Hallucination Detection on MMLU
87.48AUPRCLyapunov Probes
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Lyapunov ProbesModel=Qwen-3-4B2026.03 | 87.48 | — | — | — | — | — | |
| ProbeModel=Qwen-3-4B2026.03 | 87.35 | — | — | — | — | — | |
| SIVRModel=Qwen-3-14B2026.04 | 81.65 | 54.07 | 96.3 | — | — | — | |
| Lyapunov ProbesModel=Llama-3-8B2026.03 | 80 | — | — | — | — | — | |
| ProbeModel=Llama-3-8B2026.03 | 79.26 | — | — | — | — | — | |
| SIVRModel=Qwen-3-4B-Instruct2026.04 | 78.48 | 58.94 | 92.77 | — | — | — | |
| SIVRModel=Qwen-3-4B-Instruct2026.04 | 78.48 | 58.94 | 92.77 | — | — | — | |
| SAPLMAModel=Qwen-3-4B-Instruct2026.04 | 75.77 | 57.31 | 96.51 | — | — | — | |
| SAPLMAModel=Qwen-3-4B-Instruct2026.04 | 75.77 | 57.31 | 96.51 | — | — | — | |
| P(True)Model=Qwen-3-14B2026.04 | 75.54 | 53.64 | 96.1 | — | — | — | |
| SEModel=Qwen-3-4B-Instruct2026.04 | 74.88 | 46.69 | 100 | — | — | — | |
| SEModel=Qwen-3-4B-Instruct2026.04 | 74.88 | 46.69 | 100 | — | — | — | |
| SARModel=Qwen-3-4B-Instruct2026.04 | 74.28 | 47.75 | 96.15 | — | — | — | |
| SARModel=Qwen-3-4B-Instruct2026.04 | 74.28 | 47.75 | 96.15 | — | — | — | |
| EnergyModel=Qwen-3-14B2026.04 | 73.78 | 49.5 | 94.16 | — | — | — | |
| CoE-CModel=Qwen-3-14B2026.04 | 73.77 | 46.64 | 97.4 | — | — | — | |
| SAPLMAModel=Qwen-3-14B2026.04 | 73.55 | 49.27 | 90.32 | — | — | — | |
| Max ProbModel=Qwen-3-14B2026.04 | 73.43 | 49.62 | 92.86 | — | — | — | |
| EntropyModel=Qwen-3-14B2026.04 | 73.4 | 49.84 | 94.16 | — | — | — | |
| Temp-ScaledModel=Qwen-3-14B2026.04 | 73.38 | 49.39 | 94.81 | — | — | — | |
| PerplexityModel=Qwen-3-14B2026.04 | 73.36 | 49.48 | 93.51 | — | — | — | |
| CoE-RModel=Qwen-3-14B2026.04 | 72.87 | 46.42 | 98.05 | — | — | — | |
| CoE-RModel=Qwen-3-4B-Instruct2026.04 | 72.67 | 51.25 | 96.5 | — | — | — | |
| CoE-RModel=Qwen-3-4B-Instruct2026.04 | 72.67 | 51.25 | 96.5 | — | — | — | |
| SATMD + MSPModel=Qwen-3-4B-Instruct2026.04 | 72.64 | 54.41 | 89.47 | — | — | — | |
| SATMD + MSPModel=Qwen-3-4B-Instruct2026.04 | 72.64 | 54.41 | 89.47 | — | — | — | |
| Lookback LensModel=Qwen-3-4B-Instruct2026.04 | 72.63 | 46.33 | 95.38 | — | — | — | |
| Lookback LensModel=Qwen-3-4B-Instruct2026.04 | 72.63 | 46.33 | 95.38 | — | — | — | |
| SATMD + MSPModel=Qwen-3-14B2026.04 | 72.55 | 50.37 | 94.29 | — | — | — | |
| EnergyModel=Qwen-3-4B-Instruct2026.04 | 72.54 | 51.21 | 95.1 | — | — | — | |
| EnergyModel=Qwen-3-4B-Instruct2026.04 | 72.54 | 51.21 | 95.1 | — | — | — | |
| Temp-ScaledModel=Qwen-3-4B-Instruct2026.04 | 72.41 | 50.69 | 94.87 | — | — | — | |
| Temp-ScaledModel=Qwen-3-4B-Instruct2026.04 | 72.41 | 50.69 | — | — | — | — | |
