Teuken 7B

Unicode watermarking methods tested against LLMs, with mixed results

A new paper from arXiv analyzes the security and detectability of Unicode text watermarking methods against various large language models. Researchers tested ten watermarking techniques across six models, including GPT-…