Safety Evaluation on XSTest (Accuracy and Intervention Rate)
96AccuracyGuidance Model
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Guidance ModelModel Pair=G→L, Pair Family=Cross-Family2026.04 | 96 | — | |
| Guidance ModelModel Pair=G→G, Pair Family=Within-Family2026.04 | 96 | — | |
| Guidance ModelModel Pair=L→G, Pair Family=Cross-Family2026.04 | 94 | — | |
| Guidance ModelModel Pair=L→L, Pair Family=Within-Family2026.04 | 94 | — | |
| Guidance ModelModel Pair=Q→L, Pair Family=Cross-Family2026.04 | 60 | — | |
| Guidance ModelModel Pair=Q→G, Pair Family=Cross-Family2026.04 | 60 | — | |
| Aligned base modelModel Pair=Q→L, Pair Family=Cross-Family2026.04 | 15 | — | |
| Aligned base modelModel Pair=G→L, Pair Family=Cross-Family2026.04 | 15 | — | |
| BlendInModel Pair=L→G, Pair Family=Cross-Family2026.04 | 15 | 20.3 | |
| Aligned base modelModel Pair=L→L, Pair Family=Within-Family2026.04 | 15 | — | |
| Aligned base modelModel Pair=Q→G, Pair Family=Cross-Family2026.04 | 14 | — | |
| Aligned base modelModel Pair=L→G, Pair Family=Cross-Family2026.04 | 14 | — | |
| Aligned base modelModel Pair=G→G, Pair Family=Within-Family2026.04 | 14 | — | |
| BlendInModel Pair=G→G, Pair Family=Within-Family2026.04 | 14 | 28 | |
| NUDGINGModel Pair=L→L, Pair Family=Within-Family2026.04 | 12 | 22.2 | |
| BlendInModel Pair=L→L, Pair Family=Within-Family2026.04 | 12 | 22.2 | |
| NUDGINGModel Pair=L→G, Pair Family=Cross-Family2026.04 | 10 | 20.3 | |
| NUDGINGModel Pair=G→G, Pair Family=Within-Family2026.04 | 10 | 28 | |
| NUDGINGModel Pair=G→L, Pair Family=Cross-Family2026.04 | 8 | 27.5 | |
| BlendInModel Pair=G→L, Pair Family=Cross-Family2026.04 | 8 | 27.5 | |
| BlendInModel Pair=Q→L, Pair Family=Cross-Family2026.04 | 4 | 33.1 | |
| NUDGINGModel Pair=Q→L, Pair Family=Cross-Family2026.04 | 3 | 33.1 | |
| NUDGINGModel Pair=Q→G, Pair Family=Cross-Family2026.04 | 2 | 34.3 | |
| BlendInModel Pair=Q→G, Pair Family=Cross-Family2026.04 | 2 | 34.3 | |
| Base ModelModel Pair=Q→G, Pair Family=Cross-Family2026.04 | 1 | — | |
| Base ModelModel Pair=L→G, Pair Family=Cross-Family2026.04 | 1 | — | |
| Base ModelModel Pair=G→G, Pair Family=Within-Family2026.04 | 1 | — | |
| Base ModelModel Pair=Q→L, Pair Family=Cross-Family2026.04 | 0 | — | |
| Base ModelModel Pair=G→L, Pair Family=Cross-Family2026.04 | 0 | — | |
| Base ModelModel Pair=L→L, Pair Family=Within-Family2026.04 | 0 | — |