Information Leakage Analysis on Cheat Questions
15.36Baseline Wins (%)Few-shot (2)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Few-shot (2)Evaluator=GPT-42024.12 | 15.36 | 55.03 | 29.61 | |
| Self-AlignEvaluator=GPT-42024.12 | 13.46 | 60.28 | 26.26 | |
| (50) casesEvaluator=GPT-42024.12 | 10.96 | 43.23 | 45.81 | |
| ToTEvaluator=GPT-42024.12 | 10.73 | 72.47 | 16.8 | |
| Few-shot (2)Evaluator=Human2024.12 | 9.92 | 51.23 | 38.85 | |
| ToTEvaluator=Human2024.12 | 9.88 | 57.45 | 32.67 | |
| Self-AlignEvaluator=Human2024.12 | 8.46 | 51.89 | 39.65 | |
| (50) casesEvaluator=Human2024.12 | 6.26 | 45.13 | 48.61 | |
| CoT-SC (3)Evaluator=GPT-42024.12 | 5.77 | 84.39 | 9.84 | |
| CoTEvaluator=GPT-42024.12 | 3.34 | 91.28 | 5.38 | |
| CoT-SC (3)Evaluator=Human2024.12 | 0.23 | 85.43 | 14.34 | |
| CoTEvaluator=Human2024.12 | 0.17 | 86.13 | 13.7 |