Yashodhar29/synthetic-mirrors-human-ai-cnn-dailymail-v1

Synthetic Mirrors: Human vs AI (CNN DailyMail)

📌 Overview

Synthetic Mirrors is a multi-model, research-grade dataset designed for detecting humanized AI-generated text.

This dataset aggregates AI generations from multiple open-source and closed-source LLMs, paired with corresponding human-written content from the CNN DailyMail domain.

The core idea is to treat each AI model as a synthetic mirror — reflecting distinct stylistic and probabilistic artifacts that help detectors generalize to unseen models.

🤖 AI Models Used

• —GPT-3.5 Turbo
• —Qwen/Qwen2.5-1.5B-Instruct
• —HuggingFaceTB/SmolLM2-360M-Instruct
• —HuggingFaceTB/SmolLM2-1.7B-Instruct
• —Google/Gemma-2-2B-IT
• —Meta-LLaMA-3.2-1B-Instruct
• —Meta-LLaMA-3.1-8B-Instruct

🧱 Dataset Schema

Each row corresponds to one text sample.

🎯 Intended Use

• —AI-generated text detection
• —Humanized AI content analysis
• —Cross-model generalization studies
• —Leave-one-model-out evaluation
• —Training robust AI detectors

⚠️ Limitations

• —Single domain (news)
• —English only
• —AI generations may reflect prompt biases from original datasets

📖 Citation

If you use this dataset, please cite: @dataset{syntheticmirrors2025, title = {Synthetic Mirrors: Human vs AI (CNN DailyMail)}, author = {Chavan, Yashodhar}, year = {2025}, platform = {Hugging Face} }