Overall Certainty Distortion Analysis on SPICED and MIMIC
49.4Average MetricLlama 3.3
Evaluation Results
| Method | Links | |
|---|---|---|
| Llama 3.3size=70B, sampling=top-p=0.9, T=12026.06 | 49.4 | |
| Gemini 3.1version=fl, sampling=top-p=0.9, T=12026.06 | 46.7 | |
| Claude 4.5version=H, sampling=top-p=0.9, T=12026.06 | 43 | |
| Qwen3size=8B, sampling=top-p=0.9, T=12026.06 | 37.5 | |
| gpt-5-nanosampling=top-p=0.9, T=12026.06 | 30 |