LLM Watermarking
LLM watermarking aims to embed imperceptible signals within text generated by large language models (LLMs) to verify authorship and deter misuse. Current research focuses on developing robust watermarking techniques, analyzing their susceptibility to spoofing and removal attacks, and evaluating the trade-off between watermark detectability and the quality of generated text. These efforts are crucial for addressing concerns about intellectual property, misinformation, and the responsible deployment of LLMs, impacting both the security of AI systems and the broader societal implications of AI-generated content.
Papers
Hey, That's My Model! Introducing Chain & Hash, An LLM Fingerprinting Technique
Mark Russinovich, Ahmed Salem
Building Intelligence Identification System via Large Language Model Watermarking: A Survey and Beyond
Xuhong Wang, Haoyu Jiang, Yi Yu, Jingru Yu, Yilun Lin, Ping Yi, Yingchun Wang, Yu Qiao, Li Li, Fei-Yue Wang