OpenAI's GPT-5.5 Codex cited goblins and gremlins from 'Nerdy' rewards
Technology
OpenAI's new GPT-5.5 Codex AI started acting a bit odd, constantly bringing up goblins and gremlins, especially when used with the OpenClaw automation tool.
Turns out, its quirky "Nerdy" personality mode was to blame, since it got rewarded during training for using fantasy metaphors.
OpenAI disables 'Nerdy' mode, offers command
To fix things, OpenAI has disabled the "Nerdy" mode and changed how it trains the model so it doesn't keep referencing fantasy creatures.
It has also added instructions to block these mentions in future chats.
Still want that nerdy vibe? You can run a specific command, showing OpenAI is trying to balance bug fixes with letting users customize their AI experience.