self-monitor

# Self-Monitor Dataset

This dataset contains supervised fine-tuning (SFT) data used in the research paper "Mitigating Deceptive Alignment via Self-Monitoring" ([arXiv:2505.18807](https://arxiv.org/abs/2505.18807)).

## Overview

The self-monitor dataset is designed to train language models to develop self-monitoring capabilities that can help mitigate deceptive alignment behaviors. This dataset contains examples that teach models to reason about their own outputs and detect potential deception or misalignment.

## Paper Information

**Title:** Mitigating Deceptive Alignment via Self-Monitoring
**arXiv:** [2505.18807](https://arxiv.org/abs/2505.18807)
**Project Website:** [https://cot-monitor-plus.github.io/](https://cot-monitor-plus.github.io/)

## Dataset Description

This dataset is specifically curated for supervised fine-tuning of language models to enhance their self-monitoring capabilities. The data focuses on scenarios where models need to:

- Monitor their own reasoning processes
- Detect potentially deceptive or misaligned outputs
- Develop better alignment with human values and intentions
- Improve transparency in their decision-making processes

```bibtex
@article{ji2025mitigating,
title={Mitigating deceptive alignment via self-monitoring},
author={Ji, Jiaming and Chen, Wenqi and Wang, Kaile and Hong, Donghai and Fang, Sitong and Chen, Boyuan and Zhou, Jiayi and Dai, Juntao and Han, Sirui and Guo, Yike and others},
journal={arXiv preprint arXiv:2505.18807},
year={2025}
}
```

## License

This dataset is released under the Apache 2.0 License.

# 自我监控数据集（Self-Monitor Dataset）

本数据集包含用于论文《通过自我监控缓解欺骗性对齐》（*Mitigating Deceptive Alignment via Self-Monitoring*，arXiv:2505.18807）研究的监督微调（Supervised Fine-Tuning，SFT）数据。

## 概述
本自我监控数据集旨在训练语言模型习得自我监控能力，以助力缓解欺骗性对齐行为。数据集包含若干示例，用于指导模型对自身输出进行推理，并检测潜在的欺骗或对齐偏差问题。

## 论文信息
**论文标题：** 《通过自我监控缓解欺骗性对齐》（*Mitigating Deceptive Alignment via Self-Monitoring*）
**arXiv编号：** [2505.18807](https://arxiv.org/abs/2505.18807)
**项目官网：** [https://cot-monitor-plus.github.io/](https://cot-monitor-plus.github.io/)

## 数据集说明
本数据集专为语言模型的监督微调任务构建，用于增强其自我监控能力。数据聚焦于模型需要完成以下任务的场景：
- 监控自身推理过程
- 检测潜在的欺骗性或对齐偏差输出
- 优化与人类价值观及意图的对齐效果
- 提升决策过程的透明度

bibtex
@article{ji2025mitigating,
title={Mitigating deceptive alignment via self-monitoring},
author={Ji, Jiaming and Chen, Wenqi and Wang, Kaile and Hong, Donghai and Fang, Sitong and Chen, Boyuan and Zhou, Jiayi and Dai, Juntao and Han, Sirui and Guo, Yike and others},
journal={arXiv preprint arXiv:2505.18807},
year={2025}
}

## 许可证
本数据集采用Apache 2.0许可证发布。