Alignment with Human Preferences on Chatbot Arena English-only
91.67Spearman CorrelationAuto-Arena
Evaluation Results
| Method | Links | |
|---|---|---|
| Auto-ArenaPeer Battles=included, Committee Discussions=included2024.05 | 91.67 | |
| Auto-ArenaCommittee Discussions=w/o2024.05 | 88.33 | |
| Auto-ArenaPeer Battles=w/o2024.05 | 86.67 | |
| Arena-Hard2024.05 | 85.71 | |
| MT-Bench2024.05 | 82.86 | |
| LC-AlpacaEval2024.05 | 82.14 | |
| MMLU2024.05 | 56.36 | |
| GPQA2024.05 | 36.84 | |
| OpenLLM2024.05 | -15.39 |