AI, GPT & the Immediate Applications in Finance
0 likes170 views
This document discusses how leading financial services companies are applying generative AI models like GPT-3 and GPT-4 to address immediate business needs. Stripe has implemented 3 prototypes produced by GPT-4 into their platform to better understand customers, answer support questions, and detect fraud. Morgan Stanley is using GPT-4 to power an internal chatbot for employees to search company documents. Bloomberg has developed its own model, BloombergGPT, to generate SEC filings, analyze markets, automate financial news, and improve risk management. The document concludes with considerations for financial institutions looking to apply generative AI, such as ensuring access to proprietary data and identifying existing chat-based interactions that could be enhanced.
AI, GPT & the Immediate Applications in Finance
- 1. AI, GPT & the Immediate Applications in Financial Services TV
- 2. Generative AI could expose the equivalent of 300 million full-time jobs to automation. How are leading financial services companies reacting right now? TV
- 3. Stripes team put together a list of 50 potential applications to test GPT-4; and after vetting and testing, 15 of the prototypes were considered strong candidates to be integrated into the platform and 3 made it to production. TV
- 4. Better understanding users' businesses Stripe uses GPT-4 to scan users' websites and deliver a summary that helps them understand how each business uses the platform. Answering support questions about documentation GPT-4 is able to understand the user's question, read technical documentation, identify the relevant section, and summarize the solution in one place. Fraud detection on community platforms Stripe maintains a community on forums like Discord. Malicious actors can find their way into these forums, often trying to gain critical information from community members. GPT-4 helps Stripe's fraud team scan inbound communications and identify coordinated activity from malicious actors. 1 2 3 TV
- 5. Morgan Stanley is using OpenAI's GPT-4 to power an internal chatbot for wealth management personnel to easily search and locate relevant information. TV
- 6. Morgan Stanley has fine-tuned and trained the model on over 100,000 of its own documents that contain information on investment recommendations, general business questions, investment strategies, market research, analyst insights and process questions that their financial advisors might have. Morgan Stanleys financial advisors can then ask questions on these topics and get answers from the model. Morgan Stanley restricts prompt topics to business-relevant issues to ensure that the outputs will come from its own curated documents. The system is being trained by more than 200 employees on a daily basis. TV
- 7. BloombergGPT is machine learning model which uses proprietary and curated datasets such as financial news, company filings and Bloomberg News content. It fragments the financial data into 363 billion tokens and and the estimated cost of the model is more more than $2.7m. Here are the envisioned use-cases
- 8. Generate an initial draft of a Securities and Exchange Commission filing With access to large amounts of data from financial filings, BloombergGPT (B-GPT) may be able to produce a provisional SEC filing, potentially reducing the cost of filing. Conduct financial market analysis: Help investors and traders analyze financial market trends and make informed decisions based on the insights gleaned from the analysis. Automate financial news reporting: Automatically generate financial news reports, potentially saving time and money for media organizations. Enhance investment portfolio management: Help investors manage their portfolios more e鍖ectively by providing insights into market trends and identifying opportunities for investment. Improve financial risk management: Help financial institutions identify and manage risks more e鍖ectively by analyzing large amounts of financial data and detecting patterns that may indicate potential risks. 1 2 3 4 5 TV
- 9. A few takeaways to consider TV
- 10. Thinking from the first principles: Data Availability: Does your organization have significant proprietary data sets on your customers, their behaviors, internal knowledge or external world? Chat as a short-term form factor What are the use-cases where your customers, stakeholders and employees are interacting in this form factor already? Possible vectors of short term applications Customer support full service & product walkthroughs automatization Internal knowledge 2.0 going from static wikis to dynamic & contextual ChatGPT-like queries Complex textual input analysis & interpretation co-pilot in interpretation of complex documentation (regulatory requirements, legal agreements, credit scoring reports, etc.) 1 2 3 TV
- 11. Thanks for reading & get in touch: www.tomasvysny.com tvysny@gmail.com @tomas_vysny Sources: OpenAI Customer Case Studies, Goldman Sachs Economic Research, Bloomberg TV