot” can these technologies be? Where is the technology headed? How will it transform investment management?

Dr Tillman Weyde: Knowledge integration and multimodality in LLMs

Abstract: Since their advent, LLMs have quickly received unprecedented levels of interest in many areas, including investment and finance. LLMs demonstrate impressive progress in language modelling and human-like fluency of the generated language. However, they are now used for their general intelligence, which is not what language models are designed or optimised for, and there are limitations that have an impact on application development. Factual correctness is one challenge that has been generally recognised. This can be addressed with knowledge integration, but it is still unclear how to reliably use knowledge bases and reasoning with language models. We will present ongoing research into these questions.

Xinyu Wang: A Scalable Framework for Table of Contents Extraction from Complex ESG Annual Reports

Abstract: Table of contents (ToC) extraction centres on hierarchically structuring documents. ESG reports pose significant challenges due to their diverse structures and extensive length. Existing large language models (LLMs), including ChatGPT and GPT-4, while capable of managing longer inputs, still fall short in processing ESG reports due to their inherent length restrictions and the absence of hierarchical modelling capabilities. LLMs are also very expensive. To address these challenges, we propose a new framework for Toc extraction, consisting of three steps: (1) Constructing an initial tree of text blocks based on reading order and font sizes; (2) Modelling each tree node (or text block) independently by considering its contextual information captured in node-centric subtree; (3) Modifying the original tree by taking appropriate action on each tree node (Keep, Delete, or Move). This approach offers a practical solution for document segmentation, allowing section headings to exploit both local and long-distance contexts pertinent to themselves.