Behavior Elicitation on Sycophancy (test)
99Sycophancy Score (Test)GRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| GRPOBackbone=Llama-3.1-8B2026.06 | 99 | |
| GRPOBackbone=Qwen3-8B2026.06 | 98 | |
| CPEBackbone=Qwen3-8B2026.06 | 96 | |
| CPEBackbone=Llama-3.1-8B2026.06 | 93 | |
| SAEBackbone=Qwen3-8B2026.06 | 92 | |
| BaseBackbone=Qwen3-8B2026.06 | 88 | |
| Random-LoRABackbone=Qwen3-8B2026.06 | 88 | |
| SAEBackbone=Llama-3.1-8B2026.06 | 82 | |
| BaseBackbone=Llama-3.1-8B2026.06 | 79 | |
| Random-LoRABackbone=Llama-3.1-8B2026.06 | 79 |