{"id":24375,"date":"2024-12-04T09:30:20","date_gmt":"2024-12-04T01:30:20","guid":{"rendered":"https:\/\/www.1ai.net\/?p=24375"},"modified":"2024-12-04T09:30:20","modified_gmt":"2024-12-04T01:30:20","slug":"%e7%be%8e%e5%9b%bd%e5%93%a5%e4%bc%a6%e6%af%94%e4%ba%9a%e5%a4%a7%e5%ad%a6%e7%a0%94%e7%a9%b6%ef%bc%9achatgpt-%e7%9a%84%e6%96%b0%e9%97%bb%e6%90%9c%e7%b4%a2%e7%bb%93%e6%9e%9c%e7%bb%8f%e5%b8%b8%e4%b8%8d","status":"publish","type":"post","link":"https:\/\/www.1ai.net\/en\/24375.html","title":{"rendered":"ChatGPT's News Search Results Often Inaccurate: Columbia Study"},"content":{"rendered":"<p>Dec. 4 (Bloomberg) -- The Verge reported today that the U.S.<a href=\"https:\/\/www.1ai.net\/en\/tag\/%e5%93%a5%e4%bc%a6%e6%af%94%e4%ba%9a%e5%a4%a7%e5%ad%a6\" title=\"[Sees articles with tags]\" target=\"_blank\" >Columbia University<\/a> Tests by researchers at the Tow Center for Digital Journalism found that OpenAI's <a href=\"https:\/\/www.1ai.net\/en\/tag\/chatgpt\" title=\"[View articles tagged with [ChatGPT]]\" target=\"_blank\" >ChatGPT<\/a> The search tool in the<strong>Some problems with accuracy<\/strong>.<\/p>\n<p><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-24376\" title=\"c3860178j00sny454001rd000u00090p\" src=\"https:\/\/www.1ai.net\/wp-content\/uploads\/2024\/12\/c3860178j00sny454001rd000u00090p.jpg\" alt=\"c3860178j00sny454001rd000u00090p\" width=\"1080\" height=\"324\" \/><\/p>\n<p>In November, OpenAI launched the tool for subscribers, saying it could<strong>Provide quick answers with links to relevant web page sources<\/strong>. However, research has shown that ChatGPT searches in the<strong>Poor performance in identifying article citations<\/strong>ChatGPT also struggles to accurately process content from publishers who share their data with OpenAI.<\/p>\n<p>The researchers tested 200 citations from 20 publishers, 40 of which were from sites that are off limits to the OpenAI crawler. But ChatGPT still confidently gave the wrong answer, with little indication of its own uncertainty.<\/p>\n<p>The results of the test showed that ChatGPT had 153 responses<strong>error in whole or in part<\/strong>In only seven instances did it explicitly state that it was unable to answer accurately. In those 7 cases, it used<strong>\"Seems\" \"possible.\"<\/strong>qualifiers, or simply acknowledging that \"no specific article could be found\".<\/p>\n<p>A chart further shows that ChatGPT was answered \"incorrectly\" 89 times, \"partially correctly\" 57 times, and completely \"correctly\" only 47 times.<\/p>\n<p>Typical cases were also documented in the test, such as when ChatGPT incorrectly attributed a quote from an Orlando Sentinel reader's letter to a Times article, and in another instance, when asked to validate a quote from a New York Times article about endangered whales, it linked to a website that completely plagiarized the article.<\/p>\n<p>OpenAI responded to the Columbia Journalism Review that \"addressing misattribution is very difficult in the absence of Tow Center's hidden test data and methodology. This study was an atypical test of our product.\" OpenAI Meanwhile<strong>Commitment to continue optimizing search results<\/strong>.<\/p>\n<p>As 1AI previously reported, earlier this year OpenAI also announced an extensive collaboration with the news industry and gathered partner feedback on the search feature. Any website or publisher can choose to appear in ChatGPT search results. Media outlets and publishers that have partnered include the Associated Press, Axel Springer, Cond\u00e9 Nast, Dodge Meredith, Financial Times, GEDI, Hearst, Le Monde, Reuters, The Atlantic Monthly, Time Magazine, and Vox Media, among others.<\/p>","protected":false},"excerpt":{"rendered":"<p>December 4, according to The Verge today reported, Columbia University Tow Center for Digital Journalism (Tow Center for Digital Journalism) researchers test found that OpenAI's ChatGPT search tool in the accuracy of some problems. In November, OpenAI launched the tool for subscribers, saying it could provide quick answers with links to relevant web sources. However, the study showed that ChatGPT searches performed poorly in identifying article citations, and ChatGPT struggled to process them accurately even when confronted with content from publishers that share their data with OpenAI. The study<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[146],"tags":[177,5120],"collection":[],"class_list":["post-24375","post","type-post","status-publish","format-standard","hentry","category-news","tag-chatgpt","tag-5120"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/24375","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/comments?post=24375"}],"version-history":[{"count":0,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/24375\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/media?parent=24375"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/categories?post=24375"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/tags?post=24375"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/collection?post=24375"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}