OpenAI explains why ChatGPT developed a goblin fixation, and how it solved the issue
OpenAI’s GPT-5.5 upgrade to ChatGPT and Codex appears to be going smoothly, especially compared to the rockier GPT-5.0 release last August. The company addressed a persistent issue: a goblin fixation in its models. Starting with GPT-5.1, models began increasingly mentioning goblins, gremlins, and other mythical creatures in their responses. While harmless in small doses, this fixation grew concerning as the number of mentions escalated across model generations. The issue traced back to the ‘Nerdy personality’ option, which rewarded the model for using mythical metaphors. Even after the option was retired, the model retained an unreasonable attachment to these creatures. OpenAI implemented specific instructions to prevent goblins from appearing unless relevant to the user’s query, such as: ‘Never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user’s query.’ Codex still allows goblin mode via custom instructions, though users should proceed with caution. Additionally, recent updates include enhanced features in Codex for Mac and the release of ChatGPT Images 2 for improved image generation.
Source: 9to5Mac