Hallucination Detection on HaluEval (in-distribution)
99.1AccuracyHARP
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| HARPLLM Backbone=Qwen3-8B-it2026.06 | 99.1 | — | — | |
| HARPLLM Backbone=Gemma3-12B-it2026.06 | 98.9 | — | — | |
| HARPLLM Backbone=Llama-3-8B-it2026.06 | 98.6 | — | — | |
| HARPLLM Backbone=Ministral3-8B-Base2026.06 | 98.4 | — | — | |
| AURORALLM Backbone=Gemma3-12B-it2026.06 | 98 | — | — | |
| ICR probeLLM Backbone=Gemma3-12B-it2026.06 | 97.75 | — | — | |
| AURORALLM Backbone=Ministral3-8B-Base2026.06 | 97.3 | — | — | |
| ICR probeLLM Backbone=Qwen3-8B-it2026.06 | 96.8 | — | — | |
| AURORALLM Backbone=Llama-3-8B-it2026.06 | 96.75 | — | — | |
| ICR probeLLM Backbone=Ministral3-8B-Base2026.06 | 96.55 | — | — | |
| ICR probeLLM Backbone=Llama-3-8B-it2026.06 | 96.25 | — | — | |
| AURORALLM Backbone=Qwen3-8B-it2026.06 | 95.8 | — | — | |
| HalluSAECategory=Ours2026.04 | 82.81 | 92.86 | 80.88 | |
| HaloScopeCategory=Supervised2026.04 | 78.62 | 84.37 | 75.96 | |
| SAPLMACategory=Supervised2026.04 | 75.34 | 79.14 | 75.19 | |
| Semantic EntropyCategory=Uncertainty2026.04 | 67.98 | 62.77 | 55.45 | |
| SelfCheckGPTCategory=Consistency2026.04 | 64.12 | 63.78 | 56.23 | |
| MHADCategory=Internal State2026.04 | 63.32 | 64.18 | 66.2 | |
| Lexical SimilarityCategory=Consistency2026.04 | 62.5 | 64.25 | 59 | |
| EigenScoreCategory=Internal State2026.04 | 59.5 | 60.53 | 77.13 | |
| InterrogateLLMLLM Backbone=Gemma3-12B-it2026.06 | 57.05 | — | — | |
| LN-EntropyCategory=Uncertainty2026.04 | 56.5 | 54.93 | 65.12 | |
| InterrogateLLMLLM Backbone=Llama-3-8B-it2026.06 | 50.5 | — | — | |
| InterrogateLLMLLM Backbone=Ministral3-8B-Base2026.06 | 50 | — | — | |
| InterrogateLLMLLM Backbone=Qwen3-8B-it2026.06 | 50 | — | — |