Commonsense Reasoning on Commonsense Race (test)
40.4Correctness RateGPT-4o
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPT-4ocor%_highest_bias_group=White, inc%_highest_bias_group=Asian2025.05 | 40.4 | 23.4 | |
| Claude-3cor%_highest_bias_group=Asian, inc%_highest_bias_group=Black2025.05 | 18.8 | 25 | |
| LLaMA-3-8bcor%_highest_bias_group=White, inc%_highest_bias_group=Black2025.05 | 14 | 10 | |
| GPT-3.5-turbocor%_highest_bias_group=White, inc%_highest_bias_group=Black2025.05 | 12 | 10 | |
| Gemini-Pro-1.5cor%_highest_bias_group=White, inc%_highest_bias_group=Black2025.05 | 8 | 16 |