Beyond Semantic Narrowing: Robust and Efficient LLM Watermarking with Hamming Neighborhoods
The proposed HammingMark is a semantic watermarking method that uses the semantic hash of the preceding sentence as a dynamic center and accepts candidates whose hashes fall within its Hamming neighborhood, which achieves strong robustness, high detectability, and near-unwatermarked generation quality.
Ze-Wen Sun, Tong-Yang Zhao, Li-Yao Xiang et al.
· 0 citations