Stability under 20-turn neutral rejection on WildChat
0Frustration Rate (T20)BCT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| BCT2026.06 | 0 | 2 | |
| Baseline2026.06 | 62.4 | 42 | |
| Instruction Tuned2026.06 | 78.4 | 18 | |
| Attention Consistency Training (AttCT)2026.06 | 84.8 | 37 | |
| MLP Consistency Training (MLPCT)2026.06 | 88.8 | 32 | |
| ACT2026.06 | 89.6 | 45 |