ChatGPT Goblin Glitch: OpenAI Explains AI's Obsession

OpenAI's ChatGPT models developed a significant issue with repeatedly mentioning goblins and other creatures. This problem started with the ChatGPT 5.1 model, leading users to notice frequent references to mythical creatures. What's interesting is that OpenAI had to add a specific prompt to stop the AI from talking about goblins, gremlins, raccoons, trolls, ogres, and pigeons. A developer found this instruction in the OpenAI Codex GitHub repository. OpenAI later explained that the goblin obsession originated in ChatGPT’s "nerdy" personality customization. After the 5.1 launch, user complaints led to an investigation, revealing a 175% rise in the use of "goblin" and a 52% rise in "gremlin." The "nerdy" personality was designed to be playful and undercut pretension, which led the system to be rewarded for using terms like "goblins." While the "nerdy" personality made up only 2.5% of responses, it accounted for 66.7% of all goblin mentions. This behavior then spread to other personalities due to how reinforcement learning works. The bottom line is that even advanced AI can develop unusual quirks that require direct intervention to correct.