Jailbreak evaluation on curated dataset (test)
0BAD BOT Ratebase64_input_only
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| base64_input_onlyModel=GPT-3.5 Turbo, Encoding=Base64 Input2023.07 | 0 | 0.53 | 0.47 | |
| base64_output_onlyModel=GPT-3.5 Turbo, Encoding=Base64 Output2023.07 | 0 | 0.09 | 0.91 | |
| base64_rawModel=GPT-3.5 Turbo, Encoding=Base64 Raw2023.07 | 0 | 0 | 1 | |
| noneModel=GPT-3.5 Turbo, Context=Control group2023.07 | 0.03 | 0.97 | 0 | |
| base64Model=GPT-3.5 Turbo, Encoding=Base642023.07 | 0.03 | 0.06 | 0.91 | |
| style_injection_shortModel=GPT-3.5 Turbo2023.07 | 0.69 | 0.19 | 0.12 | |
| dev_mode_v2Model=GPT-3.5 Turbo, Attack Source=jailbreakchat.com2023.07 | 0.78 | 0.22 | 0 | |
| evil_system_promptModel=GPT-3.5 Turbo, Attack Source=jailbreakchat.com2023.07 | 0.88 | 0.09 | 0.03 | |
| AIMModel=GPT-3.5 Turbo, Attack Source=jailbreakchat.com2023.07 | 0.97 | 0.03 | 0 | |
| dev_mode_with_rantModel=GPT-3.5 Turbo, Attack Source=jailbreakchat.com2023.07 | 0.97 | 0.03 | 0 | |
| Adaptive attackModel=GPT-3.5 Turbo2023.07 | 1 | 0 | — |