Conversational mimicry, especially reuse of prompt phrases, can make unwatermarked LLM text test positive for watermarks, and the paper argues current watermark configurations are too permissive.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Watermarking Needs Input Repetition Masking
Conversational mimicry, especially reuse of prompt phrases, can make unwatermarked LLM text test positive for watermarks, and the paper argues current watermark configurations are too permissive.