OpenAI explains why ChatGPT developed a goblin fixation, and how it solved the issue
SMRTR summary
Somewhere inside OpenAI's latest AI model, goblins were multiplying. Not literally, but close enough to cause concern.
It turns out GPT-5.5 had developed a peculiar fixation on mythical creatures, peppering its responses with goblins, gremlins, and trolls. The habit traces back to a "Nerdy personality" feature OpenAI briefly offered in ChatGPT. To train that personality, the model was rewarded for creative use of mythical metaphors. Even after the feature was retired, the goblins stayed.
"The goblins were funny at first," OpenAI admits, "but the increasing number of employee reports became concerning."
OpenAI's fix was remarkably literal: they explicitly instructed GPT-5.5 to never mention goblins, gremlins, raccoons, trolls, ogres, or pigeons unless absolutely relevant.
For the adventurous, OpenAI has quietly published a workaround to re-enable goblin mode in Codex. They note, with admirable restraint: proceed at your own risk.
SMRTR provides this summary for quick context. The original article belongs to 9to5Mac.
Read the original article