AI-generated text detection on Wikipedia OPT-13B generations (+ 60L,600)
97.2Accuracy (1% FPR)BM25
Evaluation Results
| Method | Links | |
|---|---|---|
| BM25Corpus Size=3K generations from model itself2023.03 | 97.2 | |
| BM25Corpus Size=3K generations from model itself2023.03 | 95.3 | |
| BM25Corpus Size=9K generations pooled from all three models2023.03 | 87.2 | |
| SPCorpus Size=3K generations from model itself2023.03 | 84.4 | |
| BM25Corpus Size=9K generations pooled from all three models2023.03 | 79.1 | |
| SPCorpus Size=3K generations from model itself2023.03 | 77.7 | |
| SPCorpus Size=9K generations pooled from all three models2023.03 | 74.6 | |
| SPCorpus Size=9K generations pooled from all three models2023.03 | 65.6 | |
| Watermark2023.03 | 63.7 | |
| Watermark2023.03 | 52.8 | |
| OpenAI2023.03 | 10 | |
| OpenAI2023.03 | 9.1 | |
| DetectGPT2023.03 | 0.8 | |
| DetectGPT2023.03 | 0.3 |