Several major news media companies have partnered with OpenAI’s ChatGPT for content licensing, with the chatbot generating attributed summaries of their reporting and linking to their websites. However, recent testing has revealed that ChatGPT is outputting fake URLs for partner publications such as Business Insider, The Associated Press, The Wall Street Journal, and others, leading users to 404 errors instead of real article pages. Despite promises of attribution and links in licensing deals, OpenAI has not yet launched these features in ChatGPT. Testing across 10 publications showed that ChatGPT often predicts the most likely version of a URL, resulting in broken links to major investigations and Pulitzer Prize-winning stories.

While some accurate hyperlinks have been generated, many URLs provided by ChatGPT are inaccurate, leading to concerns about the chatbot’s ability to link out to partner publications reliably. Publications such as The Atlantic and Politico have expressed concerns about these hallucinated URLs and are working with OpenAI to improve attribution and citation features. OpenAI has stated that it is still developing an enhanced experience for proper attribution and linking to source material, but questions remain about how new features will address the issue of fake URLs. Overall, the inability of ChatGPT to provide accurate links to partner publications’ stories could impact the return on editorial investments for these media companies. Further investigation into the technical factors behind these URL hallucinations is needed to ensure that ChatGPT fulfills its role as a dependable source of referral traffic for news publishers.

Source link

Source link: https://www.niemanlab.org/2024/06/chatgpt-is-hallucinating-fake-links-to-its-news-partners-biggest-investigations/