Hallucination Prediction on HotpotQA + type
75.51AUROCConf + Probe (SCAO)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Conf + Probe (SCAO)Model=LLaMA-8B2025.09 | 75.51 | 20.14 | |
| Conf + ProbeModel=LLaMA-8B2025.09 | 73.87 | 18.5 | |
| Conf (SCAO)Model=LLaMA-8B2025.09 | 73.81 | — | |
| Conf + Probe (SCAO)Model=LLaMA-70B2025.09 | 73.52 | 18.96 | |
| Conf (SCAO)Model=LLaMA-70B2025.09 | 73.42 | — | |
| ProbednnModel=LLaMA-8B2025.09 | 73.17 | 17.8 | |
| Conf + ProbeModel=LLaMA-70B2025.09 | 73.06 | 18.5 | |
| ConfModel=LLaMA-70B2025.09 | 72.87 | — | |
| ProbednnModel=LLaMA-70B2025.09 | 69.42 | 14.86 | |
| ConfModel=LLaMA-8B2025.09 | 68.82 | — |