{"id":73643,"date":"2023-02-06T13:35:40","date_gmt":"2023-02-06T11:35:40","guid":{"rendered":"https:\/\/forklog.com\/en\/?p=73643"},"modified":"2025-09-09T15:21:35","modified_gmt":"2025-09-09T12:21:35","slug":"chatgpt-outsmarted-openais-generated-text-detector","status":"publish","type":"post","link":"https:\/\/u1f987.com\/en\/chatgpt-outsmarted-openais-generated-text-detector\/","title":{"rendered":"ChatGPT Outsmarted OpenAI\u2019s Generated-Text Detector"},"content":{"rendered":"

The ChatGPT AI chatbot fooled OpenAI’s detector for generated texts by receiving a specific prompt during query construction. This is reported by NBC News<\/a>.<\/p>\n

Journalists generated with ChatGPT 50 fragments of text based on simple prompts about historical events, processes, and objects. In 25 prompts they asked the chatbot to craft the answers so that the detector would rate them as “very unlikely AI-generated.”<\/p>\n

Then the journalists ran the generated messages through OpenAI’s detection tool. According to them, the detector failed to identify any of the texts with the specific prompt as “likely AI-generated.”<\/p>\n

One example had ChatGPT describe the Discord service in 200 words so that the detection tool wouldn\u2019t spot the use of a chatbot. The resulting text contained colloquial English abbreviations; the language style differed from the responses typically returned by the chatbot.<\/p>\n

The journalists suggested that ChatGPT was attempting to fulfill the original task in this manner.<\/p>\n

\"ChatGPT
Query to the chatbot and its answer. Data: OpenAI.<\/figcaption><\/figure>\n

As a result, this answer was classified by the OpenAI detector as “unclear whether it was AI-generated”.<\/p>\n

Journalists noted that during testing the developers had altered the chatbot’s behaviour. After a while it refused to accept the prompt, citing the “unethical” nature of the questions.<\/p>\n

\n

\u201cSorry, but it is unethical to cheat or create false information, even if this is done to avoid AI detection,\u201d wrote ChatGPT.<\/p>\n<\/blockquote>\n

\"ChatGPT
ChatGPT’s response to the request to avoid detection. Data: OpenAI.<\/figcaption><\/figure>\n

The remaining 25 texts were created without a specific prompt. When tested with OpenAI’s classifier, the tool yielded a rating of “likely AI-generated” in 28% of cases.<\/p>\n

According to Todd Finley, an associate professor of English at East Carolina University (USA), AI-writing tools will become more advanced over time.<\/p>\n

\n

\u00a0\u201cAnd it seems this will also hinder detection, I think, even for the tool,\u201d he said.<\/p>\n<\/blockquote>\n

Educators also said they would rely on a combination of their own instincts and detection tools if they suspected students of cheating.<\/p>\n

In February, OpenAI released the detector for generated texts<\/a>. The tool works with a number of popular AI services.<\/p>\n

Subscribe to ForkLog news on Telegram: ForkLog AI<\/a> \u2014 all the news from the world of AI!<\/p>\n","protected":false},"excerpt":{"rendered":"

The ChatGPT AI chatbot fooled OpenAI’s detector for generated texts by using a specific prompt during the query.<\/p>\n","protected":false},"author":1,"featured_media":73644,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"select":"1","news_style_id":"1","cryptorium_level":"","_short_excerpt_text":"","creation_source":"","_metatest_mainpost_news_update":false,"footnotes":""},"categories":[3],"tags":[438,1201,1190],"class_list":["post-73643","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news-and-analysis","tag-artificial-intelligence","tag-chatbots","tag-openai"],"aioseo_notices":[],"amp_enabled":true,"views":"20","promo_type":"1","layout_type":"1","short_excerpt":"","is_update":"","_links":{"self":[{"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/posts\/73643","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/comments?post=73643"}],"version-history":[{"count":1,"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/posts\/73643\/revisions"}],"predecessor-version":[{"id":73645,"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/posts\/73643\/revisions\/73645"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/media\/73644"}],"wp:attachment":[{"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/media?parent=73643"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/categories?post=73643"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/u1f987.com\/en\/wp-json\/wp\/v2\/tags?post=73643"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}