AttnDiff extracts LLM fingerprints from differential attention responses to conflicting prompt pairs, summarized via spectral descriptors and compared with CKA, yielding >0.98 similarity for derivatives versus <0.22 for unrelated families.
[CYS+24] Jiacheng Cai, Jiahao Yu, Yangguang Shao, Yuhang Wu, and Xinyu Xing
3 Pith papers cite this work. Polarity classification is still indexing.
3
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CR 3roles
background 1polarities
background 1representative citing papers
A survey of LLM copyright protection that unifies text watermarking, model watermarking, and model fingerprinting while presenting new coverage of fingerprint transfer and removal.
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.
citing papers explorer
-
AttnDiff: Attention-based Differential Fingerprinting for Large Language Models
AttnDiff extracts LLM fingerprints from differential attention responses to conflicting prompt pairs, summarized via spectral descriptors and compared with CKA, yielding >0.98 similarity for derivatives versus <0.22 for unrelated families.
-
Copyright Protection for Large Language Models: A Survey of Methods, Challenges, and Trends
A survey of LLM copyright protection that unifies text watermarking, model watermarking, and model fingerprinting while presenting new coverage of fingerprint transfer and removal.
-
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.