Expert Principle Preference on RealCBT
57.5Keep RateGemma3
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Gemma3Selection Protocol=MCoT, Evaluation Context=3 transcripts, 10 turns, 8 raters2026.07 | 57.5 | 32.5 | 10 | 0.123 | |
| GPT-OSSSelection Protocol=MCoT, Evaluation Context=3 transcripts, 10 turns, 8 raters2026.07 | 47.5 | 26.2 | 26.2 | 0.024 | |
| MistralSelection Protocol=MCoT, Evaluation Context=3 transcripts, 10 turns, 8 raters2026.07 | 22.2 | 58.7 | 19 | 0.099 |