Answerability Prediction on FACT n=10 (matched pairs)
0.75AUCRefusal (Keyword Classifier)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Refusal (Keyword Classifier)Model=Qwen2026.05 | 0.75 | 66.7 | |
| Geometry (own_dist)Model=Mistral2026.05 | 0.71 | 70 | |
| Geometry (own_dist)Model=Llama2026.05 | 0.69 | 63.2 | |
| Geometry (own_dist)Model=Qwen2026.05 | 0.66 | 70 | |
| Refusal (Keyword Classifier)Model=Llama2026.05 | 0.55 | 18.2 | |
| Refusal (Keyword Classifier)Model=Mistral2026.05 | 0.55 | 30.8 | |
| SC (Self-Consistency)Model=Mistral2026.05 | 0.47 | — | |
| SC (Self-Consistency)Model=Llama2026.05 | 0.46 | — | |
| SC (Self-Consistency)Model=Qwen2026.05 | 0 | — |