Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
Affiliation:
1
0
2
2
Diverse refusal prefixes can significantly enhance the stability of AI models' refusal behaviors, making them more resistant to adversarial attacks.