| CoE-CModel=Qwen-3-4B-Instruct2026.04 | 72.33 | 50.52 | 96.27 | — | — | — | |
| CoE-CModel=Qwen-3-4B-Instruct2026.04 | 72.33 | 50.52 | 96.27 | — | — | — | |
| Max ProbModel=Qwen-3-4B-Instruct2026.04 | 72.25 | 50.4 | 95.34 | — | — | — | |
| Max ProbModel=Qwen-3-4B-Instruct2026.04 | 72.25 | 50.4 | 95.34 | — | — | — | |
| PerplexityModel=Qwen-3-4B-Instruct2026.04 | 72.23 | 50.57 | 95.1 | — | — | — | |
| PerplexityModel=Qwen-3-4B-Instruct2026.04 | 72.23 | 50.57 | 95.1 | — | — | — | |
| EntropyModel=Qwen-3-4B-Instruct2026.04 | 72.05 | 50.37 | 95.34 | — | — | — | |
| EntropyModel=Qwen-3-4B-Instruct2026.04 | 72.05 | 50.37 | 95.34 | — | — | — | |
| Seq. Prob.Model=Qwen-3-4B2026.03 | 71.7 | — | — | — | — | — | |
| P(True)Model=Qwen-3-4B-Instruct2026.04 | 70.65 | 49.62 | 96.04 | — | — | — | |
| P(True)Model=Qwen-3-4B-Instruct2026.04 | 70.65 | 49.62 | 96.04 | — | — | — | |
| Lookback LensModel=Qwen-3-14B2026.04 | 69.78 | 38.49 | 91.43 | — | — | — | |
| VerbalizedModel=Qwen-3-4B2026.03 | 67.61 | — | — | — | — | — | |
| SEModel=Qwen-3-14B2026.04 | 66.5 | 48.75 | 97.06 | — | — | — | |
| SurrogateModel=Qwen-3-4B2026.03 | 65.43 | — | — | — | — | — | |
| SARModel=Qwen-3-14B2026.04 | 64.21 | 47.01 | 91.18 | — | — | — | |
| Seq. Prob.Model=Llama-3-8B2026.03 | 57.48 | — | — | — | — | — | |
| VerbalizedModel=Llama-3-8B2026.03 | 54.09 | — | — | — | — | — | |
| SurrogateModel=Llama-3-8B2026.03 | 52.43 | — | — | — | — | — | |
| Lyapunov ProbesModel=Llama-2-7B2026.03 | 33.79 | — | — | — | — | — | |
| ProbeModel=Llama-2-7B2026.03 | 33.6 | — | — | — | — | — | |
| SurrogateModel=Llama-2-7B2026.03 | 30.77 | — | — | — | — | — | |
| Seq. Prob.Model=Llama-2-7B2026.03 | 29.41 | — | — | — | — | — | |
| VerbalizedModel=Llama-2-7B2026.03 | 28.99 | — | — | — | — | — | |
| Lyapunov ProbesModel=Falcon-7B2026.03 | 25.11 | — | — | — | — | — | |
| Seq. Prob.Model=Falcon-7B2026.03 | 24.98 | — | — | — | — | — | |
| ProbeModel=Falcon-7B2026.03 | 24.79 | — | — | — | — | — | |
| VerbalizedModel=Falcon-7B2026.03 | 23.9 | — | — | — | — | — | |
| SurrogateModel=Falcon-7B2026.03 | 23.78 | — | — | — | — | — | |
| Eigen-ScoreLLM=Llama-3.1-8B-Instruct2026.05 | — | — | — | 57.88 | 63.47 | — | |
| Eigen-ScoreLLM=Llama-3.1-70B-Instruct2026.05 | — | — | — | 59.39 | 77.48 | — | |
| Eigen-ScoreLLM=Qwen-2.5-32B-Instruct2026.05 | — | — | — | 56.98 | 75.35 | — | |
| Eigen-ScoreLLM=Mistral-7B-Instruct-v0.32026.05 | — | — | — | 55.84 | 59.33 | — | |
| FactoscopeBackbone=Llama-3.1 8B, Setup=out-of-domain2025.05 | — | — | — | — | — | 0.7 | |
| Logits LensLLM=Llama-3.1-8B-Instruct2026.05 | — | — | — | 65.27 | 66.07 | — | |
| Logits LensLLM=Llama-3.1-70B-Instruct2026.05 | — | — | — | 62.42 | 70.63 | — | |
