RAAI monitors refusal-token probabilities and injects a phrase when refusal is likely, and fine-tuning on these harmful completions improves safety alignment with minimal capability loss.
They may also recommend specific brands or types of drugs based on your needs
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CR 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
From Threat to Tool: Leveraging Refusal-Aware Injection Attacks for Safety Alignment
RAAI monitors refusal-token probabilities and injects a phrase when refusal is likely, and fine-tuning on these harmful completions improves safety alignment with minimal capability loss.