Persona Induction on AOL (test)
71.8Coherencepi_theta (Ours)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| pi_theta (Ours)Backbone=Gemma3-27B2026.04 | 71.8 | 94.4 | 79.1 | 81.7 | |
| GPT-oss-120bEvaluation protocol=prompt-only2026.04 | 61 | 95.5 | 65.4 | 73.9 | |
| GPT-5.1Evaluation protocol=prompt-only2026.04 | 60.5 | 94.2 | 60.7 | 71.8 | |
| PersonaXsBackbone=Gemma3-27B2026.04 | 59.7 | 86.5 | 59.9 | 68.6 | |
| PersonaXrBackbone=Gemma3-27B2026.04 | 59.4 | 71.5 | 42.8 | 57.8 | |
| Qwen3-80bEvaluation protocol=prompt-only2026.04 | 57.3 | 83.5 | 54.6 | 65.1 | |
| Claude-4.5Evaluation protocol=prompt-only2026.04 | 56.3 | 95.9 | 63.5 | 71.8 |