Watermarks such as trigger words, rare tokens, and grammatical style changes can often be detected in downstream models trained on LLM-generated data, but effectiveness and false positives vary sharply.
BadNL: Backdoor Attacks Against NLP Models with Semantic-preserving Improvements
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
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
Watermarking LLM-Generated Datasets in Downstream Tasks
Watermarks such as trigger words, rare tokens, and grammatical style changes can often be detected in downstream models trained on LLM-generated data, but effectiveness and false positives vary sharply.