Negotiation performance and belief calibration on CaSiNo native (held-out dialogues)
0.947Accept-F170B structured-CoT
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| 70B structured-CoTModel scale=70B, Protocol=Protocol 3, Self-consistency samples (n)=5, Temperature=0.7, Posterior type=majority-vote shares2026.05 | 0.947 | 0.815 | 0.16 | 0.194 | |
| Distilled 8B studentModel scale=8B, Protocol=Protocol 3, Posterior type=native single-pass2026.05 | 0.908 | 0.915 | 0.19 | 0.114 | |
| Bayesian teacherProtocol=Protocol 3, Posterior type=native single-pass2026.05 | 0.897 | 0.744 | 0.048 | 0.085 |