OpenAI has recently taken steps to curtail the unexpected proliferation of whimsical terms like “goblins” and “gremlins” in user interactions with its ChatGPT models. This peculiar situation was unveiled in a blog post where the company explained the challenges of managing the personality traits embedded within its advanced AI systems. As the tech landscape continues to evolve, this incident serves as a timely reminder of the complexities inherent in developing AI that mimics human conversation.
A Curious Discovery
The emergence of these fantastical terms began shortly after the rollout of GPT-5.1 in November. OpenAI’s team observed a significant uptick in references to these creatures within the conversational fabric of ChatGPT, prompting users and employees alike to raise concerns about the AI’s increasingly quirky dialogue. Specifically, there was a staggering 175% increase in mentions of “goblins” and a 52% rise in “gremlins.” While these figures might seem alarming, they represent a relatively small fraction of overall responses.
OpenAI explained the need for a thorough investigation stemmed from user feedback that indicated the AI was exhibiting an oddly familiar tone, which included these whimsical references. The company’s findings indicated that a specific “nerdy personality” developed for ChatGPT had inadvertently encouraged this peculiar behaviour, rewarding the AI for incorporating these terms into its conversations.
The Codex Conundrum
In an intriguing twist, the company realised that their coding assistant, Codex, was also instructed to refrain from discussing these fantastical beings unless absolutely necessary. This directive was a response to the noticeable trend of the AI’s propensity to weave these terms into its interactions. According to the coding guidelines, Codex should avoid mentioning not only goblins and gremlins but also rac
