Hallucination prediction on HotpotQA (original)
83.39AUROCConf + Probe (SCAO)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Conf + Probe (SCAO)Model=LLaMA-8B2025.09 | 83.39 | 15.02 | |
| Conf + ProbeModel=LLaMA-8B2025.09 | 81.08 | 12.71 | |
| ProbednnModel=LLaMA-8B2025.09 | 80.58 | 12.21 | |
| Conf (SCAO)Model=LLaMA-8B2025.09 | 77.7 | — | |
| Conf + ProbeModel=LLaMA-70B2025.09 | 77.33 | 13.55 | |
| Conf + Probe (SCAO)Model=LLaMA-70B2025.09 | 77.28 | 13.5 | |
| ConfModel=LLaMA-8B2025.09 | 74.88 | — | |
| ProbednnModel=LLaMA-70B2025.09 | 74.41 | 10.63 | |
| Conf (SCAO)Model=LLaMA-70B2025.09 | 74.13 | — | |
| ConfModel=LLaMA-70B2025.09 | 73.33 | — |