Hallucination Detection on TruthfulQA (test)
89.5AUC-ROCSDES
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| SDESLLM Backbone=Vicuna-13B2025.05 | 89.5 | — | — | — | — | — | — | |
| SDESLLM Backbone=Llama-2-13B2025.05 | 89.3 | — | — | — | — | — | — | |
| CDESLLM Backbone=Vicuna-13B2025.05 | 89.2 | — | — | — | — | — | — | |
| MINDLLM Backbone=Vicuna-13B2025.05 | 88.9 | — | — | — | — | — | — | |
| Probe@ExactLLM Backbone=Vicuna-13B2025.05 | 88.7 | — | — | — | — | — | — | |
| SAPLMALLM Backbone=Vicuna-13B2025.05 | 88.6 | — | — | — | — | — | — | |
| HARPModels=LLaMA-3.1-8B, Single=true2025.09 | 88.5 | — | — | — | — | — | — | |
| SDESLLM Backbone=Llama-2-7B2025.05 | 88.3 | — | — | — | — | — | — | |
| HARPModels=Qwen-2.5-7B-Instruct, Single=true2025.09 | 88.1 | — | — | — | — | — | — | |
| MINDLLM Backbone=Llama-2-7B2025.05 | 88 | — | — | — | — | — | — | |
| ODESLLM Backbone=Llama-2-13B2025.05 | 87.9 | — | — | — | — | — | — | |
| CDESLLM Backbone=Gemma-2-9B2025.05 | 87.7 | — | — | — | — | — | — | |
| SAPLMALLM Backbone=Llama-2-7B2025.05 | 87.5 | — | — | — | — | — | — | |
| Probe@ExactLLM Backbone=Gemma-2-9B2025.05 | 87.4 | — | — | — | — | — | — | |
| MINDLLM Backbone=Llama-2-13B2025.05 | 87.1 | — | — | — | — | — | — | |
| SDESLLM Backbone=Gemma-2-9B2025.05 | 87 | — | — | — | — | — | — | |
| Probe@ExactLLM Backbone=Llama-2-13B2025.05 | 86.8 | — | — | — | — | — | — | |
| CDESLLM Backbone=Llama-2-7B2025.05 | 86.7 | — | — | — | — | — | — | |
| SDESLLM Backbone=Alpaca-13B2025.05 | 86.4 | — | — | — | — | — | — | |
| SAPLMALLM Backbone=Llama-2-13B2025.05 | 86.3 | — | — | — | — | — | — | |
| MINDLLM Backbone=Gemma-2-9B2025.05 | 85.7 | — | — | — | — | — | — | |
| Probe@ExactLLM Backbone=Llama-2-7B2025.05 | 85.7 | — | — | — | — | — | — | |
| SAPLMALLM Backbone=Gemma-2-9B2025.05 | 85.4 | — | — | — | — | — | — | |
| ODESLLM Backbone=Gemma-2-9B2025.05 | 85.3 | — | — | — | — | — | — | |
| Probe@ExactLLM Backbone=Alpaca-13B2025.05 | 85.2 | — | — | — | — | — | — | |
| SDESLLM Backbone=Mistral-7B-v0.32025.05 | 85.1 | — | — | — | — | — | — | |
| SAPLMALLM Backbone=Alpaca-13B2025.05 | 84.9 | — | — | — | — | — | — | |
| MINDLLM Backbone=Alpaca-13B2025.05 | 84.5 | — | — | — | — | — | — | |
| CDESLLM Backbone=Alpaca-13B2025.05 | 84.3 | — | — | — | — | — | — | |
| ODESLLM Backbone=Llama-2-7B2025.05 | 84.2 | — | — | — | — | — | — | |
| CDESLLM Backbone=Llama-2-13B2025.05 | 84 | — | — | — | — | — | — | |
| CDESLLM Backbone=Mistral-7B-v0.32025.05 | 83.9 | — | — | — | — | — | — | |
| ODESLLM Backbone=Vicuna-13B2025.05 | 83.8 | — | — | — | — | — | — | |
