Turnitin查不出来了？娜姐这一句提示词，把AI率降得服服帖帖

随着ChatGPT等AI工具的兴起，论文除了重复率检测，又多了一项AI率检测。但是，这个AI率检测，存在好多bug，比如：

高误报率: 很多用户吐槽，他们完全由人工撰写的、甚至是在 AI 出现之前的文章，被检测为 AI 生成。美国的好几所高校也因此，相继取消了Turnitin的AI率检测。改为由教授自行酌情判断。

标准不一: 据调查，同一篇文章在不同检测工具上的得分差异巨大，如下图所示，AI率用不同工具检测，结果在33-79%之间变动。

对非母语者的不公: 也有用户反馈，非英语母语者被教导的规范统一的学术写作，恰好符合 AI 的写作模式，对非母语科研人员非常不公平。

期刊编辑也不会上来不看文章，就直接检测AI率。除非你的论文确实是机械、空洞的AI生成内容，一眼假。才会检测确认一下。

但是，AI率又是大家投稿普遍担心的问题。好多学员跟我提了这个需求。

娜姐测试过市面上的一些提示词和AI改写工具，最主要的问题是改写后面目全非：AI率是降下来了，但是改后的文本不符合学术风格，和上下文也衔接不起来。

因此，娜姐自己琢磨了既能降低AI率，同时又保留学术风格的“降AI率提示词”。这里的难点是要平衡学术风和“人味”。

来看看娜姐提示词的效果：

这篇2023年发表在European Chemical Bulletin上的文章Throat Cancer detection using canny edge detector with ANN，

文中居然出现了“As an AI language model”，这篇文章估计全文都是AI撰写的：

Turnitin检测一下：

用娜姐的提示词改写一下前言部分：

改写后的前言部分内容：

Throat cancer—including laryngeal and pharyngeal malignancies—remains a critical health concern due to its potential lethality and the challenges associated with its early detection. Prompt diagnosis is essential, as timely intervention significantly improves the likelihood of favorable treatment outcomes. Conventional diagnostic modalities, such as computed tomography (CT) and magnetic resonance imaging (MRI), are widely employed to visualize anatomical abnormalities. However, these techniques typically require expert interpretation, are resource-intensive, and may not be readily accessible in all clinical settings. 

In response to these limitations, computer-aided diagnosis (CAD) systems have garnered growing interest as tools to enhance diagnostic accuracy and efficiency. By integrating image processing algorithms with machine learning frameworks, CAD systems offer the potential to automate aspects of image interpretation and assist clinicians in decisionmaking.

Among these, edge detection algorithms are particularly relevant; by delineating structural boundaries, they can accentuate image features potentially associated with malignancy. 

This study introduces a methodological framework that combines the Canny edge detection algorithm with artificial neural networks (ANNs) for throat cancer classification. The Canny algorithm, a well-established technique grounded in gradient intensity analysis, is applied to preprocess throat cancer images, thereby enhancing the visibility of contours and margins that may demarcate pathological regions.

The edge-enhanced images are subsequently input into an ANN, a machine learning architecture modeled loosely on the interconnected structure of the human brain. The network is trained on a curated dataset of annotated images—each labeled as cancerous or non-cancerous—to enable the model to learn patterns that differentiate between malignant and benign features based on edge characteristics.

The primary aim of this work is to assess the diagnostic performance of the proposed approach. Specifically, we evaluate its classification accuracy, sensitivity, and specificity relative to existing techniques. Should the findings prove favorable, this method may contribute to the development of an efficient, automated system capable of supporting earlier and more accessible detection of throat cancer. 

The remainder of this paper is structured as follows. Section 2 surveys relevant literature on throat cancer detection and image processing methods. Section 3 details the proposed methodology, including the implementation of the Canny edge detector and ANN architecture. Section 4 outlines the experimental design, dataset characteristics, and evaluation metrics. Section 5 presents and interprets the results. Finally, Section 6 concludes with a summary of contributions, limitations, and directions for future research.

改写后的文本很好的保留了原文的结构和学术风格。

你说AI率怎么没降到0？我尝试过，如果再降，文本内容和风格就会改动太大。那就用不成了。

我这篇检测的AI率是100%，如果你的AI率没那么高，降一次就差不多了。后来又测试了Discussion部分，差不多也是100%降到50%，效果比较稳定。

娜姐调试了好多个版本，这个是最好的：既能保留和提升学术风格，又能有效降低AI率。

最后说明一下， 娜姐认为， 应该把针对AI率的检测回归到关注内容质量本身。学术论文发表的目的是为了和同行交流，文字只是载体。如果AI能帮你把文字表达梳理的更清晰、流畅、规范，有何不可？

我这个提示词只针对Turnitin的AI率检测。因为前面也说了，各大工具检测出来的AI率差异很大。期刊和出版社一般都用Turnitin，保持一致即可。