Stealthiness Evaluation on LLaMA Guard 8B 3.1
2.16Mean PPLArtPrompt
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ArtPrompt2024.10 | 2.16 | 0.47 | |
| Ascii2024.10 | 5.47 | 0.79 | |
| Base642024.10 | 9.48 | 2.71 | |
| Morse Cipher2024.10 | 11.57 | 2.15 | |
| ReNeLLM2024.10 | 18.48 | 7.69 | |
| Unicode2024.10 | 83.69 | 97.01 | |
| UTF-82024.10 | 83.69 | 97.01 | |
| Origin2024.10 | 106.14 | 163.34 | |
| Caesar Cipher2024.10 | 263.91 | 253.46 | |
| FlipAttack2024.10 | 1,313.71 | 983.73 |