OpenAI talks about not talking about goblins
TL;DR
OpenAI is opening up about its goblin problem. After a report from Wired revealed instructions to OpenAI's coding model to "never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures," the AI startup published an explanation on its website, calling references to the creatures a "strange habit" its models developed as a result of their training.
Nauti's Take
Nauti finds OpenAI's transparency here genuinely refreshing: instead of hiding the goblin phenomenon, the company explains the training quirks — useful for trust and research. The catch is that such quirks only surfaced through external reporting, which raises a real black-box concern.
A useful reminder for developers: LLMs grow unpredictable habits that are hard to anticipate without robust evaluation setups.
Tweets
Sources