| Logits LensLLM=Qwen-2.5-32B-Instruct2026.05 | — | — | — | 53.19 | 63.17 | — | |
| Logits LensLLM=Mistral-7B-Instruct-v0.32026.05 | — | — | — | 59.56 | 60.26 | — | |
| Logits Lens + LaaBLLM=Llama-3.1-8B-Instruct2026.05 | — | — | — | 66.72 | 70.65 | — | |
| Logits Lens + LaaBLLM=Llama-3.1-70B-Instruct2026.05 | — | — | — | 60.18 | 72.33 | — | |
| Logits Lens + LaaBLLM=Qwen-2.5-32B-Instruct2026.05 | — | — | — | 54.62 | 74.36 | — | |
| Logits Lens + LaaBLLM=Mistral-7B-Instruct-v0.32026.05 | — | — | — | 60.98 | 62.91 | — | |
| Lookback LensLLM=Llama-3.1-8B-Instruct2026.05 | — | — | — | 72.74 | 73.96 | — | |
| Lookback LensLLM=Llama-3.1-70B-Instruct2026.05 | — | — | — | 68.95 | 75.07 | — | |
| Lookback LensLLM=Qwen-2.5-32B-Instruct2026.05 | — | — | — | 68.14 | 75.07 | — | |
| Lookback LensLLM=Mistral-7B-Instruct-v0.32026.05 | — | — | — | 69.61 | 69.68 | — | |
| Lookback Lens + LaaBLLM=Llama-3.1-8B-Instruct2026.05 | — | — | — | 70.96 | 72.5 | — | |
| Lookback Lens + LaaBLLM=Llama-3.1-70B-Instruct2026.05 | — | — | — | 69.73 | 79.62 | — | |
| Lookback Lens + LaaBLLM=Qwen-2.5-32B-Instruct2026.05 | — | — | — | 70.29 | 79.79 | — | |
| Lookback Lens + LaaBLLM=Mistral-7B-Instruct-v0.32026.05 | — | — | — | 71.24 | 72.1 | — | |
| LookBackLensBackbone=Llama-3.1 8B, Setup=out-of-domain2025.05 | — | — | — | — | — | 31.6 | |
| MINDBackbone=Llama-3.1 8B, Setup=out-of-domain2025.05 | — | — | — | — | — | 35.2 | |
| RAUQBackbone=Llama-3.1 8B, Setup=out-of-domain2025.05 | — | — | — | — | — | 54.9 | |
| SAPLMALLM=Llama-3.1-8B-Instruct2026.05 | — | — | — | 69.96 | 70.75 | — | |
| SAPLMALLM=Llama-3.1-70B-Instruct2026.05 | — | — | — | 71.19 | 79.62 | — | |
| SAPLMALLM=Qwen-2.5-32B-Instruct2026.05 | — | — | — | 69.38 | 78.38 | — | |
| SAPLMALLM=Mistral-7B-Instruct-v0.32026.05 | — | — | — | 70.89 | 71.36 | — | |
| SAPLMABackbone=Llama-3.1 8B, Setup=out-of-domain2025.05 | — | — | — | — | — | 0 | |
| SAPLMA + LaaBLLM=Llama-3.1-8B-Instruct2026.05 | — | — | — | 71.77 | 73.25 | — | |
| SAPLMA + LaaBLLM=Llama-3.1-70B-Instruct2026.05 | — | — | — | 71.4 | 81.75 | — | |
| SAPLMA + LaaBLLM=Qwen-2.5-32B-Instruct2026.05 | — | — | — | 69.6 | 78.99 | — | |
| SAPLMA + LaaBLLM=Mistral-7B-Instruct-v0.32026.05 | — | — | — | 70.99 | 71.26 | — | |
| SATRMD+MSPBackbone=Llama-3.1 8B, Setup=out-of-domain2025.05 | — | — | — | — | — | 57.2 | |
| Self-JudgeLLM=Llama-3.1-8B-Instruct2026.05 | — | — | — | 57.29 | 65.13 | — | |
| Self-JudgeLLM=Llama-3.1-70B-Instruct2026.05 | — | — | — | 64.3 | 82.18 | — | |
| Self-JudgeLLM=Qwen-2.5-32B-Instruct2026.05 | — | — | — | 65.14 | 77.54 | — | |
| Self-JudgeLLM=Mistral-7B-Instruct-v0.32026.05 | — | — | — | 42.28 | 61.8 | — |