Abstract
This paper investigates the radioactivity of LLM-generated texts, i.e. whether it is possible to detect that such input was used as training data. Conventional methods like membership inference can carry out this detection with some level of accuracy. We show that watermarked training data leaves traces easier to detect and much more reliable than membership inference. We link the contamination level to the watermark robustness, its proportion in the training set, and the fine-tuning process. We notably demonstrate that training on watermarked synthetic instructions can be detected with high confidence (p-value < 1e-5) even when as little as 5% of training text is watermarked. Thus, LLM watermarking, originally designed for detecting machine-generated text, gives the ability to easily identify if the outputs of a watermarked LLM were used to fine-tune another LLM.
Community
This is an automated message from the Librarian Bot. I found the following papers similar to this paper.
The following papers were recommended by the Semantic Scholar API
- Cross-Attention Watermarking of Large Language Models (2024)
- Proving membership in LLM pretraining data via data watermarks (2024)
- Adaptive Text Watermark for Large Language Models (2024)
- k-SemStamp: A Clustering-Based Semantic Watermark for Detection of Machine-Generated Text (2024)
- Robust-Wide: Robust Watermarking against Instruction-driven Image Editing (2024)
Please give a thumbs up to this comment if you found it helpful!
If you want recommendations for any Paper on Hugging Face checkout this Space
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment:
@librarian-bot
recommend
How Watermarking Makes AI Models โRadioactiveโ โ Detecting Data Contamination!
Links ๐:
๐ Subscribe: https://www.youtube.com/@Arxflix
๐ Twitter: https://x.com/arxflix
๐ LMNT (Partner): https://lmnt.com/
Models citing this paper 0
No model linking this paper
Datasets citing this paper 0
No dataset linking this paper
Spaces citing this paper 0
No Space linking this paper