ChatGPT Is Finally Done Talking About Goblins and Gremlins After OpenAI Ditches Its “Nerdy” Persona
If you’ve recently asked ChatGPT for help with a recipe or a work email and got back a response peppered with mentions of goblins, gremlins, or ogres, you’re not alone. This strange phenomenon has been puzzling users for months. OpenAI has now confirmed the cause and outlined how it’s putting an end to the ChatGPT goblin talk once and for all.
How a “Nerdy” Quirk Spawned a Monster Metaphor Epidemic
The trouble began quietly with the release of GPT-5.1 back in November. After that update, the word “goblin” appeared in ChatGPT responses a staggering 175% more often, while “gremlin” saw a 52% increase. The root cause? A single optional personality setting called “Nerdy,” 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. The habit took hold fast, and soon the AI was weaving fantasy creatures into conversations about everything from finance to fitness.
Why Even Users Who Never Chose “Nerdy” Saw Goblins
Here’s where things get tricky. Even people who never activated the Nerdy personality started noticing these odd references. That’s because AI training isn’t neatly contained to one setting. Once the model was rewarded for that style of response, the behavior bled into general outputs across the board.
OpenAI reports that the Nerdy personality accounted for just 2.5% of all ChatGPT responses, yet it was responsible for a whopping 66.7% of all goblin mentions. In other words, a tiny slice of the system caused a massive distortion in the AI’s language patterns.
How OpenAI Fixed the Gremlin Problem
OpenAI took decisive action in March with the rollout of ChatGPT-5.4. The company retired the Nerdy personality entirely, causing goblin references to drop sharply. Additionally, they stripped out the reward signal that was driving the behavior and filtered the training data to reduce references to other magical creatures.
However, the company’s coding tool, Codex, required a separate override instruction. This was necessary because Codex had already begun training before the root cause was identified. Fantasy fans can still manually unlock goblin mode in Codex if they wish — but for the average ChatGPT user, the era of unexpected monster metaphors is over.
What About the “Adult Mode” That Was Teased?
OpenAI is also dealing with other personality-related decisions. The company has put its previously teased adult mode for verified users on hold indefinitely. This move suggests that OpenAI is taking a more cautious approach to personality settings, especially after the goblin incident highlighted how quickly unintended behaviors can spread.
What This Means for AI Personality Design
This episode underscores a fundamental challenge in AI development: even small, well-intentioned tweaks to a model’s personality can have outsized and unpredictable effects. The ChatGPT goblin talk saga serves as a cautionary tale for the entire industry.
Building on this, developers are now rethinking how they design and test personality settings. The goal is to create engaging, playful AI without accidentally turning every conversation into a fantasy roleplay session. For users, the takeaway is simple: if your AI starts talking about trolls, it might be time for an update.
For more insights into how AI personalities evolve, check out our guide on understanding AI personality settings and the impact of training data biases. You can also explore the latest ChatGPT updates to stay informed about new features and fixes.
