Research map / Large language models / Language, safety and interpretability
Multi-level Style Preference Optimization: An Adaptive Detection Framework for Human-Machine Hybrid Text
Zehao Wang, Lianwei Wu, Wenbo An, Hang Zhang, Yaxiong Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add phrase-level and lexical-level losses to sequence-level loss (ablation) | MSPO · GPT-3.5 polished CS | AUROC: 0.9175 -> 0.9774 | Improved |
| add lexical-level KL constraint (ablation) | MSPO · GPT-3.5 polished CS | AUROC: 0.9227 -> 0.9774 | Improved |
| add phrase-level n-gram hinge loss (ablation) | MSPO · GPT-3.5 polished CS | AUROC: 0.9350 -> 0.9774 | Improved |
| add complexity-based dynamic beta (ablation) | MSPO · GPT-3.5 polished CS | AUROC: 0.9429 -> 0.9774 | Improved |
| use MSPO instead of ImBD on rewritten text | MSPO · GPT-3.5 rewritten, PHX | AUROC: 0.7099 -> 0.8214 | Improved |
| use MSPO instead of ImBD on polished text | MSPO · XSum polished (4 LLM avg) | AUROC: 0.8208 -> 0.8809 | Improved |
| use MSPO instead of ImBD on LLaMA3.1 rewritten news | MSPO · XSum rewritten, LLaMA3.1 | AUROC: 0.5932 -> 0.7715 | Improved |
| use MSPO instead of ImBD on purely generated text | MSPO · GPT-3.5 generated, CS | AUROC: 0.9997 -> 0.9907 | No clear effect |
| use MSPO instead of ImBD on creative-writing rewrites | MSPO · WritingPrompts rewritten, DeepSeek/GPT-4o | AUROC: GPT-4o 0.6216 -> 0.7094; GPT-3.5 0.6206 -> 0.7625; LLaMA3.1 0.7082 -> 0.7329 | Mixed |
Closest papers by meaning
- Imitate Before Detect: Aligning Machine Stylistic Preference for Machine-Revised Text Detection (AAAI 2025 · similarity 0.93)
- Few-Shot Detection of Machine-Generated Text using Style Representations (ICLR 2024 · similarity 0.89)
- Detecting Machine-Generated Texts by Multi-Population Aware Optimization for Maximum Mean Discrepancy (ICLR 2024 · similarity 0.89)
- ReMoDetect: Reward Models Recognize Aligned LLM's Generations (NeurIPS 2024 · similarity 0.88)
- Position: On the Possibilities of AI-Generated Text Detection (ICML 2024 · similarity 0.88)
- DeTeCtive: Detecting AI-generated Text via Multi-Level Contrastive Learning (NeurIPS 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
