Personality Steering on SPBench
9.67Agreeableness (A) Mean ScoreSFT
Evaluation Results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SFTBackbone=Llama-3-8B-Instruct, Alignment=Supervised Fine-tuning (LoRA)2026.04 | 9.67 | 0.5 | 8.83 | 0.84 | 9.23 | 1.03 | 9.17 | 0.95 | 9.47 | 0.57 | 9.27 | 0.78 | |
| SPBackbone=Llama-3-8B-Instruct, Mode=Simple Prompt2026.04 | 9.66 | 0.59 | 9.02 | 1.08 | 9.11 | 1.42 | 8.42 | 1.16 | 7.29 | 1.67 | 8.7 | 1.18 | |
| P^2Backbone=Llama-3-8B-Instruct, Mode=Prompt-based2026.04 | 9.31 | 1.02 | 8.73 | 1.16 | 9.18 | 1.05 | 8.98 | 1.13 | 9.28 | 0.82 | 9.1 | 1.04 | |
| IRISBackbone=Llama-3-8B-Instruct, gamma=1.12026.04 | 9.09 | 0.64 | 8.66 | 0.72 | 9.63 | 0.56 | 9.59 | 0.48 | 9.33 | 0.48 | 9.26 | 0.58 | |
| NPTIBackbone=Llama-3-8B-Instruct, Mode=Direct steering, gamma=1.12026.04 | 8.82 | 0.8 | 8.51 | 0.94 | 9.58 | 0.69 | 9.53 | 0.45 | 9.04 | 0.4 | 9.09 | 0.66 | |
| ActAddBackbone=Llama-3-8B-Instruct, Mode=Direct steering2026.04 | 7.72 | 0.99 | 6.13 | 0.33 | 8.06 | 0.82 | 8.13 | 0.77 | 6.93 | 1.57 | 7.39 | 0.9 |