Learn about the current state and future challenges of watermarking Large Language Model outputs in this 21-minute lecture by Scott Aaronson from UT Austin. Explore the technical and political barriers preventing widespread deployment of watermarking technology for AI-generated content, while gaining insights into the intersection of alignment, trust, and copyright issues in LLMs. Discover key considerations and potential solutions for implementing effective watermarking systems to help authenticate and track AI-generated content.
Watermarking LLM Outputs: Technical and Political Challenges
Overview
Syllabus
Watermarking: where to?
Taught by
Simons Institute
Related Courses
-
Watermarking of Large Language Models
-
Pseudorandom Error-Correcting Codes with Applications to Watermarking Generative AI
-
Edit Distance Robust Watermarks - Beyond Substitution Channels in Language Models
-
Watermarking Generated Text for LLM Attribution
-
A Sociotechnical Approach to a Safe, Responsible AI Future: A Path for Science and Evidence-based AI Alignment and Trust
-
Differentially Private Synthetic Data for Private LLM Training
