检测LLM产生的同行评审
Vishisht Srihari Rao1, Aounon Kumar2, Himabindu Lakkaraju2
1Machine Learning Department, Carnegie Mellon University, Pittsburgh, Pennsylvania, United States of America.
PloS one
|September 22, 2025
概括
检测人工智能生成的科学评论是具有挑战性的. 本研究引入了嵌入在PDF文件中的新水印技术,可靠地识别大型语言模型 (LLM) 辅助的评论,即使对防御.
科学领域:
- 人工智能的人工智能
- 科学出版科学出版
- 信息安全 信息安全
背景情况:
- 科学同行评审的完整性对于进步至关重要.
- 大型语言模型 (LLM) 构成风险,可能会产生人工智能生成的审查.
- 目前的检测工具很难区分人工智能辅助和人类审查.
研究的目的:
- 开发一种强大的方法来检测LLM产生的同行评价.
- 解决现有检测工具在识别AI辅助内容方面的局限性.
- 确保科学同行评审过程的可靠性和真实性.
主要方法:
- 通过PDF实现间接快速注入,在LLM生成的评论中嵌入隐藏的水标.
- 开发具有强有力的统计保证的新型水标方案和假设测试.
- 控制多次审查的家庭智能错误率,以提高统计能力.
- 评估多种间接提示注入策略,包括基于字体的嵌入和模糊提示.
- 测试对常见审查员防御的弹性,并在实践中验证统计界限.
主要成果:
- 在各种LLM中嵌入水标的成功率很高.
- 拟议的水印方法证明了对常见审查员防御的弹性.
- 统计测试在实践中保持了准确性,超过了像邦费罗尼校正这样的保守方法.
- 该框架为检测人工智能生成的审查提供了强有力的统计保证.
结论:
- 一个严格的水标和检测框架有效地识别了LLM产生的同行评审.
- 该方法提供了可靠的检测和强大的统计保证,对于科学完整性至关重要.
- 这种方法提升了区分人工智能辅助内容的能力,保护了同行评审过程.
相关概念视频
Review and Preview
8.3K
In statistics, several tools are used to interpret the data. Measures of central tendency represent the characteristics of the data, such as mean, median, and mode. Additionally, measures of variance like standard deviation and range are used to find the spread of data from the mean. Relative standing measures the distance between data locations. Commonly used measures of relative standings are percentile, z score, and quartiles.
Percentiles are a type of fractile that partition data into...
Percentiles are a type of fractile that partition data into...
8.3K
Review and Preview
10.8K
Data are individual items of information obtained from a population or sample. Data may be classified as qualitative (categorical), quantitative continuous, or quantitative discrete. Because it is not practical to measure the entire population in a study, researchers use samples to represent the population. A random sample is a representative group from the population chosen by using a method that gives each individual in the population an equal chance of being included in the sample. Random...
10.8K
Leaky Scanning
5.6K
During most eukaryotic translation processes, the small 40S ribosome subunit scans an mRNA from its 5' end until it encounters the first start AUG codon. The large 60S ribosomal subunit then joins the smaller one to initiate protein synthesis. The location of the translation initiation is largely determined by the nucleotides near the start codon as there may be multiple translation initiation sites present on the mRNA. Marilyn Kozak discovered that the sequence RCCAUGG (where R...
5.6K
Quantifying and Rejecting Outliers: The Grubbs Test
3.4K
Sometimes, a data set can have a recorded numerical observation that greatly deviates from the rest of the data. Assuming that the data is normally distributed, a statistical method called the Grubbs test can be used to determine whether the observation is truly an outlier. To perform a two-tailed Grubbs test, first, calculate the absolute difference between the outlier and the mean. Then, calculate the ratio between this difference and the standard deviation of the sample. This...
3.4K
Proofreading
59.6K
Overview
59.6K


