---
pretty_name: aihumanmerged
license: mit
---

# aihumanmerged

Combined neologism detector pair dataset.

Contents:
- `train_pairs.json`
- `holdout_pairs.json`
- `metadata.json`

Construction:
- RAID unique human texts paired with random AI texts, with 500 RAID pairs held out.
- EditLens `ai_generated` vs `human_written`, with test used as holdout.
- `gsingh1-py/train`: one unique human text paired with a random AI text.
- `AlekseyKorshuk/ai-detection-gutenberg-human-formatted-ai-part2`: one unique human text paired with a random AI text.
- `coai/ai-text-detection-training`: one unique human text paired with a random AI text.