Skip to content
arXiv · Artificial Intelligence· Yerim Oh, Young-Jun Lee, Jaewoo Ahn, Gunhee Kim, Dongyeop Kang·· 3 hr agoSelectedAI score67

Science or Slop?: Benchmarking and Mitigating Scientific Slop in AI-Generated Papers

Science or Slop?: Benchmarking and Mitigating Scientific Slop in AI-Generated Papers

AI brief

该研究提出 SciSlopBench,包含 390 篇 AI 生成论文与对应人类撰写论文,衡量结构、论证与 artifacts 三类科学 slop,以 85.9% 准确率识别 AI 论文;SciSlopHarness 在不依赖人类参考目标时,将 AI-人类差距减少 63%。

Why it matters

论文构建了 SciSlopBench 并提出 SciSlopHarness,为识别与减轻 AI 论文中的科学性泡沫提供了可复现的基准与框架。

Source: arXiv · Artificial Intelligence · arxiv.org