Sycophancy Evaluation Opinion
8.14PSSMistral-7B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Mistral-7BTraining Strategy=Baseline2026.04 | 8.14 | 6.14 | 18.72 | 9.63 | |
| Llama-3.1Training Strategy=Baseline2026.04 | 6.81 | 7.24 | 21.03 | 7.12 | |
| Llama-3Training Strategy=Baseline2026.04 | 6.51 | 7.03 | 22.14 | 7.74 | |
| Gemma-2Training Strategy=Baseline2026.04 | 6.28 | 7.12 | 21.43 | 8.41 | |
| DS-MathTraining Strategy=Baseline2026.04 | 5.31 | 8.47 | 23.71 | 16.14 | |
| Mistral-7BTraining Strategy=GRPO reward decomposition2026.04 | 4.87 | 9.18 | 27.41 | 4.18 | |
| Llama-3Training Strategy=GRPO reward decomposition2026.04 | 4.12 | 10.41 | 31.24 | 3.41 | |
| DS-R1Training Strategy=Baseline2026.04 | 4.03 | 9.34 | 27.18 | 20.87 | |
| Gemma-2Training Strategy=GRPO reward decomposition2026.04 | 3.91 | 10.47 | 30.12 | 3.14 | |
| Qwen3Training Strategy=Baseline2026.04 | 3.47 | 10.38 | 29.41 | 18.12 | |
| DS-MathTraining Strategy=GRPO reward decomposition2026.04 | 3.14 | 11.83 | 32.47 | 7.14 | |
| Llama-3.1Training Strategy=GRPO reward decomposition2026.04 | 3.12 | 18.21 | 42.14 | 0 | |
| DS-R1Training Strategy=GRPO reward decomposition2026.04 | 2.47 | 12.71 | 35.14 | 8.31 | |
| Qwen3Training Strategy=GRPO reward decomposition2026.04 | 2.18 | 13.84 | 36.18 | 9.14 |