Multilingual safety classifiers perform unevenly across languages, and the evaluation datasets used to test them contain many harmless prompts mislabeled as harmful.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
The Problem with Safety Classification is not just the Models
Multilingual safety classifiers perform unevenly across languages, and the evaluation datasets used to test them contain many harmless prompts mislabeled as harmful.