A Watermark for Large Language Models
Kirchenbauer et al., 2023, A Watermark for Large Language Models (Proceedings of the 40th International Conference on Machine Learning (ICML 2023), PMLR 202)
Subscriber deck — free during early access
1 cards in “7.1 Degradation Under Attack: Span Replacement” · whole deck
7.1 Degradation Under Attack: Span Replacement
- How effective was the T5 span-replacement attack?Replacing random spans with T5-Large suggestions, without knowing the green lists, barely helped at a 10% budget (AUC dropped by 0.01); at 30% it degraded the watermark substantially, but tripled perplexity and needed many model calls.
- Kirchenbauer et al., 2023, A Watermark for Large Language Models (Proceedings of the 40th International Conference on Machine Learning (ICML 2023), PMLR 202), §7.1
Source
- John Kirchenbauer, Jonas Geiping, Yuxin Wen, Jonathan Katz, Ian Miers and Tom Goldstein. 2023. A Watermark for Large Language Models. In Proceedings of the 40th International Conference on Machine Learning (ICML 2023), PMLR 202, pp. 17061–17084. https://proceedings.mlr.press/v202/kirchenbauer23a.html