OpAI-Bench is an operation-guided benchmark for studying progressive human-to-AI text transformation in AI-text detection. Starting from human-written source documents, it provides revision trajectories with increasing AI involvement across predefined edit operations and coverage levels.
The dataset supports multi-granularity evaluation at document, sentence, token, and span levels. The main benchmark covers four domains and four generators. The ablation subsets… See the full description on the dataset page:
https://huggingface.co/datasets/OpAI-Bench1/OpAI-Bench.
