{"id":622567,"date":"2026-05-04T08:29:15","date_gmt":"2026-05-04T08:29:15","guid":{"rendered":"https:\/\/www.newsbeep.com\/us\/622567\/"},"modified":"2026-05-04T08:29:15","modified_gmt":"2026-05-04T08:29:15","slug":"chatgpt-became-so-obsessed-with-goblins-that-openai-had-to-intervene","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/us\/622567\/","title":{"rendered":"ChatGPT Became So Obsessed With Goblins That OpenAI Had to Intervene"},"content":{"rendered":"<p>\t\t\t\t<a href=\"https:\/\/www.msn.com\/en-us\/news\/technology\/chatgpt-became-so-obsessed-with-goblins-that-openai-had-to-intervene\/ar-AA22br39\" rel=\"nofollow noopener\" target=\"_blank\">The Wall Street Journal reports<\/a> that OpenAI &#8220;recently gave its popular ChatGPT strict instructions. Stop talking about goblins.&#8221;<\/p>\n<p>Recent models of the artificial-intelligence chatbot have been bringing up the creatures in conversations with users seemingly out of the blue, as well as gremlins, trolls and ogres. The goblin-speak caught the attention of programmers, who are often heavy users of the bot. Barron Roth, a 32-year-old product manager at a tech company, said the bot referred to a flaw in his code as a &#8220;classic little goblin.&#8221; He said he counted more than 20 times it mentioned goblins, without any prompting&#8230; <\/p>\n<p>Several users speculated that goblin terminology was how the model characterized itself, in lieu of identifying as a person with a soul.   Then OpenAI decided enough was enough. &#8220;Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user&#8217;s query,&#8221; reads an open source line in ChatGPT&#8217;s base instructions for its coding assistant. <br \/>The Journal calls this &#8220;a reminder that even as AI companies tout one advance after another in their technology, they are sometimes baffled by the things their own models do&#8230;.&#8221;  While training a &#8220;nerdy&#8221; personality for their model&#8217;s customization feature, &#8220;We unknowingly gave particularly high rewards for metaphors with creatures,&#8221; OpenAI <a href=\"https:\/\/openai.com\/index\/where-the-goblins-came-from\/\" rel=\"nofollow noopener\" target=\"_blank\">explained in a log post<\/a>.  And &#8220;From there, the goblins spread.&#8221;<\/p>\n<p>When we looked, use of &#8220;goblin&#8221; in ChatGPT had risen by 175% after the launch of GPT-5.1, while &#8220;gremlin&#8221; had risen by 52%&#8230; With GPT-5.4, we <a href=\"https:\/\/news.ycombinator.com\/item?id=47319285\" rel=\"nofollow noopener\" target=\"_blank\">and our users\u00e2<\/a> noticed an even bigger uptick in references to these creatures&#8230;  Nerdy accounted for only 2.5% of all ChatGPT responses, but 66.7% of all &#8220;goblin&#8221; mentions in ChatGPT responses&#8230;  The rewards were applied only in the Nerdy condition, but reinforcement learning does not guarantee that learned behaviors stay neatly scoped to the condition that produced them. Once a style tic is rewarded, later training can spread or reinforce it elsewhere, especially if those outputs are reused in supervised fine-tuning or preference data.   <br \/>It all started because the &#8220;nerdy&#8221; personality&#8217;s prompt had said &#8220;You must undercut pretension through playful use of language. The world is complex and strange, and its strangeness must be acknowledged, analyzed, and enjoyed&#8230;&#8221;   Now OpenAI calls this &#8220;a powerful example of how reward signals can shape model behavior in unexpected ways, and how models can learn to generalize rewards in certain situations to unrelated ones.&#8221; <\/p>\n<p>But &#8220;fans of goblins don&#8217;t have to fear,&#8221; notes the Wall Street Journal. &#8220;OpenAI provided a command in its blog post that would remove its creature-suppressing instructions.&#8221;<\/p>\n","protected":false},"excerpt":{"rendered":"The Wall Street Journal reports that OpenAI &#8220;recently gave its popular ChatGPT strict instructions. Stop talking about goblins.&#8221;&hellip;\n","protected":false},"author":2,"featured_media":622568,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[31],"tags":[74],"class_list":["post-622567","post","type-post","status-publish","format-standard","has-post-thumbnail","category-technology","tag-technology"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/posts\/622567","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/comments?post=622567"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/posts\/622567\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/media\/622568"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/media?parent=622567"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/categories?post=622567"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/tags?post=622567"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}