The banking industry’s reaction to recent advances in artificial intelligence (AI) platforms such as OpenAI’s ChatGPT and freshly released GPT-4 has been a mix of excitement and caution.
Morgan Stanley is among the banner clients of GPT-4, using it to enhance its internal knowledge base and handle federated search capabilities for its wealth management personnel. JPMorgan Chase, in contrast, has tried to restrict employee use of ChatGPT. We believe adoption is inevitable, especially after Microsoft integrates GPT’s capabilities into its Office products.
Awareness of the opportunities for disruption around these new AI tools, commonly termed large language models (LLMs), is spreading in financial services, with established technology vendors scrambling to integrate them into their offerings, and a surge of investment in AI and new startups.
LLMs are evolving at an incredible pace, with OpenAI pushing updates to the publicly available GPT models multiple times a month. With that in mind, here are some of applications we see emerging in the FI space:
Technical support: Stripe has also been using the technology to streamline helpdesk functions by automatically routing issue tickets and summarizing user queries.
Fraud detection: Payments giant Stripe is piloting using GPT-4 to autonomously monitor community and support forum posts and flag malicious actors who are phishing or trying to gain information on vulnerabilities.
Training/education: LLMs can help produce training scripts and quizzes, accelerate tailored training and onboarding for new employees, and flexibly answer questions and reframe content to aid learning. Duolingo uses GPT-4 to provide customized conversational practice and feedback to help users learn new languages.
Professional writing: LLMs can assist with ad copy, internal/external communication, eminence, sales outreach, document summaries, and status reporting. An increasing number of third-party tools and plugins use OpenAI’s APIs to deliver on-the-spot drafting and editing capabilities, including highly specialized and advanced revisions.
Code generation and review: While the capabilities are nascent and generally not suitable to be used without oversight, LLMs can create functional software from scratch based on high-level directions. Tremendous resources are being dedicated to refining and integrating these capabilities into development workflows.
Web/mobile design: OpenAI has demonstrated how GPT-4’s image classification function could be used instantly create a functioning webpage prototype. This can only improve as OpenAI’s technology evolves.
Reporting: Further democratization of data analytics will be driven by the capability of LLMs to create and iterate upon SQL queries and reports based on plain language instruction, helping bridge a longstanding gap in understanding between business users and technical personnel. LLMs could also help smaller FIs that struggle to use data effectively.
The Bottom Line
LLMs will find their way into nearly every FI, one way or another. Even if institutions are slow to embrace the technology, individual employees will realize how valuable these AI tools can be for their work, and existing vendors will be using them to enhance and expand their products.
There’s a quiet revolution occurring in knowledge work, and FIs must figure out how to realize the potential of LLMs and manage the risks associated with them.
SRM is here to educate and guide our bank and credit union partners in various areas, such as LLMs, prompt engineering, and risks and security. Expect more content as we monitor evolving trends and share our insights throughout the journey.
Connor Heaton, vice president of intelligent automation and AI, has years of experience in this space from prior work at Deloitte advising federal agencies and startups. At SRM, he has led numerous engagements around automation and conversational AI. He can be reached at firstname.lastname@example.org.