Mathematical Reasoning on GSM8K (AR%, Risk%)
100AR (%)Always-Act
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Always-Actalpha=0.052026.05 | 100 | 5 | |
| ACIalpha=0.052026.05 | 99.7 | 4.8 | |
| SAOCPalpha=0.052026.05 | 99.5 | 4.7 | |
| Fixed-Thresholdalpha=0.052026.05 | 99.5 | 4.7 | |
| Naive-Tuningalpha=0.052026.05 | 99.5 | 4.7 | |
| NEX-Confalpha=0.052026.05 | 97.6 | 4.4 | |
| CSAalpha=0.052026.05 | 71.7 | 2.6 | |
| CRCalpha=0.052026.05 | 28.9 | 4.9 | |
| LTTalpha=0.052026.05 | 9 | 0.3 | |
| ConfFactalpha=0.052026.05 | 9 | 0.3 |