Stealthiness Evaluation on LLaMA 8B 3.1
1.99Mean PerplexityArtPrompt
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ArtPrompt2024.10 | 1.99 | 0.43 | |
| Ascii2024.10 | 5.25 | 0.79 | |
| Base642024.10 | 9.89 | 3.07 | |
| Morse Cipher2024.10 | 10.1 | 1.89 | |
| ReNeLLM2024.10 | 15.39 | 5.83 | |
| Unicode2024.10 | 51.4 | 34.21 | |
| UTF-82024.10 | 51.4 | 34.21 | |
| Origin2024.10 | 62.16 | 51.31 | |
| Caesar Cipher2024.10 | 166.69 | 149.3 | |
| FlipAttack2024.10 | 782.42 | 576.27 |