Post-hoc Correction on ConflictBank (100 atomic claims)
89ROUGEFACTCORRECTOR
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| FACTCORRECTORBase LLM=mixtral-8x22b-instruct2026.01 | 89 | 87 | 0.73 | 87 | |
| RACBase LLM=llama-3.3-70b-instruct2026.01 | 87 | 77 | 0.65 | 77 | |
| FACTCORRECTORBase LLM=granite-4.0-h-small2026.01 | 83 | 76 | 0.58 | 78 | |
| RACBase LLM=granite-4.0-h-small2026.01 | 74 | 48 | 0.49 | 73 | |
| FACTCORRECTORBase LLM=llama-3.3-70b-instruct2026.01 | 73 | 62 | 0.46 | 60 | |
| RACBase LLM=mixtral-8x22b-instruct2026.01 | 54 | 23 | 0.29 | 87 | |
| LLM2Base LLM=granite-4.0-h-small2026.01 | 42 | 15 | -0.1 | 13 | |
| CRITICBase LLM=llama-3.3-70b-instruct2026.01 | 32 | 15 | -0.36 | 31 | |
| LLM2Base LLM=llama-3.3-70b-instruct2026.01 | 21 | 4 | -0.46 | 6 | |
| LLM2Base LLM=mixtral-8x22b-instruct2026.01 | 20 | 5 | -0.48 | 10 | |
| LLM1Base LLM=llama-3.3-70b-instruct2026.01 | 19 | 3 | -0.62 | 17 | |
| LLM1Base LLM=mixtral-8x22b-instruct2026.01 | 15 | 2 | -0.63 | 30 | |
| CRITICBase LLM=granite-4.0-h-small2026.01 | 15 | 3 | -0.69 | 20 | |
| LLM1Base LLM=granite-4.0-h-small2026.01 | 15 | 2 | -0.68 | 17 | |
| CRITICBase LLM=mixtral-8x22b-instruct2026.01 | 14 | 2 | -0.69 | 41 |