Hallucination Detection on MMLU (stratified subset, test)
0.8AUROCACSE
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| ACSELLM Backbone=Mistral-7B-Instruct2026.05 | 0.8 | 46 | 39 | 73 | 72 | |
| CAPLLM Backbone=Mistral-7B-Instruct2026.05 | 0.73 | 62 | 54 | 70 | 69 | |
| DDCRP-CPLLM Backbone=Mistral-7B-Instruct2026.05 | 0.71 | 66 | 58 | 68 | 67 | |
| SULLM Backbone=Mistral-7B-Instruct2026.05 | 0.69 | 69 | 61 | 65 | 65 | |
| EigVLLM Backbone=Mistral-7B-Instruct2026.05 | 0.68 | 75 | 68 | 64 | 62 | |
| P(True)LLM Backbone=Mistral-7B-Instruct2026.05 | 0.58 | 85 | 78 | 54 | 55 | |
| TELLM Backbone=Mistral-7B-Instruct2026.05 | 0.52 | 90 | 84 | 50 | 52 |