If you’ve been chatting with ChatGPT lately and noticed it dropping oddly specific references to goblins, gremlins, ogres, or trolls, you’re not imagining things. OpenAI has now explained why ChatGPT has developed this strange habit and how it’s getting fixed.
How a “nerdy” quirk became everyone’s problem
The problem quietly started with GPT-5.1, released in November. After that launch, use of the word “goblin” in ChatGPT responses jumped 175%, while “gremlin” rose 52%. The culprit turned out to be one of ChatGPT’s optional personality settings called “Nerdy,” which was designed to make the AI sound playful and intellectually curious.
During training, OpenAI accidentally gave the model unusually high rewards for responses that included creature-based metaphors, and the habit took hold fast.
How did a single personality setting cause so much goblin talk?
Here’s where it gets interesting. Even users who never switched on the Nerdy personality started seeing goblin references pop up in their chats. That’s because AI training isn’t contained to one setting. Once the ChatGPT model was rewarded for that style, the behavior bled into general responses across the board.
OpenAI says the Nerdy personality made up just 2.5% of all ChatGPT responses, yet accounted for 66.7% of all goblin mentions.
So how is OpenAI actually fixing this?
OpenAI retired the Nerdy personality in March with ChatGPT-5.4, which caused goblin references to drop sharply. The company also stripped out the reward signal driving the behavior and filtered training data to reduce references to other magical creatures.
Its coding tool, Codex, however, needed a separate override instruction since it had already begun training before the root cause was identified. Fantasy fans can still unlock goblin mode in Codex manually, if that’s your thing.
OpenAI is also dealing with other personality-related decisions, including putting its previously teased adult mode for verified users on hold indefinitely.