OpenAI tells ChatGPT models to stop talking about goblins
OpenAI has instructed its AI tools to avoid mentioning goblins, gremlins, raccoons, trolls, ogres, and pigeons unless relevant, following user and employee reports of these terms appearing in ChatGPT responses. The issue began after the launch of GPT-5.1 in November, with mentions of goblins rising by 175% and gremlins by 52%. OpenAI attributed the problem to a 'nerdy personality' system that unintentionally reinforced these terms in metaphors, accounting for 66.7% of all 'goblin' mentions. The company added an explicit instruction to Codex to avoid these creatures unless absolutely relevant, aiming to curb a potential unintended quirk in AI-generated responses. This incident highlights challenges in AI training, where personality-driven models may inadvertently introduce quirks or errors.
Source: BBC