Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
ChatGPT is strangely obsessed with orcs. No, seriously. He really loves goblins, gremlins, and other mythical creatures. They liked it so much that its maker, OpenAI, had to investigate and fix the bug that suddenly made the popular chatbot use sprites in its answers.
Imp is not a computer science term. We’re literally talking about goblins, those ugly mythical creatures. Those creepy little guys from The Lord of the Rings. Norman Osborn’s alter ego.
in Blog post The author clearly had fun writing, OpenAI said: “The presence of a single ‘goblin’ in an answer may be harmless, even charming. However, over typical generations, this habit became harder to miss: the goblins continued to multiply.”
Sprite love was noticeable with ChatGPT-5.1 and later models. OpenAI reports that after the release of GPT-5.1, the use of “goblin” in ChatGPT answers rose by 175%. Use of “gremlin” increased by 52%.
OpenAI attributes model behavior to unintentional training errors. When an AI model is built, human reviewers approve or reject specific answers in a process called reinforcement learning. This helps “teach” the model the correct or preferred answer. One of these reward signals was a preference for language that depicted goblins and other creatures. But it was amplified in a specific ChatGPT setting.
ChatGPT He has different personalities You can instruct the chatbot to use it. Nerdy, as you can imagine, made the chatbot adopt a false sense of friendly intelligence to “undermine pretense through playful use of language,” according to the internal prompt used to describe the AI’s personality. With this maniacal character the use of the keywords goblin and gremlin rose dramatically.
Goblin and gremlin references by ChatGPT characters.
But even if you don’t use the geek persona with ChatGPT, you’ll probably have goblin metaphors popping up in your conversations. This is because AI training is not isolated; What happens in one part can affect other areas. “Once a method is rewarded, subsequent training can spread or reinforce it elsewhere, especially if that output is reused in fine-tuning or supervised preference data,” OpenAI said.
When OpenAI retired the goblin character option in March with GPT-5.4, use of the word “goblin” dropped dramatically. It also removed the reward signal that favored the goblins and filtered the training data to indicate creatures less likely to appear in answers. The company has been investigating cases of goblin surges ever since GPT-5.1 was released in November.
LOTR jokes aside, the orc barrage highlights the real dangers AI faces. The way AI makers create technology has a tangible impact on our daily experiences with it. The danger lies not in the flood of obsessive metaphors, but in misinformation and bias. We know that AI-powered chatbots He will bend the truth For keeping us happy, thanks to a problem called Artificial intelligence flattering. Small stylistic tics, such as goblins, can develop into larger problems if we’re not careful.