A study of 12,750 arXiv papers finds AI-like writing flagged in 65% of CS papers
新智元 · wechat · 2026-07-27
A new unslop study argues that AI-written or AI-edited text has become widespread in arXiv papers, but the headline numbers are best read as a lower bound rather than a verdict.
- The team scanned 12,750 arXiv papers across ten fields from Jan 2023 to Jul 2026.
- Using its detector, it found the share of papers flagged as AI-like rose from 0.4% in the pre-ChatGPT baseline to 32% in the latest full quarter, peaking near 39% in early 2026.
- By field, computer science was flagged at 65%, while mathematics was only 0.7%.
- The authors stress the detector mostly measures the “smell” of AI writing; it can’t cleanly separate light grammar editing from full machine-generated papers, and it performs poorly on formula-heavy math papers.
- The article connects this to broader evidence from Stanford that LLM-assisted writing is already common in peer review and paper abstracts, especially in competitive fields.
More from Research
- SceneActBench tests whether VLM agents can act in full 3D scenes — Yifei Zhao · 2026-07-27
- Tencent Hunyuan maps scaling laws for native multimodal pretraining from scratch — Tencent-Hunyuan · 2026-07-27
- A training-free method fixes revisit inconsistency in generative rendering — Wenchao Ma · 2026-07-27
- NVIDIA open-sources Molt, a PyTorch-native framework for agentic RL — nvidia · 2026-07-27
- Skill Self-Play uses co-evolving skills to push LLM capability — QwenBusinessUnit-Edu · 2026-07-27
- VisCo reuses a pretrained VLM to compress visual tokens more effectively — Yupeng Zheng · 2026-07-27