LLM win-rate estimation ranking on LLM benchmark (Appendix)
1Spearman CorrelationAIPW-EM
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| AIPW-EMn=252026.04 | 1 | 100 | 0 | |
| AutoCaln=252026.04 | 1 | 100 | 0 | |
| Labeled-onlyn=252026.04 | 1 | 100 | 0 | |
| LinearCaln=252026.04 | 1 | 100 | 0 | |
| PPI++n=252026.04 | 1 | 100 | 0 | |
| Labeled-onlyn=502026.04 | 1 | 100 | 0 | |
| PPI++n=502026.04 | 0.75 | 87.5 | 0.003 | |
| AIPW-EMn=502026.04 | 0.5 | 75 | 0.006 | |
| AutoCaln=502026.04 | 0.5 | 75 | 0.006 | |
| LinearCaln=502026.04 | 0.5 | 75 | 0.006 | |
| AIPWn=252026.04 | 0.25 | 62.5 | 0.01 | |
| PPIn=252026.04 | 0.25 | 62.5 | 0.01 | |
| AIPWn=502026.04 | 0.25 | 62.5 | 0.01 | |
| PPIn=502026.04 | 0 | 50 | 0.013 |