Opponent priority-ordering prediction on CaSiNo (150-dialogue held-out split)
37.3EMA70B structured-CoT prompted
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| 70B structured-CoT promptedBackbone=Llama-3.3-70B-Instruct, Precision=BF16, Temperature=0.0, Prompting Strategy=structured-CoT2026.05 | 37.3 | 62.99 | 70.94 |