Research map / Large language models / Language, safety and interpretability
On the Salience of Low-Probability Tokens for AI-Generated Text Detection: A Multiscale Uncertainty Perspective
Yikai Guo, Bin Wang, Xilai Fan, Wenjun Ke, Haoran Luo
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 |
|---|---|---|---|
| restrict statistics to low-probability token positions (ablation) | Uncertainty++ · Reddit black-box, 12 source models | AUROC: 88.86 -> 95.60 | Improved |
| include the local tail log-probability component (ablation) | Uncertainty++ · Reddit black-box | AUROC: 63.80 -> 95.60 | Improved |
| add Renyi entropy global uncertainty (ablation) | Uncertainty++ · Reddit black-box | AUROC: 94.71 -> 95.60 | Improved |
| compute Likelihood, LogRank, Fast-DetectGPT only over low-probability tokens | Likelihood / LogRank / Fast-DetectGPT · XSum/WritingPrompts/Reddit, Neo-2.7 source | AUROC: +24.53, +18.97, +4.20 average | Improved |
| vary Renyi order alpha and weight beta | Uncertainty++ · Reddit black-box | AUROC: within 1% | No clear effect |
| vary low-probability percentile rho | Uncertainty++ · Reddit black-box | AUROC | Mixed |
| increase number of conditional samples m | Uncertainty++ · Reddit black-box | AUROC | No clear effect |
| use Uncertainty instead of Lastde | detector · Black-box 12 source models | AUROC: 86.49 -> 88.74 | Improved |
| use Uncertainty instead of Lastde on Llama3-8 source | detector · Black-box Llama3-8 | AUROC: 99.60 -> 98.43 | Hurt |
| use Uncertainty++ instead of Lastde++ | detector · White-box 12 source models | AUROC: 98.93 -> 99.07 | Improved |
| use Uncertainty++ against strong GPT-5 paraphrasing | detector · arXiv paraphrased by GPT-5 | AUROC: 94.08 -> 77.57 | Hurt |
| use Uncertainty methods under frequency penalty decoding | Uncertainty / Uncertainty++ · arXiv GPT-5 decoding variants | AUROC: +7.47 and +11.58 over Lastde and Lastde++ | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Fast-DetectGPT — Fast-DetectGPT: Efficient Zero-Shot Detection of Machine-Generated Text via Conditional Probability Curvature (ICLR 2024) · better on 2 of 2 shared results
- DetectGPT — DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature (ICML 2023) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Log-Likelihood, Simpson’s Paradox, and the Detection of Machine-Generated Text (NeurIPS 2026 · similarity 0.88)
- Position: On the Possibilities of AI-Generated Text Detection (ICML 2024 · similarity 0.88)
- Distinguishing the Knowable from the Unknowable with Language Models (ICML 2024 · similarity 0.87)
- Forking Paths in Neural Text Generation (ICLR 2025 · similarity 0.87)
- Human Texts Are Outliers: Detecting LLM-generated Texts via Out-of-distribution Detection (NeurIPS 2025 · similarity 0.87)
- Towards Understanding and Quantifying Uncertainty for Text-to-Image Generation (CVPR 2025 · similarity 0.87)
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.
