Response-level Hallucination Detection on HDM-Bench
76.3AUROCHallucinationProbes
Evaluation Results
| Method | Links | |
|---|---|---|
| HallucinationProbesDetector category=White-box, Probe=LoRA-KL probe on LLaMA-3.1-8B (lambda_KL = 0.05)2026.04 | 76.3 | |
| RAGognizerDetector category=White-box, Proxy generator=Qwen3-4B-Instruct-25072026.04 | 72.72 | |
| MiniCheck-7BDetector category=Black-box2026.04 | 71.7 | |
| HDM-2-3BDetector category=Black-box, Same-dataset evaluation=true2026.04 | 69.62 | |
| SelfCheckGPT (NLI)Detector category=Black-box2026.04 | 68.44 | |
| DeBERTa-v3 (Ent.)Detector category=Black-box2026.04 | 62.6 | |
| DeBERTa-v3 (Con.)Detector category=Black-box2026.04 | 60.9 | |
| LettuceDetect-LDetector category=Black-box2026.04 | 60.85 | |
| HHEM-2.1-OpenDetector category=Black-box2026.04 | 59.97 | |
| INSIDE (EigenScore)Detector category=White-box, Proxy generator=Qwen3-4B-Instruct-25072026.04 | 56.35 | |
| Semantic EntropyDetector category=White-box2026.04 | 55.81 |