Researchers from Northeastern University have found out a new method which can make AI detection in text easier. They found this method through analyzing the sentence structure, syntax and some other points in a text. We already know that ChatGPT tends to repeat certain words in its texts. But it isn't the only way that we can find out if a text is AI generated or not. The researchers said that instead of focusing on what words are being used in the text, it is best to analyze the syntax to find if it is AI generated or not.
The researchers asked different AI models to write on different kinds of topics. All the texts produced by AI were then analyzed. The researchers analyzed what kind of pattern all these AI models are following which they named as syntactic templates. The researchers then found out that each AI model has its own specific syntactic template. One thing common about AI models was that they often used a set of dual adjectives to describe something.
Even though humans can also have repeated syntax in their writings, it is much lower than repeated syntax in AI writings. The writing style also mattered in identifying how difficult human writing is from AI writing. In texts like biomedical writing, the difference was small because biomedical writing uses a specific pattern. But in texts like movie reviews and news articles, the difference was bigger.
75% of the writing templates that AI models follow are from their training data and are not something they generate on their own during the writing process. So, if you want to detect AI generated text, it is best to look at repeated syntax and dual adjectives. Humans tend to get more creative in their writings, but AI models have to follow a certain method.
Image: DIW-Aigen
Read next:
• How To Use Google Docs Secret Feature to Detect AI Content Writing
• How Effective Are AI Content Detectors?
The researchers asked different AI models to write on different kinds of topics. All the texts produced by AI were then analyzed. The researchers analyzed what kind of pattern all these AI models are following which they named as syntactic templates. The researchers then found out that each AI model has its own specific syntactic template. One thing common about AI models was that they often used a set of dual adjectives to describe something.
Even though humans can also have repeated syntax in their writings, it is much lower than repeated syntax in AI writings. The writing style also mattered in identifying how difficult human writing is from AI writing. In texts like biomedical writing, the difference was small because biomedical writing uses a specific pattern. But in texts like movie reviews and news articles, the difference was bigger.
75% of the writing templates that AI models follow are from their training data and are not something they generate on their own during the writing process. So, if you want to detect AI generated text, it is best to look at repeated syntax and dual adjectives. Humans tend to get more creative in their writings, but AI models have to follow a certain method.
Image: DIW-Aigen
Read next:
• How To Use Google Docs Secret Feature to Detect AI Content Writing
• How Effective Are AI Content Detectors?