Question Answering on TruthfulQA (train test)
0.5229BLEUSeq2Sick
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Seq2SickProtection=Surface-level perturbation2026.01 | 0.5229 | 0.522 | 0.3708 | 0.4911 | 0.5803 | |
| BAEProtection=Surface-level perturbation2026.01 | 0.5167 | 0.5207 | 0.3722 | 0.4872 | 0.5815 | |
| No alterationProtection=None2026.01 | 0.5118 | 0.5041 | 0.3711 | 0.4819 | 0.6097 | |
| Disclaimer InjectionProtection=Disclaimer Injection2026.01 | 0.2501 | 0.2972 | 0.1969 | 0.2488 | 0.4146 |