Community Notes Generation on Community Notes Human Evaluation N=100 (random sample)
99F1: Factual AccuracyGITSEARCH
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| GITSEARCHbackbone=Gemini-2.5-Flash2026.02 | 99 | 89 | 91 | 90 | 97 | 87 | 3.87 | |
| Supernote Litebackbone=Gemini-2.5-Flash2026.02 | 96 | 83 | 73 | 77 | 91 | 53 | 3.14 | |
| Web Agentbackbone=Gemini-2.5-Flash2026.02 | 96 | 80 | 89 | 88 | 85 | 65 | 3.44 | |
| Human2026.02 | 94 | 91 | 87 | 81 | 93 | 69 | 3.36 |