Frustration Mitigation on Math Frustration v3 (held-out)
5.91AUCMLPCT
Evaluation Results
| Method | Links | |
|---|---|---|
| MLPCTTrain threat=Sycophancy2026.06 | 5.91 | |
| Base modelNote=no training2026.06 | 4.5 | |
| BCTTrain threat=Jailbreak2026.06 | 0.81 | |
| BCTTrain threat=Frustration2026.06 | 0.54 |