TABLE A3.
Performance of Different Thresholds for Each AI Content Detector in Classifying Human-Written, AI-Generated, Mixed, and Translated Abstracts
| Detector | Human-Written (n = 100) | Mixed, GPT-3.5 (n = 100) | Mixed, GPT-4 (n = 100) | Generated, GPT-3.5 (n = 100) | Generated, GPT-4 (n = 100) | Translated (n = 100) |
|---|---|---|---|---|---|---|
| Abstracts per category classified as high likelihood of AI content with specified threshold | ||||||
| GPTZero | 0 | 4 | 1 | 100 | 99 | 0 |
| Originality.ai | 0 | 9 | 3 | 100 | 92 | 33 |
| Sapling | 1 | 13 | 17 | 99 | 95 | 3 |
| Abstracts per category classified as high likelihood of mixed AI/human content with specified threshold | ||||||
| GPTZero | 1 | 15 | 16 | 100 | 99 | 5 |
| Originality.ai | 11 | 51 | 44 | 100 | 98 | 82 |
| Sapling | 3 | 30 | 25 | 99 | 95 | 7 |
Abbreviation: AI, artificial intelligence.