Hallucination Detection on Mixture-domain TriviaQA, CSQA, Belebele, CoQA, Math, SVAMP
0.787AUROCSpikeScore
Evaluation Results
| Method | Links | |
|---|---|---|
| SpikeScoreBackbone=Qwen3-14B, Method Category=Cross-domain specialized methods2026.01 | 0.787 | |
| SpikeScoreBackbone=Llama3.1-8B, Method Category=Cross-domain specialized methods2026.01 | 0.775 | |
| SpikeScoreBackbone=Qwen3-8B, Method Category=Cross-domain specialized methods2026.01 | 0.739 | |
| SpikeScoreBackbone=Llama3.2-3B, Method Category=Cross-domain specialized methods2026.01 | 0.71 | |
| ICR ProbeBackbone=Qwen3-14B, Method Category=Cross-domain specialized methods2026.01 | 0.705 | |
| ICR ProbeBackbone=Llama3.1-8B, Method Category=Cross-domain specialized methods2026.01 | 0.702 | |
| ICR ProbeBackbone=Llama3.2-3B, Method Category=Cross-domain specialized methods2026.01 | 0.695 | |
| ICR ProbeBackbone=Qwen3-8B, Method Category=Cross-domain specialized methods2026.01 | 0.686 | |
| PRISMBackbone=Qwen3-14B, Method Category=Cross-domain specialized methods2026.01 | 0.672 | |
| PRISMBackbone=Llama3.1-8B, Method Category=Cross-domain specialized methods2026.01 | 0.67 | |
| PRISMBackbone=Llama3.2-3B, Method Category=Cross-domain specialized methods2026.01 | 0.647 | |
| PRISMBackbone=Qwen3-8B, Method Category=Cross-domain specialized methods2026.01 | 0.633 | |
| SAPLMABackbone=Qwen3-8B, Method Category=Training-based methods2026.01 | 0.569 | |
| SAPLMABackbone=Qwen3-14B, Method Category=Training-based methods2026.01 | 0.563 | |
| MMBackbone=Llama3.1-8B, Method Category=Training-based methods2026.01 | 0.545 | |
| SAPLMABackbone=Llama3.1-8B, Method Category=Training-based methods2026.01 | 0.541 | |
| MMBackbone=Llama3.2-3B, Method Category=Training-based methods2026.01 | 0.538 | |
| MMBackbone=Qwen3-14B, Method Category=Training-based methods2026.01 | 0.527 | |
| SAPLMABackbone=Llama3.2-3B, Method Category=Training-based methods2026.01 | 0.516 | |
| SEPBackbone=Qwen3-8B, Method Category=Training-based methods2026.01 | 0.512 | |
| SEPBackbone=Qwen3-14B, Method Category=Training-based methods2026.01 | 0.511 | |
| MMBackbone=Qwen3-8B, Method Category=Training-based methods2026.01 | 0.509 | |
| SEPBackbone=Llama3.1-8B, Method Category=Training-based methods2026.01 | 0.507 | |
| SEPBackbone=Llama3.2-3B, Method Category=Training-based methods2026.01 | 0.503 |