File size: 2,133 Bytes
87e82b6 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 | ---
license: apache-2.0
task_categories:
- text-classification
language:
- en
tags:
- aigc-detection
- text-classification
- qwen
base_model: Qwen/Qwen3-8B
---
# REVEAL_think_3class
**REVEAL-think-3class** is a reasoning-driven AI-Generated Content (AIGC) detection model based on Qwen3-8B. It uses a **Think-then-Answer** paradigm, generating a transparent reasoning chain (`<think>...</think>`) before outputting the final fine-grained classification (`<answer>...</answer>`).
This model is introduced in the paper: **[Reasoning-Aware AIGC Detection via Alignment and Reinforcement](https://arxiv.org/abs/2604.19172)**.
🔗 **Project Homepage & Code:** [https://aka.ms/reveal](https://aka.ms/reveal)
📚 **Associated Dataset:** [AIGC-text-bank](https://huggingface.co/datasets/bmbgsj/AIGC-text-bank)
## 🌟 Model Overview
This model performs fine-grained detection, discriminating between three categories:
- **Human**: Authentic human-authored text.
- **AI-Native**: Fully machine-generated text.
- **AI-Polish**: Human-authored text refined by AI to improve fluency and style while preserving original semantics.
## 🚀 How to Use
To run inference, simply use the [`think.py`](https://github.com/microsoft/AnthropomorphicIntelligence/blob/main/REVEAL/inference/think.py) script provided in our GitHub repository. It handles prompt formatting, `vLLM` acceleration, and automatically extracts the final prediction along with fine-grained confidence scores.
```bash
python think.py \
--model_path "bmbgsj/REVEAL_think_3class" \
--text "The rapid advancement of Large Language Models has ushered in an era where AI-generated content is increasingly pervasive..."
```
## 📖 Citation
If you use this model in your research, please cite:
```bibtex
@misc{wang2026reasoningawareaigcdetectionalignment,
title={Reasoning-Aware AIGC Detection via Alignment and Reinforcement},
author={Zhao Wang and Max Xiong and Jianxun Lian and Zhicheng Dou},
year={2026},
eprint={2604.19172},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2604.19172},
}
``` |