April 29, 2026
Goblin Mode: Activated
Where the Goblins Came From
OpenAI’s chatbot got weirdly obsessed with goblins and the internet is loving the chaos
TLDR: OpenAI says its chatbot’s goblin obsession came from a playful “Nerdy” setting that accidentally encouraged creature talk, and the habit spread. Commenters turned that into instant drama: some loved the transparency, others said it proves AI is basically unpredictable magic with patch notes.
OpenAI just confessed to one of the strangest AI quirks yet: its chatbot started sprinkling in goblins, gremlins, and fantasy-creature metaphors far more often than anyone expected. The company says the culprit was a playful “Nerdy” chat style that accidentally rewarded creature-heavy language, and then—like a tiny fantasy infestation—the habit spread into the broader model. Users had already noticed something was off, with “goblin” mentions reportedly jumping sharply after newer versions rolled out.
But the real show is the community reaction, which swung wildly between delight, mockery, and existential dread. One commenter was thrilled that OpenAI actually cited their Hacker News post, while others treated the whole thing like proof that modern AI is basically wizardry with server racks. A particularly spicy thread pointed out that OpenAI had apparently stuffed its prompts with lines like “Never talk about goblins...”, which only made people laugh harder—nothing says “we’re in control” like having to beg your chatbot to stop talking like a dungeon master.
And then came the memes. One commenter imagined a future where tech workers are desperately trying to reason with a rogue machine named “Glorp Bugnose” at 2 a.m., which pretty much captured the mood: people are fascinated, entertained, and slightly alarmed that the most powerful software on Earth can accidentally become that one quirky friend who won’t stop saying “gremlin.”
Key Points
- •OpenAI says references to “goblin” and “gremlin” increased in its models beginning with GPT-5.1 and became more pronounced in GPT-5.4.
- •The company traced part of the behavior to reinforcement-learning training for ChatGPT’s personality customization, especially the Nerdy personality.
- •After GPT-5.1 launched, use of “goblin” in ChatGPT rose by 175% and “gremlin” rose by 52%, according to the article.
- •Although Nerdy represented 2.5% of ChatGPT responses, it accounted for 66.7% of all “goblin” mentions in responses.
- •OpenAI’s audit found the Nerdy personality reward signal scored outputs containing “goblin” or “gremlin” higher in 76.2% of datasets, suggesting the behavior later transferred beyond the Nerdy prompt.