Comment Prediction on HealthBench (test)
7.31Medical Chat Annotation ScoreFeedback-to-Rubrics
Evaluation Results
| Method | Links | |
|---|---|---|
| Feedback-to-RubricsMethod Type=Rubric-based (Comment-wise refinement), Backbone=Gemini 3.1 Pro preview2026.05 | 7.31 | |
| Iterative rubric refinementMethod Type=Rubric-based (Ablation), Backbone=Gemini 3.1 Pro preview2026.05 | 7.15 | |
| Top-3 RAG with LLMMethod Type=Retrieval-based, Backbone=Gemini 3.1 Pro preview2026.05 | 7.08 | |
| Initial rubricMethod Type=Rubric-based (Ablation), Backbone=Gemini 3.1 Pro preview2026.05 | 6.72 | |
| No rubricMethod Type=Rubric-based (Ablation), Backbone=Gemini 3.1 Pro preview2026.05 | 6.62 | |
| Top-1 retrievalMethod Type=Retrieval-based, Backbone=Gemini 3.1 Pro preview2026.05 | 0.2 |