Performance Trade-offs of Watermarking Large Language Models
Preprint
Abstract
Amidst growing concerns of large language models (LLMs) being misused for generating misinformation or completing homework assignments, watermarking has emerged as an effective solution for distinguishing human-written and LLM-generated text. A prominent watermarking strategy is to embed a signal into generated text by upsampling a (pseudorandomly-chosen) subset of tokens at every generation step. Although this signal is imperceptible to a human reader, it is detectable through statistical testing. However, implanting such signals alters the model's output distribution and can have unintended effects when watermarked LLMs are used for downstream applications. In this work, we evaluate the performance of watermarked LLMs on a diverse suite of tasks, including text classification, textual entailment, reasoning, question answering, translation, summarization, and language modeling. We find that watermarking has negligible impact on the performance of tasks posed as k-class classification problems in the average case. However, the accuracy can plummet to that of a random classifier for some scenarios (that occur with non-negligible probability). Tasks that are cast as multiple-choice questions and short-form generation are surprisingly unaffected by watermarking. For long-form generation tasks, including summarization and translation, we see a drop of 15-20% in the performance due to watermarking. Our findings highlight the trade-offs that users should be cognizant of when using watermarked models, and point to cases where future research could improve existing trade-offs.Keywords
All Related Versions
This publication has 13 references indexed in Scilit:
- Watermarking Conditional Text Generation for AI Detection: Unveiling Challenges and a Semantic-Aware Watermark RemedyProceedings of the AAAI Conference on Artificial Intelligence, 2024
- SemStamp: A Semantic Watermark with Paraphrastic Robustness for Text Generation2024
- Three Bricks to Consolidate Watermarks for Large Language Models2023
- Robust Multi-bit Natural Language Watermarking through Invariant Features2023
- Frustratingly Easy Edit-based Linguistic Steganography with a Masked Language Model2021
- CoQA: A Conversational Question Answering ChallengeTransactions of the Association for Computational Linguistics, 2019
- Don’t Give Me the Details, Just the Summary! Topic-Aware Convolutional Neural Networks for Extreme Summarization2018
- RACE: Large-scale ReAding Comprehension Dataset From Examinations2017
- SQuAD: 100,000+ Questions for Machine Comprehension of Text2016
- Information Hiding Techniques for Steganography and Digital WatermakingEDPACS, 2000