CARE uses guard-model detection, token-buffer rollback, and self-reflective prompting to reduce harmful responses while preserving response quality.
If you have a specific reason for wanting to reduce slipperiness, please provide more context so I can offer appropriate advice
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
-
CARE: Decoding Time Safety Alignment via Rollback and Introspection Intervention
CARE uses guard-model detection, token-buffer rollback, and self-reflective prompting to reduce harmful responses while preserving response quality.