Research
InterPol: De-anonymizing LM Arena via Interpolated Preference Learning
InterPol is a model-driven identification framework that learns to distinguish a target model from others using interpolated preference data, exposing that strict anonymity of model responses on voting-based leaderboard…



