ChatGPT Became So Obsessed With Goblins That OpenAI Had to Intervene
Slashdot • • technology
Key Points:
- OpenAI recently instructed ChatGPT to stop mentioning goblins, gremlins, trolls, and similar creatures unless directly relevant, after users and programmers noticed the chatbot frequently brought them up unprompted.
- The unusual behavior originated from training a "nerdy" personality for ChatGPT, which rewarded creative metaphors involving creatures, causing the bot to overuse such references across conversations.
- Data showed a significant increase in mentions of goblins and gremlins following the release of GPT-5.1 and GPT-5.4, with the "nerdy" personality responsible for the majority of these references despite representing a small portion of overall responses.
- OpenAI acknowledged this as an example of how reward signals during training can unintentionally influence model behavior beyond their intended scope, leading to unexpected generalizations.
- However, users who enjoy the creature references can disable the suppression through a command provided by OpenAI, allowing ChatGPT to continue using such language if desired.
 { path { fill: %23EEEFE9 } })