Watermarks in the Sand: Impossibility of Strong Watermarking for Generative Models — mentioned by blintz

Tue, 05 Nov 2024 23:33:35 GMT

These watermarks are not robust to paraphrasing attacks: AUC ROC falls from 0.95 to 0.55 (barely better than guessing) for a 100 token passage. The existing impossibility results imply that these attacks are essentially unavoidable (https://arxiv.org/abs/2311.04378) and not very costly, so this line of inquiry into LLM watermarking seems like a dead end.

Paper and discussion history. A mention is not an endorsement.

Watermarks in the Sand: Impossibility of Strong Watermarking for Generative Models · HackerPapers discussions

Watermarks in the Sand: Impossibility of Strong Watermarking for Generative Models — mentioned by blintz