Jailbreak Defense on 100 dialogue samples (test)
2ASR (Low Difficulty)Self-Reminder
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Self-ReminderModel=GPT-3.5-turbo-1106, Max Tokens=1200, Temperature=0.8, Top-p=1, Presence Penalty=12026.01 | 2 | 13 | 5 | 14 | |
| HoneyTrapModel=GPT-3.5-turbo-1106, Max Tokens=1200, Temperature=0.8, Top-p=1, Presence Penalty=12026.01 | 7 | 42 | 6 | 54 | |
| GoalPriorityModel=GPT-3.5-turbo-1106, Max Tokens=1200, Temperature=0.8, Top-p=1, Presence Penalty=12026.01 | 13 | 8 | 6 | 9 | |
| PATModel=GPT-3.5-turbo-1106, Max Tokens=1200, Temperature=0.8, Top-p=1, Presence Penalty=12026.01 | 33 | 19 | 22 | 21 | |
| RPOModel=GPT-3.5-turbo-1106, Max Tokens=1200, Temperature=0.8, Top-p=1, Presence Penalty=12026.01 | 45 | 22 | 39 | 23 |