In the ever-evolving landscape of digital journalism, a recent study by the Tow Center for Digital Journalism has thrown a spotlight on a pressing issue: the accuracy of ChatGPT's citations.
As publishers increasingly engage in content licensing deals with OpenAI, the maker of ChatGPT, concerns are mounting over the chatbot's tendency to misrepresent or fabricate sources, regardless of whether these publishers have allowed OpenAI to crawl their content.
Conducted at Columbia Journalism School, the research examined how ChatGPT cites sources from a diverse range of publishers, including industry giants like The New York Times, which is currently embroiled in a copyright lawsuit against OpenAI, and The Washington Post, which remains unaffiliated with the AI company. The findings were less than encouraging. Out of 200 quotes tested, ChatGPT frequently returned incorrect citations, with only a handful being accurate. Even publishers who blocked OpenAI's crawlers weren't immune to the chatbot's creative sourcing.
This inconsistency poses significant risks for publishers. Incorrect citations not only mislead readers but also threaten the reputation of media outlets, potentially diverting audiences to other sources. The study highlights that even publishers who have entered licensing agreements with OpenAI cannot rely on ChatGPT for accurate representation of their content. The AI's tendency to treat journalism as decontextualized content, with scant regard for its original context, exacerbates the issue.
Want to hear more? Join Mal on the Property AI Report Podcast each week!
Access from your preferred podcast provider by clicking here
Moreover, the study found that ChatGPT's responses varied significantly when asked the same question multiple times, a characteristic typical of generative AI tools but problematic in citation contexts where consistency and accuracy are paramount. This variability further undermines user trust in the chatbot's answers.
While the Tow Center's study is relatively small, it raises crucial questions about the reliability of AI-generated citations. Publishers hoping for preferential treatment from OpenAI might need to reconsider their expectations, as the current framework offers little control over how their content is represented. OpenAI's response to the study, which labelled the research as an "atypical test," underscores the ongoing challenge of balancing AI innovation with the need for accuracy and accountability in digital content.
As the digital landscape continues to shift, the relationship between AI and journalism remains complex and fraught with challenges. For now, it seems that both publishers and AI developers have their work cut out for them in ensuring that the future of digital content is both innovative and reliable.
Want to hear more? Join Mal on the Property AI Report Podcast each week!
Access from your preferred podcast provider by clicking here
Made with TRUST_AI - see the Charter: https://www.modelprop.co.uk/trust-ai
Comentarios