This article explores the challenges of detecting AI-generated text and provides insights into the intuition and technical specification of building a detection tool. It discusses the concept of perplexity in language models and how it can be computed. It also mentions some AI-generated text detection tools like OpenAI and GPTZero. The article emphasizes the difficulty of accurately detecting AI-generated text and discusses the potential future directions for improving detection methods.
Table of contents
Examples of AI-generated v/s human written textFalse positivesThe generality of this approach3 Impressions