Lune

EMNLP Findings2024

Downstream Trade-offs of a Family of Text Watermarks

Anirudh Ajith, Sameer Singh, Danish Pruthi

Abstract

Watermarking involves implanting an imperceptible signal into generated text that can later be detected via statistical tests. A prominent family of watermarking strategies for LLMs embeds this signal by upsampling a (pseudorandomly-chosen) subset of tokens at every generation step. However, such signals alter the model's output distribution and can have unintended effects on its downstream performance. In this work, we evaluate the performance of LLMs watermarked using three different strategies over a diverse suite of tasks including those cast as k-class classification (CLS), multiple choice question answering (MCQ), short-form generation (e.g., openended question answering) and long-form generation (e.g., translation) tasks. We find that watermarks (under realistic hyperparameters) can cause significant drops in LLMs' effective utility across all tasks. We observe drops of 10 to 20% in CLS tasks in the average case, which shoot up to 100% in the worst case. We notice degradations of about 7% in MCQ tasks, 10-15% in short-form generation, and 5-15% in long-form generation tasks. Our findings highlight the trade-offs that users should be cognizant of when using watermarked models. 1

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

Questions to start from

Lune can draw on

the full paper · related work

Get the answer

Free to start. No credit card required.

Lineage

  • A Semantic Invariant Robust Watermark for Large Language Models cites Provable Robust Watermarking for AI-Generated Text
  • Provable Robust Watermarking for AI-Generated Text cites Language Models are Few-Shot Learners
  • On the Learnability of Watermarks for Language Models cites A Semantic Invariant Robust Watermark for Large Language Models
  • On the Learnability of Watermarks for Language Models cites Provable Robust Watermarking for AI-Generated Text
Builds onCited byLeft to right by year; the more cited, the higher.

Builds on10

  1. Language Models are Few-Shot LearnersTom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah et al.NeurIPS 2020 · 64,255 citations
  2. PIQA: Reasoning about Physical Commonsense in Natural LanguageYonatan Bisk, Rowan Zellers, Ronan Le Bras, Jianfeng Gao et al.AAAI 2020 · 2,916 citations
  3. Provable Robust Watermarking for AI-Generated TextXuandong Zhao, Prabhanjan Vijendra Ananth, Lei Li, Yu-Xiang WangICLR 2024 · 312 citations
  4. A Semantic Invariant Robust Watermark for Large Language ModelsAiwei Liu, Leyi Pan, Xuming Hu, Shiao Meng et al.ICLR 2024 · 108 citations
  5. On the Learnability of Watermarks for Language ModelsChenchen Gu, Xiang Lisa Li, Percy Liang, Tatsunori HashimotoICLR 2024 · 79 citations

Related papers

  1. Unbiased Watermark for Large Language ModelsZhengmian Hu, Lichang Chen, Xidong Wu, Yihan Wu et al.ICLR 2024 · 103 citations
  2. WatME: Towards Lossless Watermarking Through Lexical RedundancyLiang Chen, Yatao Bian, Yang Deng, Deng Cai et al.ACL 2024 · 2 citations
  3. Optimizing Watermarks for Large Language ModelsBram WoutersICML 2024 · 25 citations
  4. Can Watermarking Large Language Models Prevent Copyrighted Text Generation and Hide Training Data?Michael-Andrei Panaitescu-Liess, Zora Che, Bang An, Yuancheng Xu et al.AAAI 2025 · 21 citations
  5. No Free Lunch in LLM Watermarking: Trade-offs in Watermarking Design ChoicesQi Pang, Shengyuan Hu, Wenting Zheng, Virginia SmithNeurIPS 2024 · 56 citations
Dusk over the sea between two cliffs drawn in fine vertical lines