Navigating the AI Landscape: Monitoring User Interactions with Large Language Models

In today’s modern business landscape, Large Language Models (LLMs) have emerged as essential tools, enabling personalized marketing, data interpretation, and precise financial planning. However, despite their potential benefits, integrating LLMs into business processes poses inherent challenges. Issues such as misalignment, malicious inputs, harmful outputs, sensitive information disclosure, inaccurate information, and unintended biases can significantly impact their efficacy and integrity. The project aims to monitor and analyze internal user interactions with Large Language Models (LLMs) in a business context, with the objective of identifying, flagging, and remediating problematic usage patterns detected within the AI platform’s logging data. This project is crucial for the responsible deployment of a generative AI system within the organization. With all of the recent interest in the capabilities of generative AI and risks that have been identified it is crucial for the organization to move forward in a responsible and cautious manner. Having this project will help the organization better understand the tool which will encourage optimal usage. Because this tool will be given sensitive information such as client Personally Identifiable Information (PII).

Faculty Supervisor:

Xiaodong Lin

Student:

Partner:

Co-operators (General Insurance)

Discipline:

Computer science

Sector:

Finance and Insurance

University:

University of Guelph

Program: