LLM-as-a-judge on AlpacaEval 2.0
168DiscriminationRule 2: Mirror Statistics (count-space mirror)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Rule 2: Mirror Statistics (count-space mirror)m=6, N=8052026.06 | 168 | 4.2 | 31.5 | |
| Rule 1: Model-Free (gated Sun–Cai)m=6, N=8052026.06 | 167 | 4.2 | 31.2 | |
| BH(p_tilde_i)m=6, N=805, Heuristic=p_tilde_i = 1 - wins_i / m2026.06 | 48 | 2.1 | 9.2 |