Digital Tool Detects Academic Content Generated By ChatGPT with 99% Accuracy

A chemist from the University of Kansas has developed a new tool that can spot scientific text generated by ChatGPT with 99% accuracy. Heather Desaire published her research in the peer-reviewed journal Cell Reports Physical Science, alongside the source code for others to replicate the tool. She argued that AI-detection tools are increasingly necessary to maintain scientific integrity, as text generators like ChatGPT can produce believable sounding falsehoods that could pollute academic literature. Her detection method narrows the scope of writing to scientific work in peer-reviewed journals, improving accuracy over the RoBERTa detector that aims to detect AI in more general writing. Desaire hopes that others will use her code to further develop the field of AI detection, arguing that the barriers to doing so are not very high.