Authorship Discernment on Hebrew judicial reasoning (test)
84.3AccuracyGround truth
Evaluation Results
| Method | Links | |
|---|---|---|
| Ground truthDescription=real vs other judges2026.04 | 84.3 | |
| Joint-Gemma-3Type=Non-personalized baseline2026.04 | 77 | |
| Mixed-Gemma-3Type=Non-personalized baseline2026.04 | 76.2 | |
| Gemini-3-ProType=Retrieval-augmented baseline2026.04 | 73.8 | |
| Vanilla-DictaLM-3Type=Non-personalized baseline2026.04 | 71.9 | |
| Gemma-3Type=Retrieval-augmented baseline2026.04 | 70.4 | |
| Vanilla-Gemma-3Type=Non-personalized baseline2026.04 | 70.3 | |
| CLMType=Personalized model2026.04 | 56.2 | |
| Random classifierSetting=Chance level2026.04 | 50 | |
| CoLAType=Personalized model2026.04 | 49.8 | |
| ITType=Personalized model, Protocol=Instruction-Tuning2026.04 | 49.6 |