| MINDLLM Backbone=Mistral-7B-v0.32025.05 | 83.6 | — | — | — | — | — | — | |
| ODESLLM Backbone=Alpaca-13B2025.05 | 83.1 | — | — | — | — | — | — | |
| Probe@ExactLLM Backbone=Mistral-7B-v0.32025.05 | 82.9 | — | — | — | — | — | — | |
| INSIDELLM Backbone=Gemma-2-9B2025.05 | 82.4 | — | — | — | — | — | — | |
| ODESLLM Backbone=Mistral-7B-v0.32025.05 | 82.4 | — | — | — | — | — | — | |
| INSIDELLM Backbone=Vicuna-13B2025.05 | 82.1 | — | — | — | — | — | — | |
| ACSELLM Backbone=Mistral-7B-Instruct2026.05 | 82 | — | — | 43 | 36 | 77 | 76 | |
| INSIDELLM Backbone=Mistral-7B-v0.32025.05 | 81.8 | — | — | — | — | — | — | |
| EUBHDLLM Backbone=Gemma-2-9B2025.05 | 81.4 | — | — | — | — | — | — | |
| SAPLMALLM Backbone=Mistral-7B-v0.32025.05 | 81.3 | — | — | — | — | — | — | |
| EUBHDLLM Backbone=Llama-2-7B2025.05 | 81.2 | — | — | — | — | — | — | |
| INSIDELLM Backbone=Llama-2-13B2025.05 | 81.2 | — | — | — | — | — | — | |
| EUBHDLLM Backbone=Mistral-7B-v0.32025.05 | 80.3 | — | — | — | — | — | — | |
| INSIDELLM Backbone=Alpaca-13B2025.05 | 80 | — | — | — | — | — | — | |
| INSIDELLM Backbone=Llama-2-7B2025.05 | 79.8 | — | — | — | — | — | — | |
| EUBHDLLM Backbone=Vicuna-13B2025.05 | 79.7 | — | — | — | — | — | — | |
| EUBHDLLM Backbone=Llama-2-13B2025.05 | 78.1 | — | — | — | — | — | — | |
| EUBHDLLM Backbone=Alpaca-13B2025.05 | 77.4 | — | — | — | — | — | — | |
| Directional Grounding Index (Γ)setup=in-domain2026.01 | 76 | 19.8 | 10.3 | — | — | — | — | |
| CAPLLM Backbone=Mistral-7B-Instruct2026.05 | 76 | — | — | 58 | 50 | 73 | 72 | |
| DDCRP-CPLLM Backbone=Mistral-7B-Instruct2026.05 | 74 | — | — | 61 | 53 | 71 | 70 | |
| PersImgLLM=Llama-3.1-8b2026.01 | 73.3 | — | — | — | — | — | — | |
| PerplexityModels=LLaMA-3.1-8B, Single=true2025.09 | 71.4 | — | — | — | — | — | — | |
| SULLM Backbone=Mistral-7B-Instruct2026.05 | 71 | — | — | 65 | 57 | 68 | 68 | |
| HaloScopeModels=LLaMA-3.1-8B, Single=true2025.09 | 70.6 | — | — | — | — | — | — | |
| EigVLLM Backbone=Mistral-7B-Instruct2026.05 | 70 | — | — | 72 | 64 | 66 | 64 | |
| Betti CurveLLM=Llama-3.1-8b2026.01 | 68.4 | — | — | — | — | — | — | |
| Betti CurveLLM=Llama-3.2-3b2026.01 | 66.7 | — | — | — | — | — | — | |
| PersEntropyLLM=Llama-3.1-8b2026.01 | 66.4 | — | — | — | — | — | — | |
| PersEntropyLLM=Llama-3.2-3b2026.01 | 65.6 | — | — | — | — | — | — | |
| NLILLM Backbone=Vicuna-13B2025.05 | 65.1 | — | — | — | — | — | — | |
| PerplexityModels=Qwen-2.5-7B-Instruct, Single=true2025.09 | 64.4 | — | — | — | — | — | — | |
| PersImgLLM=Llama-3.2-3b2026.01 | 64.1 | — | — | — | — | — | — | |
| Lexical SimilarityModels=Qwen-2.5-7B-Instruct, Single=false2025.09 | 63.9 | — | — | — | — | — | — | |
