For i.i.d. tokens, the maximum reliable multi-bit watermarking rate equals the entropy of the LLM output distribution, growing with allowed distortion; the claimed finite-token optimal scheme is not fully supported by the proof.
A watermark for large language models,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CR 1years
2025 1verdicts
REJECT 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
Distributional Information Embedding: A Framework for Multi-bit Watermarking
For i.i.d. tokens, the maximum reliable multi-bit watermarking rate equals the entropy of the LLM output distribution, growing with allowed distortion; the claimed finite-token optimal scheme is not fully supported by the proof.