Stepwise error detection on MoreHopQA (test)
0.808AUROCGIBS
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GIBSLLM backbone=Deepseek2026.05 | 0.808 | 0.835 | 70.5 | |
| NIBSLLM backbone=Deepseek2026.05 | 0.666 | 0.776 | 64.9 | |
| GIBSLLM backbone=Phi42026.05 | 0.661 | 0.686 | 70.5 | |
| GIBSLLM backbone=Llama3.1-8b2026.05 | 0.647 | 0.669 | 56 | |
| PARCLLM backbone=Deepseek2026.05 | 0.637 | 0.686 | 65.9 | |
| PARCLLM backbone=Phi42026.05 | 0.606 | 0.676 | 65 | |
| PARCLLM backbone=Llama3.1-8b2026.05 | 0.589 | 0.608 | 51.1 | |
| NIBSLLM backbone=Phi42026.05 | 0.58 | 0.663 | 69.6 | |
| ROSCOE-saLLM backbone=Llama3.1-8b2026.05 | 0.524 | 0.565 | 54.1 | |
| NIBSLLM backbone=Llama3.1-8b2026.05 | 0.512 | 0.553 | 53.1 | |
| ROSCOE-saLLM backbone=Deepseek2026.05 | 0.492 | 0.6 | 61.2 | |
| ROSCOE-saLLM backbone=Phi42026.05 | 0.488 | 0.594 | 61.3 | |
| ROSCOE-ssLLM backbone=Llama3.1-8b2026.05 | 0.477 | 0.517 | 50.3 | |
| ROSCOE-ssLLM backbone=Phi42026.05 | 0.421 | 0.603 | 54.2 | |
| ROSCOE-ssLLM backbone=Deepseek2026.05 | 0.408 | 0.584 | 52.8 |