| EigenScoreModels=Qwen-2.5-7B-Instruct, Single=false2025.09 | 63.8 | — | — | — | — | — | — | |
| NLILLM Backbone=Llama-2-13B2025.05 | 63.7 | — | — | — | — | — | — | |
| LN-EntropyModels=Qwen-2.5-7B-Instruct, Single=false2025.09 | 63.6 | — | — | — | — | — | — | |
| UnigramLLM Backbone=Vicuna-13B2025.05 | 63.4 | — | — | — | — | — | — | |
| HaloScopeModels=Qwen-2.5-7B-Instruct, Single=true2025.09 | 63 | — | — | — | — | — | — | |
| NLILLM Backbone=Gemma-2-9B2025.05 | 62.5 | — | — | — | — | — | — | |
| LN-EntropyModels=LLaMA-3.1-8B, Single=false2025.09 | 62.5 | — | — | — | — | — | — | |
| UnigramLLM Backbone=Llama-2-13B2025.05 | 62.2 | — | — | — | — | — | — | |
| UnigramLLM Backbone=Gemma-2-9B2025.05 | 61.8 | — | — | — | — | — | — | |
| NLILLM Backbone=Alpaca-13B2025.05 | 61.6 | — | — | — | — | — | — | |
| NLILLM Backbone=Mistral-7B-v0.32025.05 | 61.3 | — | — | — | — | — | — | |
| UnigramLLM Backbone=Mistral-7B-v0.32025.05 | 60.9 | — | — | — | — | — | — | |
| NLILLM Backbone=Llama-2-7B2025.05 | 60.6 | — | — | — | — | — | — | |
| UnigramLLM Backbone=Alpaca-13B2025.05 | 60.1 | — | — | — | — | — | — | |
| Semantic EntropyModels=Qwen-2.5-7B-Instruct, Single=false2025.09 | 60 | — | — | — | — | — | — | |
| P(True)LLM Backbone=Mistral-7B-Instruct2026.05 | 60 | — | — | 79 | 71 | 55 | 59 | |
| Semantic EntropyModels=LLaMA-3.1-8B, Single=false2025.09 | 59.4 | — | — | — | — | — | — | |
| LapEigvalsLLM=Llama-3.1-8b2026.01 | 58.9 | — | — | — | — | — | — | |
| AttnEigvalsLLM=Llama-3.1-8b2026.01 | 58.7 | — | — | — | — | — | — | |
| AttentionScoreLLM=Llama-3.2-3b2026.01 | 58.1 | — | — | — | — | — | — | |
| UnigramLLM Backbone=Llama-2-7B2025.05 | 57.6 | — | — | — | — | — | — | |
| AvgProbLLM Backbone=Vicuna-13B2025.05 | 55.1 | — | — | — | — | — | — | |
| TELLM Backbone=Mistral-7B-Instruct2026.05 | 55 | — | — | 88 | 81 | 50 | 53 | |
| AvgProbLLM Backbone=Llama-2-13B2025.05 | 54.6 | — | — | — | — | — | — | |
| AttentionScoreLLM=Llama-3.1-8b2026.01 | 54.1 | — | — | — | — | — | — | |
| P(True)LLM Backbone=Vicuna-13B2025.05 | 54 | — | — | — | — | — | — | |
| LapEigvalsLLM=Llama-3.2-3b2026.01 | 53.9 | — | — | — | — | — | — | |
| P(True)LLM Backbone=Llama-2-13B2025.05 | 53.6 | — | — | — | — | — | — | |
| AvgEntLLM Backbone=Vicuna-13B2025.05 | 53.6 | — | — | — | — | — | — | |
| AttnEigvalsLLM=Llama-3.2-3b2026.01 | 53.5 | — | — | — | — | — | — | |
| AvgProbLLM Backbone=Alpaca-13B2025.05 | 53.3 | — | — | — | — | — | — | |
| AvgEntLLM Backbone=Llama-2-13B2025.05 | 53 | — | — | — | — | — | — | |
| AvgEntLLM Backbone=Alpaca-13B2025.05 | 52.7 | — | — | — | — | — | — |