The ChatGPT Privacy Tango - Dancing with Data Security and Large Language Models

Explore the intricate balance between data privacy and utility in AI-powered language models during this 50-minute sponsor presentation by Cape Privacy at PyCon US. Delve into the "ChatGPT Privacy Tango," a metaphorical approach to navigating the challenges of leveraging Large Language Models (LLMs) like ChatGPT for sensitive data while maintaining security. Learn about a system designed to protect personally identifiable information (PII) and ensure data privacy without compromising the benefits of AI. Discover the pros and cons of this approach, alternative systems, and gain insights into confidential computing, PII removal, and text de-identification. By the end of the talk, acquire the knowledge and tools necessary to implement privacy-aware applications with LLMs, striking the perfect balance between data security and AI-powered language processing.

Intro
Cape Privacy
Overview
Large Language Models
Privacy Concerns with LLMs
ChatGPT Privacy Tango
Confidential Computing
PII Removal and Text De-identification
What the client sees
System Design: Setup
System Design: Knowledge Base
System Design: User Query
System Analysis: Benefits
System Analysis: Disadvantages
Alternative approaches