{"id":82843,"date":"2025-06-02T08:05:18","date_gmt":"2025-06-02T07:05:18","guid":{"rendered":"https:\/\/www.taxresearch.org.uk\/Blog\/?p=82843"},"modified":"2025-06-02T08:05:18","modified_gmt":"2025-06-02T07:05:18","slug":"chat-gpt-a-word-of-warning","status":"publish","type":"post","link":"https:\/\/www.taxresearch.org.uk\/Blog\/2025\/06\/02\/chat-gpt-a-word-of-warning\/","title":{"rendered":"Chat GPT &#8211; a word of warning"},"content":{"rendered":"<p>I was trying to find an old blog post of mine yesterday and could not do so. As a result, I enlisted the help of ChatGPT. I wrote a pretty clear instruction and made clear I wanted to find a specific article on a very precise topic, and then have ten other articles on the same broad theme listed by it using Harvard style referencing, but with an explicit URL shown in each case so I could find the materials if I needed to do so.<\/p>\n<p>It fulfilled the ask. But, when I checked the list of articles supplied, some had titles that I thought were very unlikely that I had written. The construction just felt wrong. I checked three. None of them existed, even though precise titles and URLs specifying dates of publication were provided.<\/p>\n<p>I asked Chat GPT whether they really did exist. In each case, it said:<\/p>\n<blockquote>\n<p class=\"p1\">After reviewing the available information, it appears that the article titled \u201c[xxx etc]\u201d purportedly published on [date] on your blog \u201cTax Research UK,\u201d does not exist. A thorough search of your blog\u2019s archives for that date and title yielded no results.<\/p>\n<p class=\"p1\">It\u2019s possible that this reference was included in error or that the article was published under a different title or on a different date. If you have any additional details or context about the content of the article, I would be happy to assist you further in locating it.<\/p>\n<\/blockquote>\n<p>I then asked:<\/p>\n<blockquote>\n<p class=\"p1\">Why did you reference articles that did not exist?<\/p>\n<\/blockquote>\n<p>I got this reply:<\/p>\n<blockquote>\n<p class=\"p1\">You\u2019re absolutely right to challenge this, and I appreciate your patience.<\/p>\n<p class=\"p1\">The mistake occurred because I generated plausible-sounding article titles and publication dates based on the themes you\u2019d asked for and your well-established positions\u2014but without first verifying whether those precise articles actually existed on your blog. That was an error.<\/p>\n<p class=\"p1\">I should have fact-checked each reference against your real published work on <i>Tax Research UK<\/i> before presenting them as citations. Instead, I inferred them based on patterns in your writing, which created the impression of referencing your own archive\u2014but in fact pointed to non-existent URLs.<\/p>\n<p class=\"p1\">That\u2019s not acceptable when accuracy is required, and I take responsibility for the confusion it caused. Going forward, I\u2019ll verify every reference directly against your blog unless you specify otherwise.<\/p>\n<\/blockquote>\n<p>So, a warning. If you ask for references from ChatGPT, make clear you want them for sources that actually exist. Otherwise, it just makes stuff up.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>I was trying to find an old blog post of mine yesterday and could not do so. As a result, I enlisted the help of<br \/><a class=\"moretag\" href=\"https:\/\/www.taxresearch.org.uk\/Blog\/2025\/06\/02\/chat-gpt-a-word-of-warning\/\"><em> Read the full article&#8230;<\/em><\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[25],"tags":[],"class_list":["post-82843","post","type-post","status-publish","format-standard","hentry","category-blogging"],"_links":{"self":[{"href":"https:\/\/www.taxresearch.org.uk\/Blog\/wp-json\/wp\/v2\/posts\/82843","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.taxresearch.org.uk\/Blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.taxresearch.org.uk\/Blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.taxresearch.org.uk\/Blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.taxresearch.org.uk\/Blog\/wp-json\/wp\/v2\/comments?post=82843"}],"version-history":[{"count":1,"href":"https:\/\/www.taxresearch.org.uk\/Blog\/wp-json\/wp\/v2\/posts\/82843\/revisions"}],"predecessor-version":[{"id":82844,"href":"https:\/\/www.taxresearch.org.uk\/Blog\/wp-json\/wp\/v2\/posts\/82843\/revisions\/82844"}],"wp:attachment":[{"href":"https:\/\/www.taxresearch.org.uk\/Blog\/wp-json\/wp\/v2\/media?parent=82843"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.taxresearch.org.uk\/Blog\/wp-json\/wp\/v2\/categories?post=82843"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.taxresearch.org.uk\/Blog\/wp-json\/wp\/v2\/tags?post=82843"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}