📄 中文摘要
构建 OmniDetect 的过程中，作者发现单一的 AI 内容检测器在边缘案例中的准确性往往不可靠。通过对 211 个样本的基准测试，结果显示不同检测工具之间的结果差异显著，GPTZero、Originality.ai 和 Winston AI 在同一文本上给出了不同的判断。这种不一致性使得依赖单一工具进行内容检测变得风险较高，强调了多引擎检测的必要性。作者的研究揭示了 AI 内容检测领域中不为人知的挑战和局限性。
我构建了一个多引擎 AI 内容检测器——我对检测准确性所学到的
📄 English Summary
I Built a Multi-Engine AI Content Detector — Here's What I Learned About Detection Accuracy
While building OmniDetect, the author discovered that single AI content detectors often provide unreliable accuracy in edge cases. A benchmark of 211 samples revealed significant discrepancies among different detection tools, with GPTZero, Originality.ai, and Winston AI offering varying judgments on the same text. This inconsistency highlights the risks of relying on a single tool for content detection and underscores the necessity for multi-engine detection. The research uncovers challenges and limitations in the AI content detection field that are often overlooked.