On the Learnability of Watermarks for Language Models
read the original abstract
Watermarking of language model outputs enables statistical detection of model-generated text, which can mitigate harms and misuses of language models. Existing watermarking strategies operate by altering the decoder of an existing language model. In this paper, we ask whether language models can directly learn to generate watermarked text, which would have significant implications for the real-world deployment of watermarks. First, learned watermarks could be used to build open models that naturally generate watermarked text, enabling watermarking for open models, where users can control the decoding procedure. Second, if watermarking is used to determine the provenance of generated text, an adversary can hurt the reputation of a victim model by spoofing its watermark and generating damaging watermarked text. To investigate the learnability of watermarks, we propose watermark distillation, which trains a student model to behave like a teacher model that uses decoding-based watermarking. We test our approach on three decoding-based watermarking strategies and various hyperparameter settings, finding that models can learn to generate watermarked text with high detectability. We also find limitations to learnability, including the loss of watermarking capabilities under fine-tuning on normal text and high sample complexity when learning low-distortion watermarks.
This paper has not been read by Pith yet.
Forward citations
Cited by 7 Pith papers
-
RLSpoofer: A Lightweight Evaluator for LLM Watermark Spoofing Resilience
RLSpoofer trains a 4B model on 100 watermarked paraphrase pairs to spoof PF watermarks at 62% success rate, far exceeding baselines trained on up to 10,000 samples.
-
Optimal Multi-bit Generative Watermarking Schemes Under Worst-Case False-Alarm Constraints
Two new constructions for multi-bit generative watermarking attain the established lower bound on miss-detection probability under worst-case false-alarm constraints, fully characterizing optimal performance via linea...
-
TextSeal: A Localized LLM Watermark for Provenance & Distillation Protection
TextSeal provides a localized, distortion-free LLM watermark that enables provenance tracking and distillation detection while preserving performance and text quality.
-
TextSeal: A Localized LLM Watermark for Provenance & Distillation Protection
TextSeal provides a localized, distortion-free LLM watermark that outperforms baselines in detection strength, remains effective in mixed human-AI text, preserves model performance, and transfers through distillation ...
-
Can AI-Generated Text be Reliably Detected?
Recursive paraphrasing attacks substantially lower detection rates for multiple AI text detectors with only minor quality loss, while a theoretical analysis ties best-case AUROC to total variation distance between hum...
-
Position: LLM Watermarking Should Align Stakeholders' Incentives for Practical Adoption
LLM watermarking adoption is limited by misaligned stakeholder incentives; incentive-aligned approaches such as in-context watermarking can enable practical use in targeted domains like education and peer review.
-
AI Safety Landscape for Large Language Models: Taxonomy, State-of-the-art, and Future Directions
The paper introduces a taxonomy of AI safety for LLMs organized into Trustworthy AI, Responsible AI, and Safe AI perspectives, accompanied by a review of state-of-the-art methods, challenges, and future directions.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.