Faithfulness Detection on LatentAudit Llama-3-8B (evaluation set)
0.948AUROCGPT-4o Judge
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPT-4o JudgeLatency (ms)=∼5,3002026.04 | 0.948 | 0.881 | |
| LatentAuditLatency (ms)=0.77 (+11.2 proving)2026.04 | 0.942 | 0.869 | |
| INSIDELatency (ms)=∼3.82026.04 | 0.908 | 0.841 | |
| SAPLMALatency (ms)=∼1.52026.04 | 0.882 | 0.815 | |
| SelfCheckGPTLatency (ms)=∼28,5002026.04 | 0.871 | 0.804 | |
| Min-PerplexityLatency (ms)=0.02026.04 | 0.722 | 0.655 |