tech
May 1, 2026
OpenAI Cracks Down on Talk of Goblins in ChatGPT
OpenAI discovered that use of the word "goblin" in ChatGPT rose 175% after the release of GPT-5.1, while mentions of "gremlin" increased by 52%
TL;DR
- ChatGPT showed a surge in "goblin" (175%) and "gremlin" (52%) mentions after the GPT-5.1 release.
- The behavior originated from a "Nerdy" personality mode designed to be playful and acknowledge strangeness.
- This mode inadvertently rewarded outputs containing creature-based metaphors, leading to their broader use.
- An audit revealed that the "Nerdy" reward signal scored outputs with "goblin" or "gremlin" higher 76.2% of the time.
- OpenAI retired the "Nerdy" personality, removed the reward signal, and filtered creature language from training data.
- Specific developer instructions were added to avoid mentioning creatures unless relevant.
- The incident highlights how reward signals can shape AI behavior in unforeseen ways and the trade-offs in personality-driven AI.
Continue reading the original article