Research map / Large language models / Language, safety and interpretability
SpecDetect: Simple, Fast, and Training-Free Detection of LLM-Generated Text via Spectral Analysis
Haitong Luo, Weiyao Zhang, Suhang Wang, Wenji Zou, Chungang Lin, Xuying Meng, Yujun Zhang
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 |
|---|---|---|---|
| use DFT total energy of log-prob sequence instead of Lastde time-series features | sample-based detector · Black-box avg AUC, 12 source models | AUC: 0.8776 -> 0.8875 | Improved |
| add sampling-discrepancy z-score normalization (SpecDetect++) | SpecDetect · Black-box avg AUC, 12 source models | AUC: 0.8875 -> 0.9259 | Improved |
| SpecDetect++ instead of Lastde++ | distribution-based detection · Black-box avg AUC | AUC: 0.9246 -> 0.9259 | No clear effect |
| use energy-based features rather than frequency-based spectral features (ablation) | spectral feature choice · XSum, ELI5, WritingPrompts distributions | class separation | Improved |
| use SpecDetect instead of Lastde on LLaMA3-8 source | sample-based detector · Black-box, LLaMA3-8 | AUC: 0.9851 -> 0.9761 | Hurt |
| use SpecDetect instead of Lastde on Phi-4 source | sample-based detector · Black-box, Phi-4 | AUC: 0.7353 -> 0.7773 | Improved |
| use SpecDetect instead of Lastde for runtime | sample-based detector · XSum GPT-4-Turbo, H800 | runtime: 16.4% faster | Improved |
| use SpecDetect++ instead of Lastde++ for runtime | distribution-based detection · XSum GPT-4-Turbo, H800 | runtime: nearly twice as fast | Improved |
| use SpecDetect++ with only 10 contrast samples vs Lastde++ with 50 (ablation) | distribution-based detection · BLOOM-7.1B source | AUC | 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 3 of 3 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 0.89)
- Training-free LLM-generated Text Detection by Mining Token Probability Sequences (ICLR 2025 · similarity 0.89)
- DALD: Improving Logits-based Detector without Logits from Black-box LLMs (NeurIPS 2024 · similarity 0.88)
- DNA-GPT: Divergent N-Gram Analysis for Training-Free Detection of GPT-Generated Text (ICLR 2024 · similarity 0.87)
- AdaDetectGPT: Adaptive Detection of LLM-Generated Text with Statistical Guarantees (NeurIPS 2025 · similarity 0.87)
- Turning Up the Heat: Min-p Sampling for Creative and Coherent LLM Outputs (ICLR 2025 · similarity 0.86)
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.
