Research map / Large language models / AI and society
Sem-Detect: Semantic Level Detection of AI Generated Peer-Reviews
Andre Vicente Duarte, Brian Tufts, Aditya Oke, Fei Fang, Arlindo L. Oliveira, Lei Li
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 |
|---|---|---|---|
| combine claim-level semantic features with textual features vs best baseline EditLens | Sem-Detect · Human vs AI (ML conferences test set) | TPR@0.1% FPR: 0.606 -> 0.760 | Improved |
| use claim-level semantic comparison vs full-review embedding (Anchor) | Sem-Detect · Human vs AI (ML conferences test set) | TPR@0.1% FPR: 0.541 -> 0.760 | Improved |
| evaluate on unseen generator models (OOD-M) | Sem-Detect · 3-class distribution shift | macro-F1: 0.84 -> 0.71 | Hurt |
| also change prompts (OOD-M+P) | Sem-Detect · 3-class distribution shift | macro-F1: 0.84 -> 0.68 | Hurt |
| unseen models: AI recall | Sem-Detect · 3-class distribution shift | AI recall: 0.91 -> 0.67 | Hurt |
| unseen models: AI precision | Sem-Detect · 3-class distribution shift | AI precision: 0.93 -> 0.97 | Improved |
| apply to medical imaging venue MIDL without retraining | Sem-Detect · MIDL 2022 cross-domain | F1 | No clear effect |
| confidence threshold 0.80 for manual review | Sem-Detect · 3-class | accuracy: accuracy 94.7% on 79% coverage | Improved |
| collapse LLM-refined into non-AI class for general detectors | general-purpose detectors · (Human + LLM-Refined) vs AI | AUC: AUC <= 0.513 | Hurt |
| LLM refinement for TF Model | TF Model · (Human + LLM-Refined) vs AI | AUC: AUC 0.926 -> 0.674 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- RADAR — RADAR: Robust AI-Text Detection via Adversarial Learning (NeurIPS 2023) · better on 4 of 4 shared results
- EditLens — EditLens: Quantifying the Extent of AI Editing in Text (ICLR 2026) · better on 3 of 3 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
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
- Monitoring AI-Modified Content at Scale: A Case Study on the Impact of ChatGPT on AI Conference Peer Reviews (ICML 2024 · similarity 0.88)
- From Replication to Redesign: Exploring Pairwise Comparisons for LLM-Based Peer Review (NeurIPS 2025 · similarity 0.85)
- PiCO: Peer Review in LLMs based on Consistency Optimization (ICLR 2025 · similarity 0.85)
- Position: The AI Conference Peer Review Crisis Demands Author Feedback and Reviewer Rewards (ICML 2025 · similarity 0.84)
- CycleResearcher: Improving Automated Research via Automated Review (ICLR 2025 · similarity 0.84)
- DETree: DEtecting Human-AI Collaborative Texts via Tree-Structured Hierarchical Representation Learning (NeurIPS 2025 · similarity 0.84)
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.
