Scientific Claims on SC
87.3R ScoreHalluClean
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HalluCleanProtocol=Detection → Revision (with rationale), Backbone=GPT-3.5-turbo2025.11 | 87.3 | 79.3 | |
| HalluCleanProtocol=Detection → Revision (with rationale), Backbone=DeepSeek-V32025.11 | 87.3 | 62.7 | |
| ChatProtectProtocol=Detection → Revision (with rationale), Backbone=GPT-3.5-turbo2025.11 | 79.3 | 79.3 | |
| GPT-4o-miniProtocol=Direct Ask, Backbone=GPT-4o-mini2025.11 | 72.7 | 72.7 | |
| Step-by-StepProtocol=Detection → Revision (with rationale), Backbone=GPT-3.5-turbo2025.11 | 53.3 | 53.3 | |
| Plan-and-SolveProtocol=Detection → Revision (with rationale), Backbone=GPT-3.5-turbo2025.11 | 53.3 | 53.3 | |
| Llama-3-70BProtocol=Direct Ask, Backbone=Llama-3-70B2025.11 | 49.3 | 49.3 | |
| DeepSeek-V3Protocol=Direct Ask, Backbone=DeepSeek-V32025.11 | 35.3 | 35.3 | |
| GPT-3.5-turboProtocol=Direct Ask, Backbone=GPT-3.5-turbo2025.11 | 30.7 | 30.7 | |
| DeepSeek-R1Protocol=Direct Ask, Backbone=DeepSeek-R12025.11 | 25.3 | 25.3 |