User-aligned task completion on KitchenAmbig (OOD)
87.3AccuracyAdaptive CLIPR
Evaluation Results
| Method | Links | |
|---|---|---|
| Adaptive CLIPRLLM engine=Claude Sonnet 4.5, alpha=1.5, f=52026.05 | 87.3 | |
| GATELLM engine=Claude Sonnet 4.52026.05 | 85.3 | |
| CIPHERLLM engine=Claude Sonnet 4.52026.05 | 80.5 |