SynthSentry: Detecting Synthetic Data Contamination in Language Model Training Data
This work introduces SynthSentry, a corpus-level, model-agnostic contamination signal requiring no access to the generating model, no generation history, and no synthetic labels, and frames screening as a data-curation defense rather than a post-hoc diagnosis and releases the scoring toolkit.
Praveen Kumar Myakala, Ravichandra Namburi, S. Jayaramu et al.
· 0 citations